Teradyne, Inc. Stock

Okay. Good afternoon, everybody. Welcome to the Goldman Sachs Communacopia and Tech Conference. My name is Jim Schneider. I'm the semiconductor analyst here at Goldman Sachs, and my pleasure to welcome Teradyne and CEO, Greg Smith to the stage today. Welcome, Greg. Thanks for being here.

Great to be here.

Likewise. Greg, you operate a very diverse business today, which has high-performance test and measurement at the core. If you think about the exposure that you have across test, semiconductor test, system-level test, industrial robotics and other areas, how do you think about Teradyne business mix over the long run? And what objectives are you trying to move as a business forward with?

So it's a really interesting question. So Teradyne has, as you said, we have a semiconductor business that represents about 80% of our revenue, a robotics business that's about 10% of our revenue and a product test business that's about 10%. A little bit more for Semi Test recently because it's grown so quickly. But the kind of the uniting theme for all of those businesses right now is exposure to the AI build-out.

So Teradyne basically has exposure to the AI build-out from wafer all the way to data center. And some of that is upfront wafer to package test with our semiconductor test business. And then our product test and system-level test business really takes us through the part, through the module and the tray level. And our robotics business really picks up in the areas of robotic-assisted test, robotic-assisted assembly and robotic-assisted operations in data centers.

So we're kind of in a unique period right now where these 3 parts of Teradyne's business are reinforcing each other much more than they ever have before. And that's actually a very good thing because not only are we seeing this whole value chain coming across our businesses, but increasingly, the end market is more and more concentrated in the form of these hyperscalers, the people that are actually building out these data centers and using them and developing the chips for them, they have diverse needs across the whole range of what Teradyne does.

So we have exposure to the whole value chain, and we have a set of customers that want to deal with us across those different nodes.

Excellent. Another high-level question. If we come up here and get on stage in 5 years, what do you think is going to be the one thing that investors are surprised at retrospectively?

So this was the stumper from last year. So it's a rerun of that time to think it over. Unfortunately, the whole world is different 12 months on, right? So one of the things that we've learned is that predicting 5 years out is a really tough thing to do. We're even -- as an industry, we're even struggling to try and figure out how big things are going to be in the current year. I do think that there are some things that are pretty certain to be true looking back from 5 years from now. One is that AI is going to continue to be the dominant force across our industry.

And right now, that is mostly being driven through the build-out of this data center economy. But beyond that, there's a whole wave of physical AI, whether it's in robotics, autonomous vehicles, edge AI for mobile electronics, new types of devices that we expect to be a very big part of our business looking further out into the future.

So if we're looking all the way out to 2031, 2032, I would expect to see that this is -- that we will have experienced the most sustained period of growth that we've ever seen in our industry. I have no illusion that it's going to be perfectly monotonic that there's going to be bumps and grinds along the way. But the setup in terms of the importance of information technology and AI is undeniable.

And having ourselves positioned in this space to be able to benefit from not only the compute part of this, but the mobile and power and robotics part of it means that Teradyne is looking at a kind of a unique period in our history in terms of aligning our corporate strategy with the direction that the whole global economy is going.

So like I think looking back 5 years from now, other than feeling really, really old, I think we will have seen kind of a golden age for the semiconductor industry and for the test industry in general.

Much bigger.

Yes.

Okay. Fine. So then let's talk about sort of the shorter term. I think everything we see from at least our view of wafer fab equipment spending kind of suggests that we're going to have further revenue acceleration in '27 relative to '26. I realizing your revenue is tied to test capacity rather than units, of course. How would you handicap Teradyne's ability to grow faster next year as well? And what are the key levers that will determine whether that materializes or doesn't?

Yes. So it's a really interesting question and one that we've spent a lot of time trying to understand better, both so we can size our own business, but also so we can help people understand the factors that are at play. So one way that people are trying to wrap their heads around the size of the test equipment market is to look at it as a portion of the entire semiconductor capital spend.

And right now, the money is that semiconductor capital spend is going to be increasing pretty rapidly over the next few years, right now, kind of last year kind of $120 billion, moving on towards $250 billion or even $300 billion of semiconductor capital spend in the '28 to 2030 period of time.

Semiconductor test represents a fraction of that. If you look back in 2023, that was maybe 4% of the total semiconductor CapEx went into semiconductor test. In the first 5 months of 2026, it's actually 8% of the total is going into test. So it's a much bigger market, and we have doubled the amount of that market that is going into test.

Looking out into the future, as the market continues up to this $250 billion, we're not sure that there's -- that it's going to continue to increase as a percentage, but it's certainly going to increase in proportion to that market. So we're going to expect to see that test is a really good place to play. And the things that are driving that. So if you think about semiconductor capital spend, the most important thing that, that does is it enables new generations of process nodes and increased capacity.

Each new process node allows additional transistor density. And a way to think about this is the amount of test required for an entire wafer really is first order dependent on the number of transistors on that wafer. So if you're doing something in 1.4 nanometer versus 3-nanometer, you're going to get a whole bunch more transistors. That's going to drive a whole bunch more test per wafer.

And the other thing is to think about just how many wafers are being produced. That's the basic capacity. If you multiply those 2 things together, you get a rough idea of how much test is going to be required over the next few years. That's the part of the test market that should be a consistent percentage of semiconductor capital spend. The reason that it's accelerated over the past couple of years is because test intensity is increasing. And the reason the test intensity is increasing is because people are now putting more than one die per package. And when you start putting more than one die per package, that means that you need to ensure that each die is tested to a higher level of quality.

So think about it, if every time you test a device, you sort out the ones that you think are good from the ones that you think are bad. In the pool of ones that you think are bad are probably some good devices that you would have been able to ship. And in the pool of devices that you decide are good, there are some that have latent defects. Those latent defects are ones that will be found later on in the process. If all you're doing is taking a single silicon die and putting it in a package, it doesn't have that much of an effect if you find the fault after you package it or before.

But if you're doing something in a multi-die package, if you find a latent defect in one of those devices, it not only takes out that one device and the package, it takes out all the other die and HBM memories and everything else. So when you're trying to test things that are going into these multi-die packages, you need to invest much more heavily in the wafer test for those devices. So that's the primary reason that we believe that the test intensity as a fraction of the whole market has been going up.

Got it. Okay. Now your compute business, the part that's tied to AI GPUs, ASICs networking has obviously done very well this year. I think one of the reasons why investors have been excited about the Teradyne story is the prospect of you gaining some GPU market share with a leading customer in that market. You've already shipped some product to that customer earlier this year. How would you sort of frame for investors the next revenue opportunity for that customer and the share of the business you might ultimately be able to capture?

Sure. So in the middle of 2025, Teradyne was invited to attempt to qualify as a second vendor for test in a merchant compute account. The process for that culminated in the release of an initial part in the first part of this year and that part being loaded in production on a small fleet of testers. And that's great.

So we've demonstrated that we can provide equivalent test to the incumbent platform on our platform. Having done that release, there are now two things that are going on. One is we are developing all of the technologies and the processes necessary to keep both platforms test solutions aligned with each other as they change. And we've also begun the process of developing additional devices that will be released on both our platform and the incumbent platform.

The first part that we did as a qualification vehicle was a part that was mature. It had already ramped. They had very good production data. So the one variable was our tester. And so by using a mature part, they were able to say, okay, you got a different result here, that's your fault. You have to fix it. If you're doing that on a new part, you don't know whether it's a part problem or a tester problem or something else. By doing it this way, they were able to isolate just what was associated with our platform and we're through that.

The next parts that we're working on are earlier in their life cycle. They have not fully ramped to their full volume, and we would expect to finish those projects in the second part of this year. Once those projects are finished and released, then we would be qualified to capture volume associated with those devices. And we would expect that to happen in 2027. We expect to be in this mode of converting programs from an incumbent platform to our platform probably for the next 3 to 5 years. And over this period of time, the customer will gain more familiarity with our platform, and we'll also have all these technology transitions going to next-generation tester instrumentation and architectures where we may have the right solution at the right time so that we'd be selected as the new part qualification platform and then the roles would be reversed on that part.

But the way to think about this is for the next 3 to 5 years, we're going to slowly gain share up to maybe 30% of that compute portion of the account. And then once you get through that period and either platform is being selected for initial part qualification, then it's purely on the basis of differentiation, which vendor is going to get the most share.

Got it. Now second driver of your compute business has been custom silicon for XPUs either being done by some of the world's largest hyperscalers or sometimes custom ASICs. So maybe speak to the magnitude of the business you're expecting in that area for 2026. And are you expecting to add another customer or more customers as you welcome to 2027?

Sure. So right now, if you think about custom AI silicon there are really two very large players and a third that is at scale, but not of the same magnitude. And two of them are in cloud compute and the other is primarily in the automotive and robotics space. So those are the three players that are loading hundreds of testers. Those three, let's just call them hyperscalers for lack of a better term.

Teradyne has a -- the 2 cloud compute hyperscalers, one of them has been dual sourced for a number of years. It used to be exclusively on our competitors' platform. We began a process to qualify in 2023 that succeeded in 2024. And by 2026, we had the majority share in that hyperscaler account.

The second cloud computing hyperscaler that's at scale has given us the opportunity to qualify as a second source as a second supplier. We are in the process of doing that. We have correlated our first part, and we are in the process of releasing that to production and going on to follow-on devices. We would expect to begin capturing meaningful production share in 2027 for that hyperscaler as well and each one of these is hundreds of millions of dollars of TAM to $1 billion of TAM in any particular year. So winning a chunk of that share in any particular year is going to be very meaningful as an upside for us.

Great. Finally, networking for AI. That's another area where we have a strong presence. Talk to the size of that business you saw in 2026 and whether or not it's reasonable to expect you could actually see that business outgrow hyperscaler CapEx growth in '27?

So I think we've actually had this question in 1 or 2 of the smaller meetings that we've had today because if you look from '25 to '26, we saw really strong growth in both our networking business and in the AI ASIC business. And we didn't have the numbers in front of us, but our rough guesstimate is that they grew proportionately, that they kind of grew about the same.

Looking forward from '26 to '27, there is probably more upside potential for the ASIC business than networking. They're both going to grow strongly. But the sheer test intensity associated with AI accelerators is such that, that has the potential to really sort of blow the doors off in terms of the total capacity needed. Whereas in networking, it's a more mature market and one that has probably a more predictable TAM growth.

Yes. Finally, on related networking, co-packaged optics, that's been an area of a lot of investor debate recently, especially when it actually happens and when it ramps. What is the market size like as you see it today? And can you maybe speak to your relative market position versus your competitive advantage?

Okay. So let's start with the numbers. In 2026, we think that the test equipment TAM for silicon photonics co-packaged optics is about $100 million. And we expect by 2028 that, that is going to be between $300 million and $700 million. And that's for just the test equipment part of this opportunity. There's also optical alignment and interface hardware and material handling. That's a different TAM, but it's all kind of mixed together. Right now, this is a very concentrated market. There's one customer and a very small number of devices at that customer that are driving most of the capital intensity in 2026.

The thing that I want to try and emphasize is that this is really early days that what you're seeing is companies trying to figure out how to scale this stuff into production. And they are running into all of the things that you normally run into when you try and scale a complex new technology in production. So you see this playing out in a number of different ways. One is you can see that there is increasing excitement about copper networking for scale up. You can see that there's increasing excitement for technologies like NPO so that you don't have the same kind of yield risk associated with the full CPO device.

And you can also see it in terms of people like asking everybody who knows anybody inside of these companies who's up and who's down on a daily basis. But in terms of CPO, let me just sort of give you the lay of the land. If you're trying to build a CPO device, there are 4 major optical test insertions. There's insertion 1, where you're testing just the photonics wafer. There's insertion 2, where you're testing a photonics wafer that has the electronics ICs bonded to it. There's insertion 3 where you're testing the singulated optical engine dielets and those usually have been a mounting plate has been put on them to align the FAU.

And then the fourth is the assembled CPO device and they're insertion 1, insertion 2, insertion 3, insertion 4. For this first device, first customer, the way things have played out is that Advantest is strongest in insertion 1. Teradyne is strongest in insertion 2. A Taiwanese supplier is strongest in insertion 3 and Teradyne is strongest in insertion 4.

But what is going to happen as more customers get in and more parts are developed and the technology changes is that all of that is going to change, that the relative importance between insertion 2 and insertion 3 will change, the technology in terms of the lane counts and data rates is going to change and whether or not you go to NPO or CPO is going to change.

So the way we think this is going to end up happening is that there are going to be players in the market that are experts at the device handling and alignment. And then there are going to be players in this market that are focused on test. So Teradyne and Advantest are going to duke it out in test. There are going to be partners that we work with. Both of us will work with all partners that are handling the device alignment and device handling. And by the time you get to 2030, when you got $1 billion of test equipment being sold, that market is going to look a lot like the rest of the test market where we're competing head-to-head on who has the highest performance test equipment and other players are going to be competing on the alignment.

Got it. Okay. Memory. That's been an area, especially HBM, which has really driven a lot of momentum in your results. I think you've kind of, in the past, talked about some uncertainty well maybe we could see a digestion phase, but it sounds like you see most recently that's going to maintain its strength. As we sort of think about the next 4 or 6 quarters, how do you handicap that relative level of strength and whether or not we see any kind of pullback in that space?

Well, so a little bit of historical context, 2024 was like Yahoo time in HBM. That was like the initial capacity ramp. And from a test capacity perspective, it actually got ahead of production. So in 2025, there was a significant HBM TAM, but it was actually basically flat year-on-year. 2026 is back to strong growth above 2025. And our initial view of 2027 is that it's not going to be a digestion. It's going to be continued growth from there.

The thing that has changed in 2026 is that the rest of the DRAM market has really come alive. It's like much, much stronger than we expected coming into the year. We expected 2026 to be a strong HBM year. We did not expect as strong of a burst of activity in LPDDR and DDR. LPDDR, it's all about CPUs and SOCAMM and all of the memory that you need to do agentic AI. And so that was a happy surprise.

The other thing that we are beginning to see towards the end of 2026 is a resurgence in the NAND market. So back in the olden days, the memory market was almost 50-50 between NAND and DRAM. In 2026, it's like -- in 2025, it was like 85% DRAM, 15% flash. And the flash market used to be that the flash market was really dominated by mobile and the DRAM market was kind of split between mobile and server cloud. Now cloud is everything for everyone. And that's true for flash as well. So the big growth driver in flash is really eSSD and solid-state drives for data center applications. And that is driving TAM growth in flash in 2026, and we expect that to be an even stronger year in '27. So I guess, modest up for HBM and strong growth for DRAM and flash in '27.

Great. Finally, on your robotics business, I think there's been a lot of optimism around that business, specifically centered on like key marquee customer as we head into next year. When do you think that dynamic starts to become meaningful for you in terms of revenue?

So there's two end market dynamics and one underlying technology that get us really excited about robotics. So the two end markets, one is in logistics, and that's where our marquee customer is, is in the e-commerce logistics space. And that is like on plan. So revenue from that major customer tripled from '25 to '26. We expect it to triple again from '26 to '27, and we are still discovering new workflows within that customer that we'll be able to serve with variations of the solutions that they're developing using our products.

The second vertical that we're very excited about for robotics is in electronics manufacturing and all the way back into the semiconductor value chain. So we've been calling this test -- robotics-assisted test and robotics-assisted assembly and robotics-assisted operations for data centers. And this is the fastest growing. It's now the largest vertical that we have for our robotics business is really serving all of the contract manufacturers and other manufacturers that are helping to build out data centers. The thing -- so those are the 2 verticals.

The common thread between them is physical AI. So the reason that this e-commerce customer is able to ramp this application is because this application is actually able to recognize product without pretraining on the product itself. Like whatever the thing is, it figures out how to pick it up and what to do with it. And in order to do that, it needs a sophisticated physical AI application.

The same thing in terms of electronics manufacturing for the kinds of stuff that go into data centers. There's a lot of things that stymy traditional automation, floppy cables, gooey thermal interface materials, adhesives, screwed fasteners and very short product life cycles. All of that really needs AI so that you can get the application released and useful in a factory before the product goes obsolete. So that's become like physical AI underpinning our robotics applications is probably the most important trend that's going to drive our business over the next few years.

Great. Integrated system test, that business, maybe unpack the moving pieces there between the hard drive business and system-level test and how that segment has kind of evolved over the past few years? How much time do we have?

It makes me happy because my first job in semiconductor test, I was an applications engineer working on a hard disk drive read channel. This is like in 1995. And back then, everybody was wondering how long it was going to be before hard drives were obsolete. Like any day now, NAND memory is going to take over. Anyway, now fast forward today, 30 years later, and hard disk drives are showing the fastest exabyte growth that they've shown in years.

So the most important thing that's driving our short-term results in our Integrated System Test group is actually traditional hard disk drive test. And that's both higher exabyte growth, but also share gain that right now, we serve all 3 major manufacturers of data center class hard drives. And that's the part of the market that's growing most rapidly. So that's kind of our core.

On top of that, we have a business doing system-level test for mobile. That's another case where we've gained share. We've won additional customers. The market itself is not growing robustly in units. So we're doing mostly helping to facilitate next-generation parts on existing platforms, but we expect that to provide growth over the next few years. And then the thing that we're working on right now that we do expect to drive strong growth over the next few years is system-level test and burn-in for AI-related products.

So we're doing that for some AI accelerators for hyperscalers. We're also doing that for edge AI products that are going into applications like automotive. So very aggressive temperature ranges and very thorough testing to try and make sure that they are delivering products at a higher quality level than you get from traditional test. So those are sort of the 3 existing areas.

The last area that we are just getting into in 2026 is in the test of solid-state drives. So we test hard disk drives and we test flash memories. Before we were not testing the actual SSD. The transition of SSDs to PCIe 6 is creating a discontinuity and increasing the test intensity. So we see an opportunity there as well.

Great. And just in the last couple of minutes we got left, I want to ask you about the financial management of the company. Clearly, you've got a very strong path for growth over the next several years. How are you managing that both -- I think gross margins, 60% kind of like kind of where you've been landing. But how do you think about the OpEx growth thesis for all the things you just talked about while still getting financial leverage in the model?

Yes. So from a gross margin perspective, our margins have -- like as we've grown, we've definitely gotten sort of into this 59% to 61% range. And I would expect it to continue to operate in that range with both tailwinds from cases where we have differentiation and we get more pricing power and headwinds from material costs. So I don't expect to see a significant trend in gross margins.

Around OpEx, we have a general rule of thumb that we want to constrain our OpEx growth to less than half of our revenue growth. And obviously, if we're growing revenue very, very rapidly, that's going to be on a lagging basis. So we expect to get significant leverage out of operating expense as we get to be a larger company. But many parts of our business are really fueled by the investments we're making in OpEx. If you look backwards in time, as the semiconductor industry pivoted hard towards AI, we leaned into OpEx at the same time that we were seeing headwinds in our mobile business.

So we leaned into the investments that we needed to make to be able to win in compute eventually. And we think that, that was the right decision to make. And we're still in a situation right now where the marginal utility of additional investment is very high. So that's the stuff that's winning us a second hyperscaler. That's what's going to allow us to expand share inside of merchant compute and align to silicon photonics. So we certainly don't think we need to grow OpEx at the same rate that we're growing our revenue. But the increase that this 50-50 increase thing, we definitely have opportunities to spend on that, that are going to deliver growth.

Excellent. I think with that, we're on time. Thanks, Greg, for being here with us. We appreciate it.

Awesome. Thank you.

Ladies and gentlemen, good morning, and welcome to the Teradyne Second Quarter 2026 Earnings Conference Call. [Operator Instructions] As a reminder, today's call is being recorded.

I would now like to turn the call over to Amy McAndrews, VP of Corporate Relations for Teradyne. Please go ahead.

Thank you, operator. Good morning, everyone, and welcome to our discussion of Teradyne's most recent financial results. I'm joined this morning by our CEO, Greg Smith; and our CFO, Michelle Turner.

Following our opening remarks, we'll provide details of our performance for the second quarter of 2026 and our outlook for the third quarter. The press release containing our second quarter results was issued last evening. We are providing slides as well as a copy of these prepared remarks on the Teradyne investor website that may be helpful in following the discussion. Replay of this call will be available via the same page after the call ends.

The matters that we discuss today will include forward-looking statements that involve risks that could cause Teradyne's results to differ materially from management's current expectations. We caution listeners not to place undue reliance on any forward-looking statements included in this presentation. We encourage you to review the safe harbor statement contained in the slides accompanying this presentation as well as the risk factors described in our annual report on Form 10-K for the fiscal year ended December 31, 2025, on file with the SEC. Additionally, these forward-looking statements are made only as of today.

During today's call, we will refer to non-GAAP financial measures. We have posted additional information concerning these non-GAAP financial measures, including reconciliation to the most directly comparable GAAP financial measures, where available on our investor website.

Looking ahead, between now and our next earnings call, Teradyne expects to participate in technology-focused investor conferences hosted by Goldman Sachs and Citi. Our quiet period will begin at the close of business on September 11, 2026.

Following Greg and Michelle's comments this morning, we'll open up the call for questions. This call is scheduled for 1 hour. Greg?

Good morning. For the second quarter in a row, we delivered record revenue, and once again, AI was the driver. Total company revenue topped $1.3 billion, up over 100% year-over-year with non-GAAP EPS of $2.47, up over 300% year-over-year. All 3 of our business groups, Semi Test, Product Test and Robotics grew year-over-year and quarter-over-quarter, highlighting the AI demand across all parts of the business. At more than 60%, AI-driven revenue is the key proof point that our wafer to AI data center strategy is delivering results.

There are many superlatives that we could use to describe this quarter's results, and Michelle is going to walk you through the details. I'd like to use my time today to set context for what we believe is a multiyear growth phase for our entire business, driven by the continued AI build-out. Certainly, this is clearest in compute across CPUs, accelerators, networking and in memory for HBM and DDR. However, data centers are now also the primary growth drivers for flash memory, hard disk drives, power, board test, high-speed interconnect and Robotics. The long time line of data center investment plans has given our SOC and memory customers the confidence to aggressively invest in wafer fabrication equipment for new process technologies and additional wafer capacity.

For years, semiconductor capital investment grew slowly and ATE grew even more slowly within it. That's changing. Two factors are now tailwinds for the ATE TAM. The first is overall semi-cap investment, particularly wafer fab equipment, WFE, which has begun accelerating. More WFE means more wafers and new equipment generations enable denser process nodes. More wafers and higher density together mean more transistors to test in SOC and more bits to test in memory. WFE CapEx is now forecasted to approach $250 billion by the end of the decade, driving 5% to 10% annual growth in 300-millimeter wafer production and 15% to 20% CAGR in total transistor production over the midterm.

Total non-memory transistor production is a reasonable first order predictor of the SOC TAM. Every transistor must be tested in a given process node yields roughly the same transistor count per wafer regardless of device type. Memory bit production is the analogous predictor for the memory TAM, also forecast to grow at a 15% to 20% CAGR over the midterm. Neither of these predicts the ATE TAM precisely in any given year, but both signal direction. Transistor and bit growth rates have inflected upward and are expected to hold a steeper slope through the end of the decade.

The second factor is advanced packaging. Wafer transistor counts don't vary with die size, but the required test volume scales with acceptable quality level for the devices being tested. As accelerators, CPUs and networking devices pack in more chiplets, memories and eventually CPO, a single latent defect has a greater impact on final device yield, raising test intensity per die in multichip packages. Both package volume and dies per package are forecast to keep rising through the end of the decade, providing a sustained tailwind to the compute TAM. The result, after more than a decade of semi-cap equipment outgrowing the test TAM, the trend reversed in 2024 and test is now outpacing fab equipment.

With WFE long-term forecast becoming clearer, we're increasingly confident in long-term ATE growth. As WFE CapEx approaches $250 billion by the end of the decade, we see a path for overall ATE TAM to reach or exceed $20 billion. In addition to being in a growing market, we are confident that we will be able to gain share over the midterm. With the ramp of merchant GPU and the initial dual platform qualification in our second major hyperscaler, we are positioned to gain share in the core compute segment. Our leadership position in HBM and DRAM maximizes our exposure to DRAM-driven growth and segments where we have historically had high share like flash memory, mobile, industrial and automotive are positioned for growth driven by cloud, edge and physical AI, whether it's in cars, robots, phones or wearables.

Before I hand off to Michelle, I'd like to share a few demand highlights and their impact to share gains. First, in memory. Demand has increased from strength in HBM and DRAM and a resurgence in the NAND final test. The 2026 memory TAM is likely to be more than 40% larger than 2025 with notable growth from the first half of 2026 to the second half. A highlight in memory is the growth in HBM-based die test. Our Magnum testers have logic test capabilities that provide attractive swing tool advantages for memory makers.

Now on to compute. In the first quarter, we received our first order from -- for a merchant GPU customer, which was delivered in the second quarter. Also in Q3, we completed correlation at a second hyperscaler, increasing our confidence in 2027 market share growth. As we discussed last quarter, our expectation is that our compute business would be concentrated in the first half. We have multiple programs across networking and hyperscalers that built out capacity in the first half that is now being utilized. The next surge for these customers is expected to be in the first half of 2027.

Since dual vendor strategies are emerging at the largest compute customers, I'd like to give you a bit more color about how we think this will play out. These customers recognize the importance of derisking their supply chain. We see this as an important share driver over the next few years. The dual vendor qualification process generally runs through 4 phases: the first, an opportunity to compete; second, developing a working solution; third, correlation; and then finally, a production ramp. If successful, it takes 9 to 12 months from the start to the ramp. After the initial dual-sourced part ramps, we enter a fast follower phase where additional parts are converted and ramped.

Ultimately, the account reaches a mature dual vendor stage where either platform is used for initial part release. It can take a few years to get from fast follower to mature dual vendor. During the fast follower phase, we expect market share to grow incrementally from 0 up to about 30%. Right now, we have 1 compute customer in the mature dual vendor phase, 1 in fast follower and 1 in qualification prior to ramp.

I'd also like to add a little color around networking. Frontier models require ever larger accelerator clusters. This is driving extreme growth in networking for scale-out, scale across and scale-up applications. While the total number of network connections is growing fast, the timing of transitions from cable to backplane and from pluggable to XPO and CPO is in constant flux. Our belief is that there is robust growth in all of these technologies. This drove our acquisition of Quantifi Photonics, the development of the Photon 100 solution for optical connections and our multilane test products JV for copper connections. We are working with multiple ecosystem partners to develop leading-edge solutions from silicon photonics wafers to full data center racks. We expect CPO alone will be a $300 million to $700 million market by 2028.

Our IST business grew revenue 2.5x quarter-over-quarter on strength in HDD fueled by AI. We expect continued growth of our IST business in the second half. Given the strong forecast for greater than 20% annual exabyte growth and our design wins, we are confident this business will grow over the midterm as well. The data center build-out will drive growth for our product test and Robotics groups through the end of the decade as well. The logic chain is straightforward. More data center construction drives more rack shipments per year, which is driving growth at contract manufacturers and original design manufacturers.

This, combined with the rapid advancement of data center architectures, creates a significant opportunity for Teradyne to provide solutions well matched to the volume, quality and flexibility data center applications required. This is best reflected in the total available market for automation and test among contract manufacturers and ODMs. We believe that there is currently a multibillion-dollar market for assembly automation, test and burn-in equipment, and we expect double-digit -- mid-double-digit growth rates through the end of the decade.

By addressing these applications with enhanced production board test, optical test, backplane test and robotic-assisted test and assembly, Teradyne is uniquely positioned to follow the value chain from wafer to data center. Our wafer to data center strategy is working. Our optimism around 2026 and 2027 and through the midterm has grown. We are leaning further into investments to capture opportunities across the value chain, both organically and inorganically. And we are investing in next-generation products across our entire portfolio.

As we win business, we build out customer teams for major hyperscalers and semiconductor suppliers. The fact that we are leaning into these investments now is a sign of our confidence in the sustainability of this market growth. We expect 2027 to be another year of healthy growth for Teradyne, consistent with the transistor and bit growth dynamics I described earlier. It's clear to us that increases in WFE spend will be a primary driver to ATE TAM growth, and this sets the approach for how we will be updating our target earnings model, which we will share in our Q4 earnings call.

With that, I'll turn the call over to Michelle.

Thanks, Greg. Let me build on that with the detailed results for the quarter, starting with total company performance. Both revenue and non-GAAP EPS came in above the high end of our guidance range as strong AI-driven demand continued across all parts of our portfolio. Sequentially, total company sales were up 4% from last quarter's previous record. For the first half of 2026, we delivered $2.6 billion in revenue and $5.02 of non-GAAP EPS, up close to 100% and 275% year-over-year, respectively, driven by all things AI.

Building on that, let's take a deeper look at revenue starting with Semi Test. Our Semi Test team once again cleared the $1 billion high watermark established last quarter with revenue up $11 million from first quarter and 128% from Q2 2025. The revenue breakdown within Semi Test was SOC of $843 million, memory at $212 million and IST at $67 million. As expected, compute order timing was more than offset by another record memory quarter and IST growing over 150% quarter-over-quarter.

Within SOC, compute remains the largest portion of our SOC product revenue at 70%. Compute revenue grew nearly 600% year-over-year on strong AI-related demand. In the quarter, as Greg mentioned, we have completed correlation with a second AI hyperscaler customer, and we shipped the previously announced merchant GPU order. This further diversifies our compute portfolio, creating a foundation for future market share gains over the midterm. Auto and industrial continued to strengthen over last year, driven by power management demand increases for AI data center build-outs. Mobile grew seasonally quarter-over-quarter double digits, though it remains below historical levels and a muted part of the overall SOC portfolio.

Now turning to memory. Our memory business delivered another strong quarter at $212 million in revenue. This is another record quarter up from the previous one set in Q4 2025. This represents our third consecutive quarter of revenue over $200 million, driven by robust HBM and DRAM test solutions demand and a resurgence in NAND. Demand signals remain strong as memory manufacturers are planning capacity additions further out in time, driving our book-to-bill ratio in the quarter over 2. Finally, turning to IST. Revenue in the quarter was $67 million, up 94% from the prior year, driven by AI-related HDD storage demand from all 3 major suppliers in this space.

Now on to Product Test Group. Revenue was $107 million, up 26% year-over-year and 33% quarter-over-quarter. The group experienced broad-based growth across all end markets from production board test to optical test to defense and aerospace to scale-up networking at our newest portfolio addition, the Multilane Test Products JV, also known as MLTP. Last quarter, we announced our new production board test platform, Omnyx, which is focused on enabling earlier detection of defects that are impacting the build-out of AI data centers. The initial customer traction is strong with units already shipping in the second quarter and continued growth expected in the second half.

Similarly, momentum is building in MLTP as the need for high-speed I/O and data center interconnect test solutions is increasing. Both Omnyx and MLTP are examples of our wafer to AI data center strategy in action, both solutions focused on solving our customers' most critical pain points along the value chain. As a result, we anticipate their continued growth in the second half of the year.

Robotics revenue was $100 million, up 33% year-over-year and 9% quarter-over-quarter. Electronics manufacturing and semiconductor revenue increased by 50% from Q1 and is now the largest end market segment in this group, which includes AI data centers. Our U.S. sales increased to 32% of the Robotics overall sales. Aligned with this, our U.S.-based manufacturing center is on track for opening later this year.

Now moving down the P&L. Strong earnings results continued in the second quarter, driven by robust AI-driven volume and favorable product mix. Gross margins for the quarter were 59.8%, up 250 basis points year-over-year, driven by strong Semi Test volume and product mix, while sequentially, gross margins were down 110 basis points, driven in part by onetime benefits in the first quarter. OpEx increased as expected, driven by more R&D and go-to-market investment for 2027 growth, plus higher variable compensation on stronger results. Finally, non-GAAP operating income was $448 million, with an operating margin of 33.7%.

Now moving to capital allocation. Our first priority for capital continues to be reinvesting back into the business, aligned with the growth opportunities we see along the wafer to AI data center value chain. That means continued R&D investment in innovation and next-gen technologies as well as continued investments in cost of goods sold and CapEx aligned with scaling our operations and supporting our current and future customers with demo assets.

Beyond that, our capital allocation strategy remains consistent, maintain cash reserves to run the business and keep dry powder available for accretive M&A. We ended the quarter with cash and investments of $517 million, up over 30% from last quarter. We had free cash flow of $378 million in the quarter. For the first half of 2026, free cash flow was $579 million, up 150% from the prior year period.

Working capital, predominantly inventory investment increased in support of future sales. As discussed last earnings call, capital expenditures increased $26 million from last quarter, driven by continued investments in innovation and operations scaling. We paid $20 million in dividends in the quarter, and our share buybacks were $69 million.

Now looking ahead to our third quarter guidance. For the quarter, we expect revenue in the range of $1.2 billion to $1.3 billion and non-GAAP EPS of $1.85 to $2.15. Gross margins are expected to be in the range of 58% to 59%, reflective of product mix and new product launches. Operating expenses are expected to run at approximately 29% to 30% of third quarter sales, driven by continued investments in R&D and go-to-market. The non-GAAP operating profit rate is expected to be between 28% and 30%.

Based on current customer order visibility, we are updating our first half weighted revenue to 50% to 52% of annual revenue. The outlook for the second half has strengthened, and we've narrowed the range from 3 months ago with increased visibility and continued robust demand signals from our customers. The range also takes into account the timing of orders that could impact revenue timing across quarters or years.

To aid in your modeling, we expect growth in second half in comparison to first half in memory, auto and industrial, IST, Product Test and Robotics groups. We expect this growth will be offset by softness in mobile and order timing in compute. We anticipate a resurgence in growth in 2027, driven by overall ATE TAM expansion and market share gains. Aligned with the strength in demand signal, we expect OpEx in Q4 to be comparable to Q3 guidance, positioning us for further growth in 2027.

So in closing, once again, our team delivered exceptional financial results, meeting our customers' increasing AI demand needs through strong execution and a maniacal focus on customer satisfaction. We remain confident in achieving our target earnings model at an accelerated pace and look forward to providing an update in our Q4 earnings call.

I'll close by thanking our Teradyne team for their execution and discipline this quarter, delivering for both our customers and shareholders.

With that, we'll open the call for questions. Operator?

[Operator Instructions] Our first question today comes from Timothy Arcuri with UBS.

Greg, in correlating the test TAM to the WFE TAM, typically, it's been about 8%. And if I look at your TAM slide, it looks like the upper end of the range is like $16 billion to $17 billion for next year when some of us like me think that WFE is going to be $200 billion. So it seems like that number holds next year. And when I look out to 2028, I mean, some of us think WFE is going to be $250 billion. So that's like $20 billion, which is quite a bit higher than even the upper end of what this slide shows. So I guess the question is like, is 8%, like in your mind, when you think about it, is 8% the right number? And what are the puts and takes on that?

Yes, Tim. First of all, I want to thank you, the investor meetings that you helped mediate for us recently really gave us a ton of insight in terms of how people were thinking about our market and our model. And so we've been going back and doing some homework on this. The thing that you're bringing up in terms of that 8% is a really interesting trajectory that if you look back to 2023, it was down near 4% of the total CapEx spent on test equipment. And by 2025, that had gone up to 7%. The first 5 months of 2026, it's 8% of total semi CapEx has been on test equipment.

So it's kind of this rocket trajectory. As I said in my prepared remarks, there are some reasons why we think that is happening connected to advanced packaging. But right now, we're not sure whether it's going to settle -- that we think it's going to settle somewhere in the 7% to 9% range, but it's not going to continue to go up from there. So we're -- like I think we're kind of thinking it could settle down in the 7% to 8%, but it could go up to 9%.

Got it. Okay. Great. And then from your slide, it looks like you think the range of like 12% to 14% is it's pretty similar to the new TAM range that Advantest put out last night because they don't include burn-in in their TAM. If I use those TAM numbers and I use your loading for the back half, which basically implies that C Q4 is basically flat Q-on-Q. I get your test share for this year is up just a touch, but it's still basically flat sort of in the 37% range. So the question is, when do all these new [ quals ] start to add to share gains? You talked about these [ quals ]. And like what's the right long-term share to use? Like is 40% your sort of aspirational share? Is 42% the right number? And like when will these start to show up in share gain?

So, sort of taking that bit by bit, I think that the numbers that Advantest put out for the total TAM, both SOC and for memory are kind of in the right ZIP code that -- we were surprised when they hadn't brought it up last quarter because it seemed like the writing was on the wall that the TAM was increasing pretty strongly. The other commentary that they had was that they believe that they were going to continue to gain share in SOC and that they were going to lose a bit of share in memory. I have to say I agree with their commentary about memory. But I think in terms of SOC, it's probably going to be pretty flat, maybe a slight incremental gain for us year-on-year.

So I would say that the -- we are gaining share in most of the segments of the market, but the part of the market that is growing most strongly is the part of the market where we're starting with the lowest share. The important thing for us in 2026 is that our share in the compute segment is actually stabilizing and starting to inflect upward.

So I would say, to answer your question directly, I think you'll start to see some of the effect of share change in 2027, but it's going to be gradual. It's a socket-by-socket thing, and we are riding a big upward wave.

We'll take our next question from Mehdi Hosseini with SIG.

Two from my end. Greg, when I look at the way you are laying out the company's strategy, especially how you have illustrated in your slide, it seems like you're looking at this as holistic. Starting from wafer and going all the way to the rack and data center. And what I wanted to ask you is to what extent are you still focused on consolidating the test insertion, especially before we get to the complete package? Are you facing challenges? Or is that -- does that get lower priority and you're more focused on a holistic approach that you've laid out in your slide?

So I think the -- as you look at this wafer to data center journey, there's a through line of Teradyne is a company that tests things and our Robotics help build things. And just about at every step of that process, we are a participant in an ecosystem. So when we're building test equipment, it's being put into test cells with material handling equipment that come from a number of different suppliers. The same thing is true at every step. And so we're trying to be very mindful that our customers appreciate what that open ecosystem has delivered. So the best example of that ecosystem developing is really in the whole silicon photonics space that it's early days where there's individual partnerships sort of against the initial customer ramps. But ultimately, that's going to turn into the same open ecosystem where all of the -- like each test company works with all of the suppliers of material handling and vice versa.

I don't know if that answers your question. I do think that there is some potential pull-through or customer value that customers can get by getting more of their test from the same supplier. So in terms of leveraging the work that's done from wafer sort to final test to system test to burn in into the board test and beyond, there is some advantage to being able to leverage things across those boundaries. But our customers are quite adept at choosing the best solution at each stage. So we have to really compete for everything that we're getting. Did I get what you were talking about?

Yes. I'm not sure if there's a very concise answer. Is the evolving supply chain with some disruption as to how it's all going to come together is to be determined. Does that summarize your answer?

So I think the supply disruption is definitely on the mind of everybody in this space that we're all only as strong as the weakest link in our whole supply chain and our customers feel the same way that -- and that's one of the key things that's underpinning the drive towards a dual vendor strategy from players in the compute space. So they are working to qualify multiple test suppliers, because they want to be sure that they're going to be able to get the capacity that they need when they need that capacity. And it's not just a matter of how big any one test supplier can get. It's also that they need to have the assurance of different suppliers to be able to do each step in the chain.

Got it. And I don't want to monopolize your time, but if I may ask my second question, which has to do with Robotics. This is a part of business that is not getting as much headline, but it is part of the AI is one of the verticals. Has there been any update to the strategy? I know you have been expanding capacity in the U.S. to capture some of the opportunities there. But how should we think about the trajectory of AI recovery, especially as it fits into your longer-term model?

So Robotics, we believe that Robotics is positioned to essentially grow in proportion with the rest of the company over this midterm. And the important stuff that's going to drive that growth are physical AI applications. Our fastest-growing segment in Robotics is really electronics, manufacturing and semiconductors. And that's like -- as Michelle says, that's part of the all things AI that the data center build-out and the desire to reshore a lot of the data center build-out is pushing automation of the assembly process and the test process with a fair amount of budget behind it. So we think that Robotics is going to basically follow the same kind of growth path as all of the chunks that are connected to data center build-out.

Our next question comes from C.J. Muse with Cantor.

I guess first question on memory. You talked about growth half-on-half. I'm curious if you could speak to any changes in the drivers there, whether you're seeing an uplift in NAND and D5 versus HBM? And then perhaps more importantly, how are you thinking about memory into '27? And should we expect lumpiness in the first half? Or will that strength continue in the first half?

So this year, I think the -- coming into the year, we expected that HBM was going to be significantly stronger that if you remember, 2024 was a big, big year for HBM. 2025, there was a little bit of digestion, but it was clear from the long-term capacity add plans for HBM that '26 was going to be very strong. The thing that surprised us a little bit was just how strong the DDR business is. And I think that is, to a certain extent, coupled to the resurgence in like CPU-oriented applications, so Agentic AI. There's also use -- like some of the like SOCAMMs use LPDDRs that are driving that market harder than we would have expected. So the DDR market is stronger now than we thought it would be in January.

In January, we were talking a little bit about that we anticipated that we would see a strengthening in NAND, but we didn't see -- we didn't hear that from our memory customers yet that like we were waiting to hear that they're going to need more capacity in NAND because there was a gap between what data center demand was and what memory capacity planning was providing. Like what's different now is we actually have seen the beginning of that inflection and are actually getting pull from our customers for more NAND capacity.

Looking into 2027, I don't think we are looking at particular lumpiness in memory. I think that, that's going to be -- the capacity adds are -- there's significant WFE investment increase in memory. There are fabs that are coming online. There are wafers that are being planned for output, and they are all making their test capacity add plans out into 2027 now.

Perfect. And I guess as a follow-up, if you could speak to gross margin and the guide and driving -- what's driving kind of the 130 bps headwind. You talked about kind of new products, but would be curious how much is from new products? How much is just kind of a mix shift? And then any help into thinking the trajectory into December and '27 would be very helpful.

C.J., it's Michelle. So I'll start and then Greg, feel free to add any additional color. So just taking a step back, let me share some context, especially for some that are new to the Teradyne story. So overall gross margins within our business tends to be variable quarter-to-quarter. So if you look at past like the last 5 years, you'll see about 400 basis point swing when you look quarter-on-quarter. However, when you look at kind of the annualized gross margin amount, we tend to be pretty tight within about 200 basis points. So part of what you're seeing first half to second half is just kind of the normal quarter-on-quarter variability that we experienced. But we also did have some nonrecurring nonoperational impacts that were favorable in the first half that aren't repeating.

So as you look at going from Q2 to Q3 in the guide at 58% to 59%, part of that is product mix. So coming off the high volumes that we had in Semi Test and compute as we're starting to ramp the other parts of our portfolio. So just going back to our opening remarks, we are going to have more auto and industrial in the second half, more IST, more Robotics as well as more Product Test along with memory.

And so all of that netted together gets us to a full year gross margin range, which is just shy of our target earnings model, so right around the 59%. So there's going to be a product mix element. There's going to be the new product introductions. And there are going to continue to be some pricing elements as well as we know and you guys know as well, memory is going to continue to be a strain from an overall margin perspective, and we expect that, that will continue into 2027.

We'll take our next question from Vivek Arya with Bank of America Securities.

Greg, I just wanted to dig into how tight this correlation is between WFE and your growth prospects. When I look at the current year, your sales are roughly growing twice as fast as WFE growth this year. So I'm curious as to why that is. And then if I were to carry that argument into '27, if WFE does indeed grow 30%, what scenarios would make you grow faster or slower than that pace?

So the correlation is -- I would say the correlation is strong over kind of 3- to 5-year periods that it's not at all correlated when you're down to a quarter level. And when you look at the annual data, when we sort of drew out the charts ourselves, we saw that there can be like a year lag between a significant increase in WFE and when it echoes in the impact in the ATE space. So it's -- there's like a time lag in the correlation as well.

So I think it's -- like that's why we're saying it's not a great year-by-year predictor, but it does -- the thing that it does is it gives us confidence that we're not -- we were trying to figure out whether 2026 was like an unusually strong nonsustainable market. And when we were looking at the WFE data, it certainly gave us confidence that we were looking at something that was different than what we've seen in prior cycles, that there's this very heavy lean into WFE investment, and that's going to drag things up. So that's the correlation part. Could you repeat the second part of your question? I want to make sure that I get it right.

Sure. So if we look at 2027 and the expectations are for WFE growth to be, say, 30% plus year-on-year, under what scenarios would ATE TAM or Teradyne for that matter, over or undergrow that growth rate?

So I think -- right now, we have -- like we're outpacing WFE growth in '26. So if you draw the line and sort of use the WFE spend to project what the 2026 TAM should be, it should be smaller than that model would predict. Looking ahead to 2027, I think that WFE is essentially catching up a bit. And the way you sort of think about that in the statistics is that the percentage of semi cap that is test equipment has gone from 7%. And so far in 2026, it's at 8%. It's possible that next year, that could revert down to 7% or so. So I would expect that we're going to see an increased TAM in 2027, but we're still trying to work out kind of whether that percentage is going to stay flat or how far it will go down.

I guess my question is to why. Why would it go down?

Just, it's the time lag. The thing is -- this is correlation, not causation, right? Or well, there is causation, but there's a time lag. So when people buy fab equipment, the money changes hands about a year before the wafers start coming out of the factory because they have to go into front-end facilities that are being built and commissioned and piloted. Once those wafers are coming out, there's about a -- our lead times are in the 16-week-ish range. So it's 3 quarters on from when the revenue for the fab equipment has happened to when the revenue for the test equipment happens.

And there's also unpredictability about the required test intensity on a part-by-part basis. So you can have situations where due to yield issues or quality issues, there's significant over buying in ATE or you can have other cases where there's higher efficiency test time reductions. So you're not going to see a tight-tight link between this in any particular period, but it's more of an overtime guidance.

And for my follow-up, Greg, I'm curious to understand your position when it comes to CPU, server CPU testing, right? That has become a new area of growth, right, $200 billion-plus type TAM that different people have mentioned. So I know historically, Teradyne has been more exposed to ARM rather than the x86 ecosystem. So how do you see that developing? And if ARM CPUs take share, does that help Teradyne also take share in that market?

Yes. So we certainly are seeing increased strength in CPU business. We definitely benefit more like as the share mix of server CPU shifts towards ARM, that gives us a greater share gain opportunity right now. But we are working hard to try to gain share in the x86 space as well. But I think it's a fair assumption that if that market twists more towards ARM, that positions Teradyne for greater share gain in the compute space.

Our next question comes from Krish Sankar with TD Cowen.

I have two of them too. The first one, I just want to follow up on the CPU side. I understand you're more exposed to the ARM ecosystem, but it seems like the test intensity of the CPUs are 25% that are for GPUs. So is that fair to assume that the merchant GPU opportunities are larger even with one customer than trying to get more on the CPU test side?

Well, the -- it's a great question, Krish. So the accelerator market is definitely the one that has the highest test intensity. And I think your 1:4 ratio is probably not far off. I think that's a reasonable assumption. The key thing is the ratio of accelerators to CPUs. So as the number of CPUs increases, that becomes more important.

The other thing that I think is -- like one of the ways that we're looking at this is there are a number of bottleneck resources through the supply chain, the supply of substrates for CoWoS, the supply of HBM memories and basic advanced node fab capacity. And our customers are trying to optimize the utilization of what allocation they get of those scarce resources. And the test suppliers, Teradyne and our competitors are positioned to -- like no matter what, all of the wafers that get fab are going to need to get tested. And the share between VIP accelerators, merchant GPU and CPU, all of those are places where we are positioned to gain share. And so we are not overly hung up about whether CPUs grow faster than GPUs because at the end of the day, most of the test seconds are at the wafer level and the same number of wafers are going to be produced.

Got it. Got it. Very helpful, Greg. And then as a quick follow-up, I had a question on the silicon photonics testing. Correct me if I'm wrong, but I think the general view is that Advantest is strong in insertion 1 and you're more strong in insertion 2. A, is that correct? And part B of the question is there is some view in the industry that insertion 2 can be skipped if you have known good die and you're going to dice it? Any thoughts on that would be helpful.

Sure. So I want to tell you a funny anecdote. We were having a conversation with a customer in this space. And we were a good 15 minutes into the conversation before we discovered that we and the customer had an entirely different definition of what insertion 2 was. So like there's certainly a pilot customer in this space that is ramping, and there's a certain definition of the different test insertions for that customer. But it's so early in the maturation of that production process that I think it's a mistake to draw broad conclusions.

But directly to your question, getting a known good die at insertion 1 doesn't help you all that much in terms of the quality of the optical engine die that you need to attach to the CPO. The production steps between insertion 1 and insertion 2 are the -- bonding the electronics IC to the photonics IC wafer and then applying the lens, there's a polymer lens that gets applied on top of the photonics to make the connection with -- to where the FAU will connect to the device. And the types of optical testing that you can do with insertion 1 are limited because the lens assembly is a critical part of the optical path. So many of the loss and polarization tests that you need to do can't be done until you have the completed wafer.

The real question on our mind is how much of that testing needs to happen at the wafer level post bonding and how much of it can be deferred to do once those optical engines are singulated. But you can't do everything that you need to do for -- to validate the quality of the device at insertion 1.

We'll go next to Jim Schneider with Goldman Sachs.

Maybe following up on the -- on your prior comments and last question, Greg. You talked about the sort of CPO TAM alone being $300 million to $700 million, I believe, in 2028. Obviously, that's a very wide range. I'm sort of curious how you would think about maybe what the size could be or likely be next year, if at all, or if material? And maybe talk about the sort of certainty level you have in your overall networking business growing sort of strong double digits over the next 3 years on a compounding basis or at least every year.

Okay. So the wide range, the $300 million to $700 million in 2028 is -- that's our attempt to be as honest as we can be. Like we really are looking at a number of different market sources in terms of the number of CPO ports that are going to be produced, and they're all over the map. And so we're -- it really comes down to how quickly the initial higher volume ramps of CPO primarily for scale out achieve success over the 2027 time frame. That's what's going to drive that 2028 TAM towards the -- if that goes well, then it will drive it towards the $700 million. If it doesn't, it will still be closer to the $300 million.

Next year, I think we're probably like probably aiming towards more of the low side of that $300 million. So if you draw a line from the $100 million to the $300 million next year, kind of the low end would be in the $200 million range. I don't think there is as much upside next year as there is upside in 2028.

Now your last question was around networking and networking in general, is an area where we have a lot of confidence of better -- like we're talking about transistors growing at 15% to 20% over this midterm. We feel pretty strongly that the networking TAM is going to be growing proportionate to that. So -- and it's not just CPO. It's copper connections going from cable to backplane, pluggables are still growing and will transition not just to CPO, there's NPO and XPO stages along that path.

And so Teradyne's strategy is really to make sure that we're positioned to benefit from the growth in all of those segments. So that's why we did Quantifi. That's why we did the MLTP JV. And it's why we are in the market with the Photon 100 because we think that CPO is going -- is the technology of the future, but there is so much growth in just basic networking that we really want to make sure that we're covering all the modes.

The other thing that I'll say is behind all of that, is the networking silicon, the switch silicon. And Teradyne has a terrific position in that market that we think will grow over the midterm independent of the physical layer.

That's very helpful color, Greg. And then maybe just a quick clarification for Michelle. Relative to your prior commentary on gross margin sort of variability, I just want to make sure that you don't see anything in 2027, whether that be customer mix or product mix, et cetera, that would drive a material headwind to gross margins on a year-over-year basis?

No. It's going to be pretty consistent as we talked about in terms of the ranges annually. And it will really come down to, Jim, the things that we noted, it will come down to product mix. where compute is going to be heavy within the year, along with the rest of the portfolio growing. So we're not anticipating any fundamental changes in gross margin.

Our next question comes from Shane Brett with Morgan Stanley.

My first question is on memory. So if I assume your memory test revenue growth half-on-half, we get to full year growth north of 70%. Slide 6 of your presentation deck shows the memory test TAM growing at approximately half of the rate of memory bit growth. Two-part question here. But one is how should we think about memory test growing 2x bit growth this year? But also how relevant is test growing at half the run rate of bits given the test intensity of HBM and just extremely low NAND in the last few years?

I think I'm going to need to take that question offline. You're interpreting a lot of information from that chart that I'm not sure we were intending to communicate. So the chart shows sort of the bit growth trajectory and the ATE TAM history. We weren't trying to project the memory ATE TAM.

So -- but I guess to answer the principles behind what you're saying, in memory, there is a correlation between bit growth and ATE, but it's not as -- like there's a game, there's a factor involved in that because the -- in memory test, the part of the market where you test every bit is served by much lower priced, less differentiated equipment than the final test part of the market. So there's a big chunk of the market that follows bit growth. There's also a part of the market that follows technology shifts, the HBM3 to HBM4, DDR5, DDR6 and then next generations of flash technology.

So memory has sort of 2 flywheels. One is bit growth and the other is technology change, but it is also a very efficient test market where -- just in terms of the overall size, right now, the SOC TAM is 5x the size of the memory TAM even though the memory revenues are much -- have caught up because of ASP changes. So there's a lot of things going on. Our take is that the memory TAM is likely positioned for similar growth rates over the midterm that the SOC TAM is.

Understood. And for my follow-up, so you talked about auto industrial being better half-on-half. Just can you talk a little bit about what the path back to prior peak would look like? I would also appreciate if you could talk about how much of prior peak was China driven, if there's sort of an element of China localization risk, but also some of the tailwinds you're seeing just on the back of the Infineon strategic partnership back in January of 2025.

So, I think we are -- it's likely '27 is going to be up to or exceeding prior peaks in that space. We haven't done all of the work, but that's my gut feeling around that. There's some China localization, like I would say that our China power mix between prior peak and now is kind of about the same that we're in that market. There is some local competition, but there are also some players that we're doing quite well with.

In the power market overall, there -- the part of that market where the Infineon deal that we did is most helpful is in wide-bandgap discrete test. So that's a smaller chunk of that market, but one that is expected to grow robustly over this period of time. And I would say that the technologies and the people that we have in that group is positioning us to establish a real leadership position in the testing of wide-bandgap discretes.

Thank you. At this time, we've reached our allotted time for questions. This will conclude today's Teradyne Second Quarter 2026 Earnings Call and Webcast. You may now disconnect your line at this time, and have a wonderful day.

Welcome back to the session at the BofA Global Tech Conference. I'm Vivek Arya Karia from BFA semiconductor semi-cap equipment. -- research team. I'm really delighted and honored to have the team from Teradyne join us, Greg Smith, the President and CEO. And as always, I'll go through my questions, but please feel free to raise your hand if you would like to bring up something.

Greg, really warm welcome. Really happy to see you at our conference.

Great to be here. Thank you.

Wonderful. Let's just start with kind of a state of the union. And if you could help us frame the addressable opportunity, right, for automated test equipment. And I say that because it seems the semiconductor industry has -- is close to achieving its $1 trillion kind of milestone much, much sooner than we thought wafer fab equipment demand is very strong.

How do we connect the dots, right, from those 2 data points to what the opportunity is for you and how it's going to develop if the industry semi industry gets to, I don't know, say, $2 trillion in a few years?

$2 trillion. Okay. So there's -- if you look just over the last couple of years, you're definitely seeing an inflection in the test equipment buy rate. if you look back in history, the semiconductor buy rate has been hovering around or slightly below 1% of semiconductor revenue. And over the past couple of years, we've actually seen it begin to inflect.

And that's pretty amazing given the fact that the fastest-growing part of the semiconductor market, sort of the 3 fastest-growing parts of the end semiconductor market are GPU which is a very high margin part; HBM, which is the highest margin part and a shortage of memory, which has resulted in a significant increase in revenue without a significant increase in the number of parts being tested.

So like the setup for buy rates to actually go down was actually strong because the inflection in revenue was a lot higher than the inflection in unit volume. So it's pretty remarkable that we've seen this kind of an increase in the test intensity along with the increase in the revenue to sort of allow that buy rate to go up.

Now the other way to look at the test equipment opportunity is as a slice of a pie of semiconductor capital equipment in total. And that's a market that back in the old days would be $100 billion. Now it's well on its way to $200 billion. And we've typically -- in a $100 billion semi cap environment, we would have gotten a TAM of $5 billion to $6 billion, like 5% or 6% of that total pie.

As the wafer front end market is expanding, it's early days, but the percentage of that total market that ends up being test could actually get larger. If you think about the -- there have been some times in the semi cap space, where there have been a significant increase in capital intensity without a significant increase in wafer output.

So the two best examples is the transition to EUV, it made that part of the chain a lot more expensive relative to where it was. The other was multilayer flash that like the number of layers in the flash device we're increasing radically the number of flash devices was not increasing at the same rate, and the capital intensity for some process steps was related to layer count.

So those are two examples where you could see WFE spend not turning into test equipment TAM. But now the WFE spend is not really around these technology transitions, it's more incremental technologies, but a lot more wafer output, And more wafers means more test. So that's a driver.

And the other is a greater percentage of all of the wafers being produced, the die being produced are actually being packaged into multiple die end products. So for example, if you are building a DRAM to go into HBM, the test intensity for that is much higher because of the stacking and the quality requirements downstream.

The same thing in terms of an AI accelerator. If you are going to have 2 big compute die plus 6 HBM stacks plus a [ Coles ] interposer; the cost of having a failure downstream is so high that you're willing to invest to get to a higher level of test quality upstream. So that's the whole shift left.

So because of advanced packaging and because of the wafer output, I think we're in a position to see the test market rising at least proportionately with the whole WFE space.

Okay. And WFE this year is expected to be about 140 million, 145-ish billion. And the large WFE companies have not given a forecast. But if, let's say, semi industry does get to $2 trillion, right, we could be talking about $250 billion, $300 billion, right type. So if all we knew was WFP would be $250 million or $300 billion by the end of the decade, what does that tell you? What -- how large they market?

I feel like one of those movie trailers or like in a world where there's $250 billion of front-end investment. So like right now, in $140 billion the -- like if you're talking about 2025, the overall test equipment TAM was about $9 billion.

Versus 110.

Versus 110. For 2025 was $110 billion Yes. So we're talking about like 8-ish percent. So -- and so if you're talking about $250 billion, you'd be talking about easily a $20 billion ATE TAM. So the ATE TAM would go up by 150% from where it is.

Versus the 12% to 14%, right, although I think that...

So like the -- like our -- the model that we -- so what you're referencing is our long-term model. And for that model, rather than projecting a growth rate and giving a model for, say, 3 years from now, we admitted that we don't have great visibility about how fast the TAM is actually going to grow because like we were talking about before we came on stage, like you go to sleep at night, you wake up in the morning and there's a new headline about higher [ in OpEx ] or higher wafer front end.

So we don't have great visibility about how fast the overall semiconductor market is going to grow. And therefore, we don't have great intelligence about just how fast the TAM will grow. But what we have some high confidence about is -- if there is a semiconductor ATE TAM of $12 billion to $14 billion, that Teradyne would be a $6 billion company. So like roughly double in size from 2025 in that market. And that would represent both of participating in this growing market, but also increasing share in that market as well.

So we believe that we're positioned over the next few years to be able to grow with that market and gain share. so essentially outgrow the pace of the ATE TAM.

Got it. What is your share now for, let's say, 2025? And -- because in a 12 to 14 just shy of like 50%, right?

No, no. So this is an important math. So if you look at Teradyne, Teradyne has 3 businesses: Semiconductor, our semiconductor test business represents about 80% of our revenue. There's 10% in robotics, 10% in product test. And now in addition to that, there's also a portion of our semiconductor test business which does system-level test, which doesn't actually accrue to the semiconductor ATE TAM, all right?

So the percentage, the share in the ATE space in 2025 was 30-ish percent. And so our expectation in a $12 billion to $14 billion market that we would have 36% of that $12 billion to $14 billion, so 35% to 38% in that range. And the other parts of the company would essentially grow proportionally with that core. So it's not 50% of 12% to 14%. It's more like 36%.

Okay. So what helps drive that share gain? Are you -- is it your investment in specific end markets? Is it that you already have design wins in certain categories that help you drive that share gain?

So the -- like 36% share is nowhere near the high watermark for our share. So we had significantly higher share than that back in the mobile boom in like 2021. And from 2021 out to 2025, there was very little competitive shift in terms of which customers did business with which ATE company.

The thing that changed from that high share point to 2025 was the segment growth in the market. And so the part of the market that grew the fastest was DRAM and compute and especially AI compute GPUs. And those were two parts of the market where Teradyne's core share, segment share was lower than our share in other parts of the market that were not growing as strongly. So there was this colossal shift in the market towards these two segments.

And what's happened since then is Teradyne has actually gained share inside of compute, and we've gained share inside of DRAM and HBM. So we're in this situation right now where these are very big segments. We've increased our share in our position for additional share gain against compute and DRAM.

And the other parts of our business, mobile, power and industrial and flash; are all positioned to have significant growth over the next few years. So we'll have some of the segment shift helping to pay us back that we lost over the past few years.

Got it. Makes sense. So we talk about the GPU, but it's interesting that the CPU has been a more active area of discussion at our conference so far. What's the Teradyne opportunity in CPU testing? And when you hear of these large TAMs, every day, there is a larger TAM than the day before; what does that tell you by way of -- how much of CPU testing has been internal so far? How much has been at your -- like what is your share today of CPU testing? And what we...

Let's actually -- let's start earlier in terms of the test TAM associated with CPU, right? So the easiest way to talk about this is the relative test intensity for an AI accelerator and a data center CPU okay? And right now, that's kind of 4 to 1, in that it takes 4x as much test for an accelerator as it does for a data center CPU. And right now, the ratio of accelerators to CPUs and data centers is probably greater than 4:1, like 4 or more accelerators to each CPU.

So today, like where we are in 2025, say, that the amount of test TAM associated with accelerators is like 16:1 relative to CPUs. Now the CPU renaissance that we're talking about is going to change that 4:1 ratio to a 2:1 ratio or maybe even aggressively a 1:1 ratio somewhere in the future, right? But if you get to 2:1, that means that you have 8x as much TAM associated with accelerators as you do with the CPU.

So like it's a big and important change in the market, and it is a segment that's going to be growing, but it's still not going to be the biggest thing in the Q1.

Now let's talk about share and players. For x86, one of the X86 manufacturers is primarily on my competitor's platform. And the other x86 player had its own test solution that is sourced. We don't expect -- we don't see a near-term opportunity for us to penetrate the CPU player that uses my competitor's equipment.

For the one that uses its own equipment, our understanding is that they are not investing in future generations of their internal tester. And so we think that there's going to be a fight in 2028 for who's going to win that account. So we're positioning to compete for that account for 2028 on.

Now the other story in CPU is the balance between x86 and ARM. And when it comes to ARM, Teradyne is either qualified or in the qualification process for all major ARM CPUs. Okay. So our share gain opportunity in ARM-based data center CPUs is significant. So I guess I'm kind of hoping that ARM does really well because our exposure there is much higher.

Understood. So just to -- yes, please.

Maybe driving over time [indiscernible]...

Yes, yes. So we've talked about is that AI comes in waves. And the primary hardware associated with up through 2026 has been around building out general purpose data centers, data centers that can be used for training or for inference. And that was very important because the sort of the relative allocation of compute was pretty dynamic between research tasks and revenue tasks.

Now that the large LLM players are scaling more aggressively in revenue and -- like look at Anthropic the revenue growth that they're showing there is a need to do inference at scale. And so that's behind the CPU renaissance, and it's also behind silicon specifically to support inference, and that would be like [ Rock and Cerberus ]. That's wave #2. And that's really kicking into high gear in 2026, and it's going to be a major factor out through 2030.

The next wave after that is the edge-deployed AI. And there's three real categories of that. There's autonomous driving, there's mobile electronics, so phones, glasses, things that bring edge AI real-time AI to individual users; and then physical AI or robotics.

And Teradyne is positioning itself as a destination platform for physical AI applications in robotics. We have a lead customer in this space in e-commerce. That is deploying a physical AI application for distribution center, both stowing stuff in a distribution center and also picking orders to make up kits that are being delivered.

And that has really helped us understand what the right platform characteristics are to be able to succeed with other players in physical AI. So our kind of #1 priority right now is to win the labs. So the people that are kind of determining the physical AI future are people like skilled and generalists, intrinsic, Microsoft and NVIDIA.

They are adapting the research that they've done in the cloud environment towards physical AI tasks. And what we're trying to do is make sure that we can provide a -- like not to borrow the pharmaceutical term, a safe and effective platform because from a safe perspective, even though the AI job runs on our platform, our platform actually has a safety layer that prevents the robot from operating beyond the safe parameters for the job. That's true for our AMRs and for our [ arms ].

And then effective, you need to be able to control the robot from another piece of software through an API. And so you need scalable compute, you need modular accessories and you need the operating system to both accept commands at a high frequency and provide feedback at a high frequency. So think of that as like the sensory and motor nervous system of the robot has to be able to be interface to these brains.

So I guess you could say, our strategy is to bring the muscle and the generalists and intrinsics and NVIDIAs are going to be bringing the brains.

On the CPU cycle, acetate, I think you have mentioned that there, we think that TAM is about $500 million that we can test. Can you break it down further like this definition [indiscernible] $500 million? And where can you gain the share that you're confident in the past?

So right now, in 2026, there's probably going to be about $100 million of test equipment purchased for co-packaged optics. In addition to that, there's probably an equal amount equal amount of revenue associated with the device handling and optical alignment. So another $100 million or so of business that will go to players that are not Teradyne or [ Advent ] test.

And all of that investment is going to produce a vanishingly small amount of co-packaged optics in 2026. So like a crazy high buy rate. So over the next few years, there's a bunch of exponentials that are going to happen. The number between now and 2028, the number of scale-out CPO connections is probably going to multiply by more than a factor of 100 from '26 to '28 because like it's growing from nothing. So a very high growth rate.

The -- right now, each CPO connection represents minutes of test time. There's a very, very extensive test lists that are very characterization oriented. And so between now and 2028, the test intensity is going to be going -- like they're going to test less. And the test equipment is going to become much faster because it's going to be focused on the production parameters that are most important.

So you've got this exponential growth on one side, and you've got a rapid increase in efficiency on the other side. So when you -- like it's always high error uncertainty when you're trying to match up things that are growing in an exponential rate.

So what we've actually said is that by 2028, we think that the TAM associated with just the test equipment, so what's $100 million now; will be somewhere between $300 million and $700 million.

You pick the midpoint and like if you want to -- like they say that if you're in marketing, like if a guy in marketing says that something is going to drop at that spot on the floor, something heavy, stand right there because you're going to be safe. because like we're never going to get it exactly right, but we can give you the band. And so it's going to be that 300 to 700.

And the other thing that's interesting is that is really associated with just initial production of scale-out networking. What happens post '28 is when scale-up networking happens. And by the time you get to 2030, there's going to be like 3x as many CPO ports for scale up as there is for scale out.

And so $500 million at that midpoint, that $300 million to $700 million, that is not where this TAM ends up. This TAM ends up bigger than $1 billion. And the scale-out stuff is high -- like we have a high exposure to that part of the market because of our share in networking. And my competitor has high exposure to scale up.

And so both of us are going to be creating into this $1 billion market, and neither one of us is going to give up. And so what we'll be doing is we'll be competing on the effectiveness of our testers, the throughput, the accuracy and I've seen this play before. And what ends up happening is a highly competitive duopoly where we're splitting the share.

Is it most [indiscernible] wafer side that we sell caters to or actually other income?

So right now, there are 4 different test insertions -- there's a pick wafer on its own. There is a combined PIC in EIC wafer. There's a a singulated optical engine dilate test and then there's a test when the whole CPO unit is tied together.

Right now, most of the TAM is in the first and second insertion. And they are unique test lists, and you can't do all of the tests that you need to do before the EIC and the PIC are put together. So it's like there's going to be movement of tests in that chain, but you can't arbitrarily do some of the later tests earlier. And our biggest share is in that second insertion.

Just on that point, there was also a hoteling equipment with device handling alignment, which you don't do here to assume that in the prototyping R&D, working out which product work is, is somebody that ECM equipment will get used more and maybe the basic equipment that more when we get into volume production of these units. Is that fair to say?

So I think that's fair. I mean one point that I'd like to make is that for the rest of the whole ATE industry, there is an open ecosystem, where test equipment vendors work with multiple device handling vendors, so like Teradyne works with [ Acrotech and Tel ] and Advent test and other players to help create test cells.

The same thing we believe is going to happen in the co-packaged optics space that Teradyne is going to work with multiple handling vendors, and those handling vendors will work with both Teradyne and [ Advantest ].

Do I think that the TAM for handling equipment will -- I think it's probably going to grow it close to the same rate as test equipment that like at the end of the day, the number of CPO dilets that are going to be built is going to be growing dramatically and the specialized equipment that are required to deal with those at like -- not only insertion 1 and insertion 2, but handling those -- the optical engine dilets and getting the fiber bundles in for that third insertion and then the handler for a CPO device, that's like a world-class switch device it's going to be burning a kilowatt, tens of thousands of electrical connections and then you also have to meet to like optical connections on there. That's going to be an expensive piece of kit. So I think that this is going to be a part of the handling market that is likely to grow really significantly.

Follow-up on your comment and your competitor is taller I mean just given no one's quite doing...

No, no, no, no. What I was saying is that, like if you think about scale out, scale out is adding CPO to network switches. If you think about scale up, it's adding CPO to AI accelerators. And if you look at share in AI accelerators versus share in network switches, Teradyne is really high in network switches, my competitor is really high. But you're right, like the future is yet to be written in terms of CPO share for both of these markets.

All right. Wonderful. On the topic of that large customer, right, you received initial kind of production orders. What's the right to elect to think about your market share journey at that customer? What are the milestones? And then as part of the $12 billion to $14 billion, right opportunity that you laid out, what share does that imply?

Yes. So the way we've been talking about the merchant compute opportunity in this specific account is that there are 3 phases. The first phase is qualification. And that started midyear last year and it culminated with qualification of the UltraFLEX to be able to test a particular device.

And that was -- the end of that process was two things happened. One, there was an order for a couple of dozen tools that we are going to be doing high volume production of this specific device. And the other was the -- them asking us to engage in additional follow-on projects for devices that are earlier in their life cycle.

So we are now in a phase that I'm calling the fast follower phase. So every new part is going to first be brought up on my competitor's platform. And for particular devices, especially devices that they see high volume, they are going to be asking us to convert that test solution to be working on the Teradyne platform. So over time, we are going to convert more and more of these devices as they are coming to market, and that will allow us to incrementally gain share.

While we are following, we believe that we have kind of a hard ceiling at about 30% share because like we're the second kid to the dinner table, right, that like the company that brings up the part in the first place is always going to get the initial part of the [indiscernible]. So that fast follower kind of has a natural high limit.

After about 3 or 4 years the customer will likely have enough experience with our platform, and we will get into what I call a mature dual vendor strategy. And in a mature dual vendor strategy, they will generally be selecting a an NPI platform and then the other company would be the follower. And that would be like either Teradyne or the incumbent would be selected for new devices on the basis of platform capability, throughput, reliability. Like they're making decisions on the actual differentiation versus incumbency. And that's like 3 or 4 years in the future when we get to there.

So inside of this $12 billion to $14 billion TAM, part of it is like how much merchant compute is in there. Part of that depends on how fast we get to $12 billion to $14 billion, that if we get to $12 billion to $14 billion next year, then we will still be at a very early stage in this process and we would have lower share. If it took us 2 years, we would be higher. If it took us 3 years to get to $12 billion to $14 billion, then we would be getting close to that 30% but it's going to be a gradual increase over the next 3 to 4 years to get to 30%.

Right. And just the last question. You do expect to maintain or expand your share in networking because that's been an area of strength?

So dual vendor is a customer strategy against their entire portfolio of parts it definitely presents a downside risk in networking and an upside risk in merchant compute. But given the size of the opportunity, we're kind of delighted to take the risk against networking in order to increase our share in that larger part of the account.

We also believe that we have a platform that is more performing, that we actually have differentiation in production around reliability and throughput. And so we believe that we're going to be able to defend a high share in networking.

Would we be able to defend 100% share? Probably not. But would we be able to defend a high share, we think so. SP-7

Got it. With that, thank you so much, Greg. Really appreciate your time.

Well, thank you all.

Ladies and gentlemen, good morning, and welcome to the Teradyne First Quarter 2026 Earnings Conference Call. [Operator Instructions] As a reminder, today's call is being recorded.

I'd now like to turn the call over to Amy McAndrews, VP of Corporate Relations for Teradyne. Please go ahead.

Thank you, operator. Good morning, everyone, and welcome to our discussion of Teradyne's most recent financial results. I'm joined this morning by our CEO, Greg Smith; and our CFO, Michelle Turner. Following our opening remarks, we'll provide details of our performance for the first quarter of 2026, and our outlook for the second quarter. The press release containing our first quarter results was issued last evening. We are providing slides as well as a copy of these prepared remarks on the Teradyne investor website that may be helpful in following the discussion. Replays of this call will be available via the same page after the call ends.

The matters that we discuss today will include forward-looking statements that involve risks that could cause Teradyne's results to differ materially from management's current expectations. We caution listeners not to place undue reliance on any forward-looking statements included in this presentation. We encourage you to review the safe harbor statement contained in the slides accompanying this presentation as well as the risk factors described in our annual report on Form 10-K for the fiscal year ended December 31, 2025, on file with the SEC.

Additionally, these forward-looking statements are made only as of today. During today's call, we will refer to non-GAAP financial measures. We have posted additional information concerning these non-GAAP financial measures, including reconciliation to the most directly comparable GAAP financial measures, where available on the Investor page of our website.

Looking ahead between now and our next earnings call, Teradyne expects to participate in technology-focused investor conferences hosted by Bernstein, TD Cowen, Stifel, and Bank of America. Our quiet period will begin at the close of business on June 12, 2026. Following Greg and Michelle's comments this morning, we'll open up the call for questions. This call is scheduled for 1 hour. Greg?

Good morning. With revenue of approximately $1.3 billion and non-GAAP EPS of $2.56, Teradyne delivered record results in the first quarter of 2026. Our previous high watermark was in the consumer-driven mobile peak of Q2 of 2021. In Q1 of 2026, our revenue was $200 million or 18% higher than that previous record. This new record comes from durable AI demand drivers and the continuing acceleration of our Wafer to AI data center strategy. This strategy is delivering demand across Teradyne's portfolio.

In Q1, AI-related demand accounted for nearly 70% of our revenue, up from about 60% in Q4 of 2025. Our strategy continues to be anchored across 3 broad trends; verticalization, electrification, and AI. Verticalization is the concentration of our business into extremely large vertically integrated technology companies. The verticalization trend was clear by 2024 and continues to accelerate. This includes companies like hyperscalers, but also huge AI ecosystem enablers like foundries, merchant compute, memory and networking companies.

Many of these companies are customers of all 3 of our businesses; Semiconductor Test, Product Test and Robotics. And this product portfolio enables us to serve their needs from wafer to data center. While these massive customers are driving strong growth, it also means that the business is increasingly concentrated to these customers and to a smaller number of very large ASIC and commercial device programs. This concentration also increases the risk that bottlenecks in other areas could shift demand for our products, which can lead to short-term demand peaks and valleys superimposed over a long-term strong growth trend. In other words, it's lumpy growth.

The electrification trend continues. In the Auto/Industrial segment, 46% of our revenue came from data center devices in the first quarter, which historically has been dominated by automotive and industrial devices. It goes without saying AI is the dominant force shaping our business. We think about the opportunity presented by AI as 3 superimposed waves, each building on the one before it. We are in the heart of the first wave, which focused on the build-out of general purpose AI data center capacity. This was behind the massive increase in data center spend in 2025.

In 2026, we are entering the second wave. While there is still huge investment in general purpose AI data centers, these data centers are being augmented with compute silicon optimized for inference at scale. This wave will grow to a high run rate over the next few years. Still yet to come is the edge AI, physical AI wave as the technologies for silicon, packaging, memory and AI models improve, compelling use cases for AI at the edge will be emerging. Obvious examples of this are self-driving cars, robotics, PCs, wearables, and smartphones.

These waves are broad-based, and we expect them to stack on top of each other, driving significant ATE TAM growth over the full midterm. Because of Teradyne's wafer to data center strategy and our historic strength in mobile, automotive and industrial, we are well positioned to ride each of these waves as they arrive. Back in our January call, we shared that we expected robust double-digit year-over-year growth. We still expect that the compute TAM and revenue will grow significantly from an already strong 2025 base. We're seeing healthy engagement with both networking and VIP compute customers and our pipeline of new design wins remains robust.

Aligned with this momentum, I am pleased to share that we have received our first multisystem production test orders for merchant GPU in Q1. We expect these systems to ship, be installed and be in production in Q2. Customer engagement remains strong, and we are well positioned to capture further share as we bring up more devices on our platform. In automotive and industrial, we're seeing moderate but steady recovery in both TAM and revenue. There are signs of strength in automotive, primarily ADAS, and we're seeing increased demand for power going into AI data centers.

As of now, mobile appears a bit weaker with memory pricing and availability affecting end market demand, especially outside the iOS ecosystem. Memory test demand appears to be even stronger than our view in January, with AI compute demand for both HBM and DRAM continuing to act as an accelerator. We're also beginning to see increasing flash test demand driven by SSD. The overall memory market is on track for solid TAM growth for the year, and we expect to gain low single-digit share.

In 2025, our IST Group expanded its HDD customer base and entered the SLT compute market. Now in 2026, IST is on track to deliver against this expanded opportunity. We're seeing strength in HDD driven by greater than 20% annual exabyte growth fueled by AI. This translates into longer test times per drive and a larger HDD TAM and revenue for Teradyne.

In Robotics, we delivered our fourth consecutive quarter of sequential growth. This is particularly notable because Q4 is typically our strongest quarter and Q1 is typically down. We're seeing strong customer engagement across e-commerce, electronics manufacturing and semiconductor end markets. Robotics is a key part of our wafer to AI data center strategy with robotic-assisted assembly, test and data center operations. Our robots are being used in environmental sensing in data centers, and we recently demonstrated a complex physical AI work cell in partnership with Generalist as part of the recent NVIDIA GTC.

In prior calls, we have often talked about the investments that we are making to capture growth opportunities coming from our wafer to data center strategy. In Q1, these investments have resulted in 2 significant new product introductions. The first is Photon100, which is our platform for silicon photonics and co-packaged optics testing. The Photon100 is based on our proven UltraFLEXplus tester and is bringing SiPho testing from lab to fab. I'll remind you that silicon photonics and co-packaged optics are in the very early stages of a ramp that will likely be substantial. There is uncertainty about the timing and the slope of this ramp, but as optical interconnections are increasingly used for scale out and then scale up networking, it is going to be a big chunk of the total networking TAM.

As this market grows, we also expect to bring significantly more efficient test solutions online, so it would be a mistake to linearly extrapolate from today's test strategies and economics. That being said, we expect that this is a meaningful TAM expansion opportunity, which could reach $300 million to $700 million per year over the midterm. The second product introduction is Omnyx, which is a new production board test platform designed to address the unique set of test challenges for server boards and tray assemblies. This platform uses power, thermal, optical and TDR test capabilities from across all of Teradyne to enable earlier detection of defects that are plaguing the build-out of AI data centers.

In addition, we continue to pursue inorganic opportunities to grow our business. Our MultiLane Test Products joint venture closed on April 8, and we believe this partnership will accelerate the development of high-speed I/O and data center interconnect test solutions, a critical test need as AI data centers transition from cable-based connections to backplane and midplane architectures. Additionally, we closed the acquisition of TestInsight 2 weeks ago. TestInsight is the leading provider of test development tools that are used with our testers and competing platforms. This acquisition strengthens Teradyne's design-to-test software capabilities, enabling us to build a virtual test environment, which will reduce time to market for complex AI and networking devices.

In summary, Q1 2026 was a record quarter for Teradyne. We're executing our strategy, capitalizing on secular growth drivers and delivering value for our customers and shareholders. Our team, especially our operations team and manufacturing partners, went above and beyond to hit this ramp, and I'm grateful for their hard work and skill. We came into the second quarter with a lot of momentum and confidence that 2026 will be a strong growth year, and we are well on our way to achieving our target earnings model. With that, I'll turn the call over to Michelle.

Thank you, Greg, and good morning, everyone. Today, I will cover our first quarter financial results and our second quarter 2026 outlook. Starting first with Q1. First quarter sales were $1.282 billion with non-GAAP EPS of $2.56, both above the high end of our guidance range. Total company sales were up 87% from first quarter last year and up 18% sequentially from last quarter. Non-GAAP earnings per share was up 241% from first quarter last year and up 42% sequentially from last quarter. This represents a record financial performance for the company, driven by all things AI across all 3 of our business groups.

In the quarter, we continue to have 2 specifying customers and 1 purchasing customer greater than 10% of our revenue. Building on that, let's look a little deeper at revenue, starting with SemiTest. Revenue was $1.1 billion, breaking the $1 billion threshold for the first time, up 26% sequentially from last quarter and over 100% year-over-year versus Q1 2025. The revenue breakdown within SemiTest was SoC at $882 million (sic) [ $883 million ], memory at $203 million and IST at $27 million. The key drivers were continued AI strength in compute segments and memory. At roughly 75%, compute is the largest portion of our SoC product revenue. This continues the evolution of our test portfolio from being mobile-centric shifting to AI dominant.

Within auto and industrial, revenue nearly doubled sequentially from a low base last quarter, driven by power management demand increases for AI data center build-outs. Mobile revenue was roughly flat with fourth quarter 2025 and remains a muted impact to our overall results with the increasing importance of compute in our SoC portfolio. Aligned with our strong top line performance, operationally, we have more than doubled our UltraFLEXplus shipments over the last 9 months while sustaining our 12- to 16-week lead times. Our multisource strategy, primarily leveraging contract manufacturers provides ultimate flexibility for our customers while ensuring capacity continuity in today's dynamic environment.

Moving on to memory. Our memory business delivered another strong quarter of $203 million in revenue, relatively flat to our record last quarter, driven by robust HBM and DRAM test solution demand. We also successfully ramped the newest generation of our memory tester, Magnum 7. IST revenue of $27 million was relatively flat year-over-year, though we are seeing early indicators for potential growth driven primarily by HDD in the second half and continuing into 2027. Product Test Group revenue was $80 million, up 8% year-over-year. Growth was led by sustained defense and aerospace demand and production board test.

Robotics revenue was $91 million, up 32% year-over-year, representing our fourth sequential quarter of growth. Our one sales team approach is delivering results with revenue strength across end market verticals in e-commerce, electronics manufacturing and semiconductors, including in AI data centers. Shipments associated with our large e-commerce customer increased sequentially and AI revenue increased to 15% of the quarter's sales.

Now moving down the P&L. A confluence of positive factors delivered record earnings results, including peak AI-driven volume, favorable product mix and nonrecurring one-time benefits. Gross margin for the quarter was 60.9%, up 370 basis points sequentially, driven by strong SemiTest volume and product mix and nonrecurring operational impacts. OpEx declined sequentially from last quarter and was favorable to guidance due primarily to the timing of nonrecurring engineering. Non-GAAP operating income was $480 million with an operating margin of 37.5%, both all-time financial records.

Now moving on to capital allocation. Our capital allocation strategy remains consistent, and that is to maintain cash reserves to enable us to run the business and have dry powder for M&A. We ended the quarter with cash and investments of roughly $400 million. Working capital, predominantly in accounts receivable, increased in support of the revenue growth delivered in the quarter. Capital expenditures were flat year-over-year with the expectation that Q2 will increase, driven by continued investments in innovation and operations scaling. We paid $20 million in dividends in the quarter, and our share buybacks were de minimis.

As Greg mentioned, we closed on 2 important inorganic asset opportunities this month. On April 8, we closed on our previously announced MultiLane Test Products joint venture. The results of this business will be consolidated into the product test group, and our EPS will reflect our share of the results of this business. On April 16, we closed on the acquisition of TestInsight business, furthering our wafer to AI data center product penetration. Combined, these 2 deals used roughly $165 million of cash in the second quarter, which we funded via our credit revolver.

Looking ahead to our second quarter guidance. For the quarter, we expect revenue in the range of $1.15 billion to $1.25 billion and non-GAAP EPS of $1.86 to $2.15. Gross margins are expected to be in the range of 58% to 59% normalized for peak volumes and onetime benefits. Operating expenses are expected to run at approximately 27% to 28% of second quarter sales. The non-GAAP operating profit rate is expected to be between 30% and 32%. Based on current customer order visibility, we continue to expect first half weighted revenue with approximately 55% to 60% of annual revenue expected in the first half. This expanded range from 3 months ago recognizes the continued strong demand signals we are hearing from our customers while also balancing potential order lumpiness that could impact revenue timing across quarters or years.

For the year, we have line of sight to about $50 million in revenue for merchant GPU, but our visibility into the second half is quite limited with increasing contributions over the midterm period. So in closing, our teams delivered exceptional financial results, reflecting strong execution and robust demand across our portfolio aligned with our wafer to data center strategy. We remain confident in the full year trajectory and our target model of $6 billion in revenue and $9.50 to $11 in non-GAAP EPS.

I want to thank all of our Teradyne team members for their performance and operational discipline in delivering for our customers and shareholders.

With that, we'll open the call for questions. Operator?

[Operator Instructions] And we'll take our first question from Timothy Arcuri with UBS.

Greg, I guess my question is just on the back half of the year. So it's a bit of a disconnect. It sounds like the demand signals if anything gotten better over the past 3 months, but you're not raising guidance for the back half of the year. Is this -- this kind of came up on your competitors' call as well. So is this to some degree, like factoring in some constraints that maybe are downstream of your business? And I know you did talk about this VIP stuff can be very lumpy. So maybe that's part of it. So if you can talk about that.

Tim, it's Michelle. I'll start and then Greg can add some specifics from a customer perspective. So let me start with where we're at today. Q1 was an exceptional quarter, a record quarter. You heard that throughout our script for the company. I mean when you look at our Q2 guidance, it's equally strong. So revenue of $1.15 billion to $1.25 billion. This represents about an 84% year-over-year growth at the midpoint after coming off of a really strong Q1 of 87% growth.

So as a result of the strength in the first half, we have expanded our first half revenue range to 55% to 60%. So this is a change from January where we've given a point estimate of 60%. So when I think about the ranges, just to kind of bound this, the low end of the range really reflects the potential for timing impacts. So this is either lumpiness in terms of large customer ordering patterns or it could also be hiccups in the AI data center build-out, the ecosystem, if you will, in terms of when our testers actually get accepted. So these dynamics, as you know, can impact revenue within the quarter or across year boundaries. And so that is part of the dynamic that's playing out in the second half.

When I think about the high end of the range, this really reflects continued strength that we're seeing from a demand perspective across compute, networking and memory. And then the other element I would add to this is in terms of visibility. So we talked in our January call around improved visibility from a customer ordering perspective. That is consistent with where we're at today. So historically, this business has had about 13 weeks of visibility. Coming into this year, we improved that. We can now see into another quarter out, although not as strong as the current quarter. And so we still do have some undefined parts as we think about Q4.

Yes. So -- and Tim, let me give you a little bit more color in terms of sort of how the first half, second half polarization breaks down by sort of group or technology. The part of our business that we believe is most first half weighted is VIP compute. And that's -- like we have pretty good visibility into the timing of programs associated with that and the specific customers that we have. We think that, that's like -- it's really, really strong in the first half of the year and the next wave of that for the next generation of that technology is early '27. It might start bleeding in or pulling into the end of 2026, but we really don't know about that.

When you go from then to like networking, networking has started off very, very strong. And we think it's going to stay at a reasonably strong level through the year, but our visibility isn't as strong into the second half. And we've historically seen that sort of filling in more as time goes on. Like as you look beyond 2 quarters, you start to see that sort of -- that tending to go up, not down. So there's potential upside in the networking space.

When you look at memory, that, I think, is actually going to end up being more back half weighted than front half weighted. So that's a counter thing. And the stronger that memory gets, the more we're going to be able to trend towards that 55% end of the range that we gave. And then if you look beyond the semiconductor test part of this -- well, actually, in semiconductor test, just to sort of finish the story around auto and industrial, right now, data center is hot, hot, hot in that segment, and we expect that to continue. What we're seeing and hearing from those customers is that the rest of their portfolio that there are -- like there's reduced inventories and potential demand increases, but we haven't seen that translate into increased demand for capital equipment in the auto and industrial part of that beyond data center.

So moving beyond that, getting into IST, we definitely think that's stronger second half than first half. But I mean, we're talking about coming off of a base at $27 million in Q1. So there's a lot of upside to go before it really moves the needle at the enterprise level. Product test similarly will be back half weighted. But again, it's like a smaller percentage of the total. And typically, our second half is much stronger in robotics than the first half, but we've started this year at a really good run rate. And so we -- like the signs are encouraging, but we've learned to be very careful about predicting what's going to happen beyond lead time in the robotics space. So does that help a little bit?

It does, Greg. Yes. Yes. Yes. I guess, I just then wanted to ask you about the TAM for the year. So you didn't give us a TAM. I know Advantest is saying like low 9s for SoC and sort of low to mid-2s for memory, so kind of a total of like $11.5 billion. You've usually been pretty close to their number, a bit lower in memory and a bit higher in SoC. So is that total $11.5 billion for the year? Is that like a reasonable TAM number for the year?

Bearing in mind how far off both Teradyne and Advantest were about the 2025 TAM when it was April, so make that the big asterisk on this answer that like because a lot changed last year and the TAM strengthened significantly from the April view through the full year view. Now for 2026, there's 3 -- like just at a logical level, there are 3 possible things. One is that people have overcorrected in terms of estimating what the TAM is. People have gotten the TAM exactly right or it could follow the same pattern that it followed in 2025. I would say that like we are not talking about a TAM publicly because we feel really uncertain about which of those timelines we're living in. I don't think the numbers that Advantest gave are absurd, but I don't feel confident enough in our forecast to share them.

We'll take our next question from C.J. Muse of Cantor Fitzgerald.

I guess first question, and again, congrats on your first merchant GPU win. Curious how to think about the follow-through there? What kind of -- are the steps to try to ascertain greater percentage penetration there as well as how you're thinking about custom ASICs from here beyond your one very large customer?

Yes. So first on GPU, what we've said previously, I think, is the way it is going to -- is tending to play out, that the first project is the hardest project because it involves qualifying the test platform and converting all of the underlying libraries that support the testing. So that part of this project is behind us, and we are into the phase where that initial qualifying part can go into production. The next phase is what you could call the fast follower phase. So we're going to begin working on projects that are earlier in their life cycle and that we will be able to complete more quickly than the first qualification project.

So those are projects that probably have a time line where we would be releasing into production late in this year. And so whether that capacity ramp starts to hit at the end of '26 or into '27 is an open question. And the thing that I want to caution is like the long-term view, over the midterm, we expect that a dual source customer is going to be managing share in that 30% to 70% range. It's going to take us a few years to get there because there are so many different part types, so many different SKUs that as an incumbent platform, there's a lot of flexibility about what -- like if you need capacity for a particular device, you know that you have the solution for it on the incumbent platform.

So during the fast follower, we're really competing on the basis of differentiation of the platform and an ability to serve spot demand more quickly than our competition. So we're trying to be very responsive. We're trying to get as many SKUs converted over to our platform as possible. But I would expect it's going to take -- like it's going to take us a few years to get from low single digits in 2026 to sort of entering that 30% to 70% range over the midterm.

Very helpful. And then maybe a question on gross margins. You're taking a downtick here. And historically, the business has not been fixed cost. It's been much more product cycle driven. So curious what is precisely driving that downtick in June? And how should we be thinking about modeling the second half of the year?

Hang on, like my colleagues have reminded me that I neglected the second half of your question about custom ASICs. So we'll take your gross margin question right after. I just want to hit the custom ASICs.

So right now, there are 2 hyperscaler programs -- 2 compute hyperscaler programs that are at scale. If you include like edge automotive, there are 3 hyperscalers that are at commercial scale and driving tons of volume for us or our competitor. We are actively competing for parts that have not yet ramped and also for dual source status against the hyperscalers that have already ramped. And the timing for that would be more 2027 than 2026. But stay tuned for news of that as we go on. And I'll pass it over to Michelle for the gross margin commentary.

Well, I think technically now C.J. has 3 questions now as a result of this, but we'll go with it.

But that's my error.

Yes. So from a gross margin perspective, we did have a really strong Q1. So 60.9%, again, another record for the company. And there were several favorable factors coming into play simultaneously that drove this. One was related to the AI demand. So really strong SemiTest volume, which was 87% of our overall portfolio within Q1. Along with that, we also had favorable product mix and some benefit from some nonrecurring operational benefits. So as you think about the shift from -- or the step down from Q1 into Q2, at the midpoint, that's about 240 basis points. About half of that is driven by the onetime nonrecurring kind of operational benefits that we had.

And then you couple that with what we're seeing from a mix perspective within Q2, this is somewhat of a normalization. The one thing I will highlight, however, is when you look at first half, overall margins will be around 59.7%. This is at the low end of our target model range. And so I think it's important to kind of keep in context that you should expect to see our margins move around a bit. They are lumpy like our revenue. We typically will see up to 400 basis points within a year. However, when you look year-on-year, it's a much tighter range. It's within like 200 basis points. So some of this is noise just within the first half is how I would think about it from a modeling perspective.

We'll take our next question from Mehdi Hosseini with SFG.

I also have 2. The first one has to do with how you have been managing quarterly guide, and your performance has actually been exceeding on a consistent basis. And with that as a background, my question to you is, are you seeing a consistent trend where late in the quarter, you get the rush order, you get the programs in line and the last month of the quarter becomes the source of upside? Or is there something fundamentally different in the way you communicate with the Street? And I have a follow-up.

I'll start and Michelle, if you want to chime in with some additional color. When we sit down to do our guide, we try to give the very best view in terms of balancing the risks and opportunities that we see in the current quarter. And we are typically carrying some upside capacity that if we get quick turn orders that we -- that -- like if we can help improve customer satisfaction by serving orders within lead time and we have the capacity to do it, we are going to do that. So in a strengthening demand environment, we will tend to overperform.

At the same time, we, like everyone else, are working really hard to solve the supply chain issues that come from ramping capacity. And so there are supply chain risks that our operations team does a great job solving. But when we go into the quarter, we don't know that we have a solution for all of those. The other thing that I'll say at a like minor level is we are still being quite cautious about robotics that we are seeing much improved -- like improved predictability and improved growth from that unit, but we don't want to get ahead of ourselves in terms of assuming that we are in like a bold new future there. We want to make sure that we really understand exactly what our funnel of opportunities looks like and our conversion rate. And like we're really happy about having 4 good on-track quarters in a row, but we are still quite careful in terms of predicting like higher rates of growth against that business until we get a little bit further down the road.

Yes. And the only other thing I would add to that, to Greg's point about the supply chain variability, that's not just within Teradyne, but also with our customers. And so to the extent, for example, all the parts of a test cell are not coming together, they're not looking to take our testers and install them. There could be slips that happen within the quarter. And so that could be either upside or downside in terms of how it plays out within a particular week. And so you'll see this sitting in our receivables at the end of Q1, where we shipped a lot of units that originally we had been told to kind of push out. So I think it's important to note that we're seeing variability not only on the order side, but also in terms of the supply chain from a customer perspective.

Got it. And then since last earnings conference call, agentic AI has become the new buzzword, and there are a number of existing and new semiconductor companies that are trying to capitalize on tokenization and offer a new kind of a CPU. And my question to you is, given the fact that historically, x86 has been more of a -- driving more of an in-house test solution and now you have a diversification of a CPU because of agentic AI. Does that also add a new layer of opportunity for you? Was that already embedded in your longer-term forecast? Or is this something new and could potentially provide some upside?

So I think of it more as providing potential upside than something that's in our plan that we are testing primarily ARM-based CPUs for data center applications. I think we have active design-in opportunities that I expect to be able to convert in this space as well. I would add that in addition to sort of new demand for CPU in agentic, there's also a big trend towards this optimizing data centers for inference at scale. I talked about that a little bit in the prepared remarks that if you look at what NVIDIA is doing to try to decreased time to first token on inference and other players are doing, I think we have reasonable exposure to those types of devices as well. So as things shift towards more inference and more agentic, then I think we have potential long-term upside.

We'll take our next question from Shane Brett with Morgan Stanley.

My first question is on networking. Regarding CPU and silicon photonics, can you talk about where you stand in terms of share now and how you anticipate your share tracking? I'm asking as your competitor announced that they received their first high-volume ATE order for silicon photonics.

Thanks for the question. So our best guess so far in 2026, that share is quite balanced between us and our competitor. And I think the -- like the only difference is whether we're talking about single large orders or multiple smaller orders. But I think share is kind of balanced, and that's where we think things are right now. As -- but it's really early days that right now, there is -- there are very few end customers that are trying to ramp CPO into production. And the share split right now is mostly around which test insertion is being done by what companies. And so our strength is primarily in insertion 2, where you're actually connecting electrically to a compound wafer on the top for electrical and looking at light down in the bottom. We are in the process of releasing solutions for that in production. And we are working with partners to do that.

So it's really kind of a 4-way partnership involving foundry, end customer, flight contact and Teradyne. We're working that with the team in Taiwan, Israel, Germany, North America to bring this technology to production. We think that it's -- this is a very important market. And as -- like what -- the way we see this happening is that in '27, this is primarily going to be associated with scale-out kind of networking. The scale-up networking is likely to be even higher volume, but over a longer period of time, like '28, '29. And we are going to be rapidly changing the efficiency of test at all 4 insertions and the balance of where things are going to be done across those 4 insertions is really going to be driven by the end economics.

So if they are finding a lot of faults at a particular level, they will keep doing that test. If they have very high yields, then they will look to see if they can eliminate that. But I -- personally, I think efficiency is going to go up by like a factor of 10 over the next couple of years. And despite the fact that, that efficiency is going to get that much higher, I think that this is still going to be $300 million to $700 million worth of equipment once you get a few years into this midterm.

That's really insightful. And for my follow-up, one of your auto industrial customers talked about a bit of a tester shortage on their earnings call. And when I was listening to that, my interpretation was, oh, there may be a fight among customers to get in the queue at Teradyne. Just where do we stand right now in terms of capacity and utilization? And how much capacity are we looking to expand over the next year or so?

So if that customer is having trouble getting testers, they're obviously not buying them from Teradyne, that we are able to serve the demand that we have. Our capacity has ramped rapidly where we have multiple contract manufacturing partners that are enabling our production. So we were surprised by that. We -- for this particular end customer, we have a very good relationship. And for the parts of their business that we serve, we're able to deliver the testers that they need in the lead time that they need it.

So I think this is another reason that more and more customers are really looking at supply chain resilience all the way back to their test equipment capital supplier. And that kind of strategy isn't something that is just solved by increasing capacity that what customers really need is to be able to get the capacity that they need when they need it for the parts that they are ramping. And you don't know how that demand is going to overlap as a supplier. So I think it's -- I think the trend in the future is that for high-volume devices, we're going to increasingly see this trend towards multiple sourcing of test equipment. And overall, I see that as a real positive for Teradyne.

We'll take our next question from Krish Sankar with TD Cowen.

I have 2 of them. First one on the silicon photonics. Michelle or Greg, you spoke about the $300 million to $700 million opportunity. I understand it's a midterm. How big is the market this year? Is it like tens of millions of dollars this year? And along the same path, is the ficonTEC being acquired by Chinese entity an issue? Or is it a non-issue for you? And then I had a quick follow-up.

Yes. So this year, we're probably looking at silicon photonics right around 100-ish, maybe a little bit less, maybe a little bit more. It's substantial, but it's pretty early days. With regards to ficonTEC, ficonTEC has been an independently operated unit of a Chinese corporation since 2021, I think. I mean this is not a new thing. It is -- and our relationship with ficonTEC is with both the unit, which is in Germany and also the Chinese company that it's a part of RoboTechnik, we have a great relationship with the CEO there. And they are one of the absolute world-class providers of active alignment for silicon photonics assembly and for test. That -- there are a lot of hard things that you need to do when you're building silicon photonics. One of the most difficult is achieving alignment in the assembly process for electro-optic modules or for CPOs.

And ficonTEC is like one of the world leaders in that technology, and they are working to build their business in semiconductor capital equipment, like the test part of it. And that is something that is a high priority for us, high priority for them. And I think that the press reports that came out are -- like they've been refuted by ficonTEC, and we don't see any evidence on the ground that it's at all true.

That's very helpful. And a quick follow-up for Michelle. I know you gave some color on how to think about gross margins. But just visually thinking as auto analog industrials cyclically starts rebounding, is it fair to assume that's a huge tailwind for your gross margin given those legacy Eagle testers are pretty high margins?

So I think the short answer is no. When you think about kind of our full year expectations and what's really going to influence the end results even over the midterm, right? So we've given a target model of 59% to 61%. We're sitting at 60.9% in Q1. We expect the first half to be about 59.7%. I still think that, that's within the range. And auto -- when you think about auto and industrial, it's not going to be the biggest swinger in terms of our overall margin portfolio. I'm just going to go back to just the tightness in our overall margins within 200 basis points year-on-year. So when you use the words like huge or significant, I probably react to that a little bit. I do think it can be somewhat of a tailwind. But just given the size of it to the overall portfolio, I would say no in terms of the significance.

Yes. I mean over the years, we've seen most of our product lines converging towards similar margins. And so it's -- I would not expect to see -- even if we saw us like a significant increase in the percentage of total revenue that's coming from auto and industrial, I don't think we'd see that as a big mover. Now the other thing to remember is that one of the things that goes into auto and industrial is ADAS. And so if you squint at an ADAS tester, it looks a hell of a lot like a VIP compute tester. So just because things are aggregating into particular end markets, it doesn't necessarily reflect the platform that's being sold to serve it.

We'll take our next question from Vivek Arya with Bank of America Securities.

Greg, on the GPU engagement, when we look at the large customer, the demand is clearly increasing. The number of SKUs across training and inference is going up. So I'm curious, what is the gating factor to getting to that, let's say, whatever, 20%, 30% market share? What is your share assumed in your $6 billion target model?

Let me start at the beginning. So the thing that is setting the timeline to get to, say, 25% share of GPU is how efficiently we execute our fast follower strategy, how fast we can bring up test programs and test solutions for devices that are early enough in their life cycle that we capture a significant portion of the ramp. So that ultimately is the limitation. As time goes on, the ultimate phase of fast follower is something that I've referred to as tester-agnostic development. So what many of our customers are talking about in this space is that they really want to develop their test solutions against a -- like think of it as like a virtual test system.

And by flicking a switch in software that they can target that towards our platform or towards another platform. When we get to that, then we are going to be in a world where it's much more about the differentiation on throughput and performance and availability than incumbency. And I think that we have certain advantages around test coverage for elements of the device. I think we have some advantages in terms of platform reliability. I certainly think that we have some advantages in terms of responsiveness to demand.

So -- and what we've seen, like one of the reasons that we're -- that we feel like we know what we're talking about here is that this is basically the world that we live in, in high-performance memory, that we were later to market than our competitor when it came to HBM performance testing. But once we released a platform that delivered better economics and better performance, then we began to see significant share gains because the customer can choose which platform they're going to buy, and we were able to capture that. As the AI accelerator world migrates more towards this idea where there are solutions existent on both platforms, then we think that we can compete on being able to get them the capacity that they need and getting the most parts out of each test cell because at the end of the day, it's like the limitations are turning into things like floor space and number of probers and handlers that they can buy. So having very productive test equipment is a potent advantage.

Now in terms of the part of the $6 billion, I think that we are looking at -- I don't think we need to be much -- like we can get to this, our model in the fast follower phase of this, where we're just doing specific part conversions. And I think we would only need kind of low double-digit share in order for us to hit our model.

And for my follow-up, I just wanted to get back to this visibility question. AI is over 70% of your sales. But when we look at any other semiconductor company involved in AI, logic or networking or memory, right, even the semi-cap front-end players, they all claim to have great visibility, not just for this year, consistent sequential growth and then visibility even in 2027. And I'm curious, why is that not translating into stronger visibility for the testing part, right? Because you guys are an important part of that supply chain also. So how come everyone else have great visibility and confidence, but we are not hitting that, right, from the testing side as much?

So I think the point that Michelle made is really important that the lead time for a tester is on the order of the same as the lead time for the actual wafer, not the wafer front-end equipment. And these are -- our customers are rapidly trying to build out capacity to be able to support all of the phases of production. And so we're -- like at the end of the day, these customers are definitely leaning hard into creating the front-end capacity to increase the number of wafers that go through. But until they actually are seeing the wafers go through, they are holding back on the orders for the test equipment.

I mean we definitely are working against a long-term plan around capacity expansion for our products. We have an idea of our -- think of it as like a strategic forecast that we're working against, but that's very, very different from the level of commitment that we get from our customers around what they'll buy and when. And since the -- like testers are sold for particular devices and particular device ramps, those ramps can move significantly if you have an issue with -- like if it's an ASIC, if the first silicon doesn't work, then that can inject a 2-quarter delay in a ramp. That would have a meaningful effect on the timing of our growth. It wouldn't have a meaningful effect on the long-term growth that we'll achieve. So I think the front end is less lumpy, but growth in the front end inevitably leads to growth in the back end. It's just uncertain in the timing.

We'll take our next question from Jim Schneider with Goldman Sachs.

Relative to the CPO opportunity, I think you did a good job kind of outlining where you believe the market is going. Just kind of 2 follow-ups on that. One is, could you maybe level set us for where you expect your CPO revenue to land in terms of a range for this year, 2026? And then you talked about kind of a 4-way kind of tie-up for -- to pull off some of the second insertion solutions. Do you have any kind of plans or thoughts about how you might integrate that a little bit more under one roof in some way or another?

So I don't think we're publicly disclosing our expectations for 2026 revenue, mainly because there's so much uncertainty about where in the test flow the investment will go. So will they lean harder into insertion 2 or insertion 1 or insertion 3. So there's a lot of noise on that data right now. So we're not trying to make a prediction. The 4-way partnership is actually -- that's kind of the way the world works normally that if you think about non-CPO devices, there's a fabless specifier, there's a foundry, there's an OSAT, there's a handler or prober provider and there's a tester provider and there's a probe card provider. There's -- like there are a lot of -- there's a whole ecosystem that makes these test cells work.

And we are a real strong advocate of this open ecosystem long term. We think that, that's what our customers want, and it's how we want to try and work. And we also want to make sure that all of the providers in this space for probers, handlers, whatever, that they feel like we are being -- that we're treating people alike, that we're not showing favorites. So what we're really trying to do is we're trying to lean hard into helping ficonTEC, build their capability in the semiconductor test equipment space, but we're acknowledging that they are the world experts in active alignment. So they're a great company. They know what they're doing. We want to help stand them up to be a great member of this ecosystem versus trying to integrate them.

That's helpful. And then just as a follow-up for Michelle. Clearly, the growth and margins and everything else are kind of turning out very well and it seems like you will indeed probably get to your model at some point. Just from a philosophical standpoint, from an OpEx perspective, to the extent you do tend to grow on a multiyear basis a lot stronger, would you tend to believe that you could actually underpunch the OpEx intensity? Or would you plan to devote more OpEx resources to R&D over time?

So the answer is yes. And so we've historically talked about growing OpEx at 50% of revenue. And we still believe in that model. We believe in investing back in the business, particularly in today's environment. As we think about the pain points in our kind of wafer to data center strategy, there's lots of opportunities for us to help our customers. So we're going to look for those opportunities to reinvest back in the business, which will impact OpEx.

When we think about the short term, however, in the current year in such a hyper growth mode, we wouldn't expect that same translation to play out within 2026. But definitely over the midterm and the longer term, that's the aspiration that we would be driving to.

Yes. The only color that I'd add to that is there is such a high rate of technology change in this end market and so many interesting opportunities like there is a ton of waste due to yield issues and quality escapes. It's kind of a perfect environment for a company that is delivering something to help other companies get to high quality. So we are like for a while, we were in a world where like the marginal utility of additional R&D spend was not that great. Like if you look back a few years, now we have just a really long list of great things that we can invest in that will drive long-term revenue growth.

So we're going to do our very best to constrain the growth of our G&A. We're going to do the very best that we can to manage the growth of our sales and marketing so that most of that is customer-focused technical investments, and we're going to lean into R&D because it's a target-rich environment.

This concludes our Q&A session. I would like to now turn it back to our presenters for any additional or closing remarks.

Thank you, operator. So thanks, everyone, for joining the call. We are really looking forward to Q2, another really strong quarter. And I just want to reiterate our thanks to the team around the performance for Q1. It was -- we're calling 2026, the year of execution. And we are like hitting on all cylinders. The team is doing really, really well, and we appreciate you having the interest in the company.

Thank you. This concludes today's Teradyne First Quarter 2026 Earnings Call and Webcast. You may disconnect your line at this time, and have a wonderful day.

My name is C. J. Muse, semiconductor equipment analyst with Cantor Fitzgerald. Very pleased to have Teradyne. And with us is Greg Smith, President and CEO. Welcome.

Thank you.

Always good to see you in New York. So I guess, near term, wanted to start off with how we ended the last quarter, which was very strong visibility first half, some uncertainty around the second half. I'm curious how that visibility or the dynamics there have evolved over the last 4, 5, 6 weeks.

Yes. So one of the things that we've been trying to help describe is that we feel like we're in a new era in the way that our market works that for a very long time, we were working against a sort of a calendar sequence, that there was -- like Q2 and Q3 were always strong, and that was driven by primarily like mobile and consumer. The new pattern that we're in doesn't really obey year boundaries. And so what we're in right now is like a very strong period that started -- the leading edge of it was in the third quarter of last year, and we have visibility that it's going to continue at a really good pace kind of through the second quarter this year.

And we're cautious about the year after that, not because we think there's going to be any kind of a pullback or that there's like a market turn or anything, but just that many of our largest customers are heavily investing during this 4-quarter boom, and we think that there's going to be brief period of digestion before they get into next-generation programs and ramps and everything else. So part of it is that we can't see all of the short-cycle business that will occur in the second half of the year. And there certainly could be strengthening in some parts of the market, like the place that I'd be looking most for like a second half boost tailwind would be like in the memory market. But a lot of the bullets that were in the gun are getting fired in the first half of the year. So we don't have as many opportunities to overperform in the second half.

Makes sense. Maybe moving to AI compute. I want to start with VIP. And you've talked historically around 50% market share for Teradyne. And you did a while ago talk about an $800 million-plus kind of framework for revenues by 2028. I think since then, things have obviously improved. Not asking kind of for an update on that revenue number, but how we should be thinking about the relative growth contributions and whether that share is still attainable?

Yes. So I think the plus is doing a fair amount of work in that sense that we definitely think that there's upside to the $800 million by '28. The thing -- in 2025, we think we held that sort of 50% share in this space. But there are only a very small number of hyperscalers that are driving a lot of volume. And so depending on the timing of program ramps between those 2 hyperscalers, we'll see the share swing. So we've got majority share at one. Our competitor has majority share of the other. As those programs take turns ramping, we'll see share move around. I think that we're going to continue to see over the midterm, additional hyperscalers ramp, and we think that we have a reasonable shot at capturing share in those hyperscalers as they ramp. So I think we feel pretty good that the TAM is probably trending above the $800 million mark. And we think that over time, that we're likely to maintain that 50% share, but it's going to be pretty erratic because it's very lumpy.

So I think historically, the business has been dominated by one large customer. And I think in our conversations in the past, you talked about good breadth of XPU attach as well in there. Curious, is there potential to secure other hyperscalers that are currently using your competitors' tester? Or are there other potential large players entering this market?

Yes. So there are a couple of things that are going on. One, that we think that hyperscalers are going to continue to show a limited amount of loyalty to the design partners that they're working with. So just because the hyperscaler is working with a particular ASIC company in this generation doesn't mean that they'll continue to work in that pattern. And those are not -- I wouldn't call them jump balls, but they are a much better opportunity for us to gain share than if it's just the next generation of exactly the same supply chain.

The next thing that's going on is as these programs ramp to scale, the amount of tester capacity that they need tends to go up by a lot, and that makes supply chain assurance more important to them. So the idea that if they are struggling to get capacity from one supplier that they can get capacity from a different supplier gives them multiple shots on goal.

So I think that's both a positive and a potential negative to Teradyne that like when a customer is sole source on our competitor, we think dual sourcing is a great idea. When we are the one that has the sole share, we don't think it's such a great idea. But overall, because of our share position in compute, we think that this trend towards dual sourcing is a net positive overall. We also feel pretty confident about our customer satisfaction and the value proposition for our testers. So we're not really worried about significant share erosion due to dual sourcing.

And are you seeing dual sourcing on the ASIC side? Or is that something you see in the future?

Oh, no, no, it's going on right now. So I mean, our strongest hyperscaler account is dual sourced, and they leave a fair amount of the decision to the OSATs and the foundries, and we do really well in that environment.

Got you. Maybe moving to the GPU side of things. You've talked at a very high level around initial win kind of this year and then the hope for incremental share next year and beyond. And I guess, what can you share with us today in terms of kind of customer feedback? And what is the kind of road map for success?

So I'm trying to figure out like we are in a -- like no news is good news kind of a mode. Nothing has gone particularly wrong in the process of achieving qualification. We still expect to achieve qualification in the first half of this year and be in production in the second half of this year. We believe that will start a long process of share gain in the account. So in 2026, we're talking about low-single digits of share of GPU in this merchant account. But the other thing that happens once qualification is achieved is you get follow-on projects.

You begin to start on new parts that you are going to bring up on the tester and we'll be in what I refer to as a fast follower mode that like the engineering to introduce a new part to production would happen on the incumbent platform, but we would be able to work on a fast follower strategy from a much earlier point in that product's life cycle. So that is going to start as soon as we have qualification. And those parts when we are finished with our solution would still be in a much earlier part of their volume ramp.

So that is kind of the story that what will happen in 2027 is we'll have more parts earlier in their life cycle, which will drive our share higher. But it's probably going to take us 3 or more years to get to a mature competitive dual sourcing environment where what we've seen in other accounts is that's like 2 competitors duking it out and the share is managed in a 30% to 70% range. And the company that gets the 70% is the one that delivers the right solutions for leading-edge technologies, delivers the best reliability, best time to market, best customer satisfaction.

So it's not like -- when I say 30% to 70%, I want to make sure that people understand that we don't see like 70% is the lowest that the incumbent goes and 30% is the highest that we go to as a challenger. It's that the share would get managed in that range on the basis of customer satisfaction.

Makes sense. And if I look back at mobility, where you would have typical kind of dual sourcing, it was not for the same part, but different parts. So are you expecting it to be the same? Or would one actually -- let's make up an example, a logic die. Would there be an opportunity for 2 different testers for that part or?

For sure. I mean one of the big differences between mobile and the AI space is the capital intensity against one specific SKU. So -- and also, when you are building up to this competitive dual sourcing, since we will be doing this fast follower approach, we will not have any solutions on our tester that don't already exist on the incumbent tester. So by definition, those are going to be one part hosted on 2 different platforms. Do we expect that there are going to be some parts where we have much higher share than other parts? Yes, we do. But it's not like in mobile, it's like you get RF, you get digital, you get power management. It's going to be much more intertwined than that. And it may end up being like this die it goes to competitor A, this die it goes to competitor B. The final test is a competitive decision between the 2 platforms. That kind of thing could certainly happen.

Got you. If I think about the future road map for front-end manufacturing, high-performance compute is catching up to mobility. And it certainly feels like with Backside Power, GSM that they're going to leapfrog mobility. And so curious, as we see that transition, what impact does that have kind of on test intensity and test insertions. How are you thinking about the high-performance compute impact to the growth rate for Semi test?

Yes. So it's interesting because for like a decade, mobile was the place where the latest and greatest process technology was used in SoC, that like the first into 5-nanometer, the first into 3-nanometer and still the first into 2-nanometer. But the hunger for compute, sort of like anything that you can do to increase the tokens per watt is -- there's a very strong proposition to try and do that. So we think that despite the fact that these advanced nodes typically have lower yields and the die sizes in compute are huge, that these customers are willing to accept that proposition in order to leverage that technology in the data center.

That's great news for tester companies because if you need to test 4 parts to get good part, you need a lot of wafer sort capacity to try and pull that off. And that -- so that's a positive to us. The other is in order to try to achieve the levels of compute per watt that they're looking to do, a lot of that has to do with die-to-die connectivity, right, that it's a scale up. And the basis of the scale up, the most efficient scale-up that you can do is by co-mounting as many compute die into the same package as you can before you even get to network-based scale up.

So that means that these packages are going to go from having 2 accelerator dies to 4 accelerator dies. That puts an amazing amount of pressure on die quality. So not only do you have lower yield, but you are testing out to a higher level of coverage, which also expands the test time. So like this trend towards using the latest node for compute is a really good thing for the overall TAM.

Perfect. You mentioned scale up. So a good segue to AI networking, historically kind of the bread and butter for Teradyne. Are you worried about this market dual sourcing? Or how are you thinking competitively?

I think like I think it's pretty clear that dual sourcing is to be a thing for most important high-volume sockets. So we have to look at the networking business as something that needs to be defended with product differentiation and customer satisfaction. So I don't think we have a monopoly on like we'll be the sole beneficiary from dual sourcing. But we have a very good product for this space. And I think that the other thing that goes on in networking is if you look at the ratio of big networking die to accelerator die like in a rack, it's like 10:1. And so one of the key things that enables dual sourcing is you have a high volume, right?

That -- and so the volume mechanics are different in the networking space than they are in the accelerator space. And that's one of the things -- that's one of the levers around like when would a company say, this part needs so much capacity that I need the supply chain assurance of dual source. So I think with networking, there's a pretty big order of magnitude difference in terms of the amount that you need to get there.

No, I think part of your networking growth rate was kind of muted a bit because of excess mobile capacity and kind of reallocation. I imagine that -- I believe that's now behind us. So I'm curious, with that no longer as a headwind, how are you thinking about kind of the relative growth rate here? And as part of that, would love to hear as networking becomes more complex, how are you seeing kind of the attach rate of networking dollars relative to compute dollars?

So we came into 2024 with a belief that networking was kind of like 15-ish percent of the space. And what we saw in 2024 and 2025 was a really rapid expansion of the TAM for the accelerators for the GPU itself. And later in 2025, we are now seeing networking catching up again. So moving forward, I think we're at kind of a new normal in terms of the size of all of these markets. And our assumption is that they're going to grow relatively proportional in terms of the core technologies today. And the X factor is how rapidly CPO and optical technology will enhance the networking TAM. So that's like on top of that sort of 15-ish percent ratio.

Makes sense. So you acquired Quantifi building out your portfolio for silicon photonics. I guess how do you see that kind of ramping as we focus on scale up?

Yes. So I feel like I'm like Mr. Cold Water at this conference around silicon photonics, even though it's something that I'm incredibly excited about, right? So if you look at silicon photonics right now, there's really no large-scale deployment of CPO in any way in 2025 and at the beginning of 2026. By the end of 2026, I think we will see the beginning of commercial scale deployment of CPO for scale-out applications. And that is like -- and I think it could grow geometrically from there, just in scale out, like '26 to '27, just scale out could be like an order of magnitude difference in terms of the size of -- like the number of CPO ports that are delivered.

And then scale up would be another multiple beyond that. But that's just one factor in terms of the size of the test market. If you look at how CPO devices are tested today, it is very early days for that market. So all of the technologies around optical alignment, the tests that our customers are performing are very characterization oriented. The instrumentation required to do that testing is more sort of laboratory-focused instrumentation versus production focused.

And when we've seen new technologies like this introduced into the ecosystem, what we see is a really rapid improvement in test efficiency. So between more efficient test equipment, reduced test lists, improved efficiency for alignment and eventually testing more than one device at the same time, the -- like the test intensity for these devices could go down by a factor of 10. And both Teradyne and Advantest are leaning into developing production-oriented optical test equipment. That's why we did Quantifi. That's really around what testers are going to look like in 2027.

So those testers are going to -- for both Teradyne and for Advantest are going to be -- their ASP is going to be lower. We're going to be competing with each other. But we will be all native content or mostly native content. So right now, we're buying a lot of instruments from other people, putting them in a tester and then selling them again. That's a lower margin proposition than for natively developed testers.

So the revenue in -- so what I think is going to happen? To directly answer your question, what I think is going to happen is that this initial deployment of scale-out is going to drive a rapid increase in the size of the TAM at the end of '26 and into '27. That TAM is probably going to reach somewhere in the mid-hundreds of millions of dollars. And then as the volume in the end market scales, we're going to simultaneously see this reduction in test intensity and reduction in capital required. And so even though we'll ship 10x as many CPO ports in 2028 as we do in 2027, the TAM is only going to be incrementally larger than not multiplicatively larger than what it was.

Makes sense. Maybe moving to memory. Teradyne has done a great job securing wins at 2 of the leading DRAM players for HBM and obviously, the third does internal testing. Maybe if you could update kind of where you've had success? And perhaps maybe more importantly, as we start thinking about HBM4 and start moving kind of higher layer counts, what that means for test intensity and perhaps incrementally more test insertions.

Yes. So let's -- first of all, let's just sort of talk about the insertions involved in building an HBM. So an HBM has multiple layers of DRAM die and then a base die. Each of those layers needs to be tested at the wafer level and the base die needs to be tested at the wafer level. Then you stack the DRAM die on the base wafer, you test it again for performance as a stack assembly, then you singulate it. And right now, some -- for some of that volume is being tested again in a singulated form and then you ship it off and it gets put on to a substrate to be used with an accelerator.

So there's 3 major test insertions. We are essentially getting -- like our biggest impact is in the stacked wafer test -- performance test and the singulated stack test. The wafer sort is like the wafer sort has always that we have a strong presence in some manufacturers, not a strong presence in others. And so there isn't a lot of share movement there, whereas we have been gaining share in the post-stack tests.

So the question is, okay, so what happens to that string as you go from HBM3 to HBM 4 to 4E and then when you go from 8 high to 12 high to 16 high. As the layer count increases, that primarily drives test intensity at that first step. You need to build more wafers to build 1 HBM stack. So that's kind of normal by customer share, not a lot of movement. The technology change is what drives the later insertions. And we have a tester that has higher data rates and higher signal fidelity. So as we get to the -- to like 4E, we believe that we're going to be able to continue to increase our share in that space and differentiate with our customers around letting them get to higher yields.

And you have the one company that does in-house, and I think they publicly stated that they're going to build the logic die at TSMC. Does that represent an opportunity for merchant test player?

Yes. So I mean, logic dies already get tested. And in most cases -- well, and sometimes they get tested on our platform, sometimes they get tested on an Advantest platform. And sometimes they get tested on memory testers. As we go to HBM4E, the complexity of that base die is increasing. And it is an incrementally larger part of this whole value chain, but it really is dwarfed in comparison to the memory investment, the memory ATP investment to support the whole chain.

Got you. Makes sense. Maybe conventional DRAM, obviously, a place of strength that we're seeing today despite shortage. Just curious if you've seen sort of any movement there.

So 2025 was a remarkable year for DRAM manufacturers because they were able to remarkably increase their revenue while shipping basically the same number or DRAMs. So not a lot of unit volume change in that space, but much higher ASPs. So from a test perspective, it was kind of normal capacity increases, not like proportional to the kind of revenue increases that you're seeing. The thing that we're looking at is there are big foundry investments going into memory for stuff that's going to come online in '27. And I think we're kind of in the -- the memory market that we're in right now is probably going to be stronger in '26 than it was in '25.

But I think the breakout year for memory is really '27 when there's a lot more wafers that need to get tested. So that's kind of the way I'm looking at it is that we've got a little way to wait before you see a huge bump. The one short-term thing that I think is really interesting is there's more of a reliance on SOCAMM in the next generation of accelerators, content addressable memory. And that is often being built out of LPDDR, LPDDR5. And because we have a really good position in DRAM final test for that kind of a technology, that's a net positive for us. So the more SOCAMM that's going into the market, the better it is for us.

You said foundry investment in memory. Is that your way of saying kind of new greenfield investment by memory guys?

Yes, new fabs.

Okay. Perfect. And maybe on NAND, are you seeing any sign of life on the enterprise SSD side or elsewhere and do you have any kind of view on KV Cashe and what that might do?

So we are like -- we believe that we should be seeing those signs, and we are not that right now, the memory market is wafer constrained, wafer capacity constrained and the allocations of wafers towards flash have not been increasing. The equation doesn't balance, though, because the amount of flash required per rack in next-generation data centers is like multiples of what's required in current generation.

So it's possible that flash is going to turn into a significant enough bottleneck that you'll see some changes in behavior, but again, I think the 2027 is when there's going to be enough incremental wafers to really make that flash market grow. I -- we have been seeing more strength in the HDD market. And I think some of that has to do with the inability to source enough flash to be able to do what they really want to do with flash.

Makes sense. It's surprising to a minute -- 2 minutes left to hit on your more cyclical business, but I think it speaks to how great AI is today. So in the last few minutes that we have, I would love to hear kind of your latest thoughts around mobile, auto, industrial.

So I think that like one thing that is becoming clear to me, and I think it's important is the next generation of AI accelerators are going to need next-generation equipment from both us and our competitor. And that's because you need more power, you need more pattern depth, you need things like PCIe 6 capabilities to test those generations of parts. What that means is there's a lot of really good testers, testers that have been introduced into the ecosystem even this year that are not going to be useful for leading-edge AI accelerators anymore. And I think mobile is going to get the hand me downs.

So I'm expecting that the mobile TAM is going to be somewhat suppressed because of this really high buy rate for next-generation equipment and compute. So we've never been -- like we've been pretty cautious in terms of a recovery in mobile. And I think that's another reason why were kind of staying cautious about how much that could pop because there's going to be capacity coming available. The other -- like when you look at the other segments like automotive and industrial, I think we are kind of at the dawn of a cyclical recovery. Inventory levels are down. We're hearing more optimism in the way that those companies are talking about their results. So I think we'll see a modest recovery there.

Perfect. Well, I think we've run out of time. Thank you very much. Great. appreciate it.

No, it's always great to talk to you. Thank you.

Hi, I'm Shane Brett, U.S. semiconductor equipment analyst. I have a quick disclosure to right. For important disclosures, please see the Morgan Stanley research disclosure website at www.morganstanley.com/researchdisclosures. If you have any questions, please reach out to your Morgan Stanley sales representative.

I'm honored to host Greg Smith, President and CEO of Teradyne. So let's get started.

I want to start with a few longer-term questions about Teradyne and the broader test industry as there's just some remarkable developments underway. So just beginning with kind of the industry environment right now, semiconductor test was about $9 billion in 2025. And it surpassed the 1% buy rate relative to the semi industry. Could you walk us through the kind of key drivers behind this increase in test intensity to sort of these levels we haven't seen before?

So it's a really good question. So the first thing that I'll say is that like buy rate as an engineer, I kind of object to the notion of by rate because it doesn't -- like it's accidentally right, but it doesn't reflect the actual dynamics of the market. So people need to buy new capacity to support revenue growth, not to support the same level of revenue that they had before.

So what you'll see in markets that are strongly growing is that the buy rate relative to revenue is actually going to go way up. But at -- when the revenue ends up being stable at that higher level, then the buy rate drops down. But it's not that the capital intensity is reducing. It's that you need new stuff to build that increased capacity. And so what we're seeing right now is a massive increase in the capital intensity associated with AI compute. And that's in the memory space and in the SOC space.

And it's even more remarkable because if you're talking about a buy rate, that's the ratio of revenue, IC revenue on the bottom and test your revenue on the top. If you look, the things that are driving the revenue are actually the high-margin part of the memory business, so HBM and the high-margin part of the SOC business, which is the AI accelerator stuff.

So if you look at the test intensity as like a ratio of the device COGS, it's a much larger increase because so much of the money that's being made in memory is on ASP increase and so much of the money that's being made in the AI accelerator space is on the high margin in those devices. So it's actually -- like it's crazy, high increase in test intensity.

So I guess to kind of sum up the argument, buy rate is probably something we should look at through cycle, and the kind of buy rate relative to COGS will be the homework that we should do.

I think the thing that I would -- for model builders, I would say that the thing you want to do is you want to look at it's a derivative market. So you want to look at revenue growth as the primary thing that drives ATE TAM, not total revenue.

Got it. I kind of want to -- on that sort of -- I'm sorry, I'm going to dig into this buy rate a little bit, but it doesn't appear that the sort of test intensity is set to slow down anytime soon. But we also to remember what happened in 2020 and 2021. Just how is the current cycle different from the kind of SOC boom we saw 5 years ago? What differentiates this environment from that period?

Yes. So in -- so by the way, like I think of this as what happened in 2022 and 2023. So in 2020 and 2021, there was a really rapid increase in the size of the ATE TAM and Teradyne revenue growth was really terrific. And it was really all around the mold boom. And what happened during that period of time was it was a period of rapid improvement in mobile phones, which drove a higher refresh rate. People bought phones more often. We had a global pandemic, and so disposable income, like people no longer were spending money on going out to restaurants. And so all of a sudden -- but they were spending hours every day staring at their phones. And so they wanted to -- they wanted the best latest phones. So there was an increase in the refresh rate and also an increase in the mix of phones towards the premium tier.

All of that combined to really create a surge of demand around the mobile segment. But it was very, very concentrated and very driven by the external factors around the COVID pandemic. What we see here is something that is much broader and driven by factors that have much more longevity. Essentially, we are already hearing about NVIDIA talking about demand firming out in '27. The time line to build up data centers is such that people are like buying the land, trying to get the gas turbines, all this stuff out multiple years and the rate of growth of AI revenue for the AI players is such that they're leaning very, very hard into this capital expenditure.

For sure, there are downside risks against this cycle, but it looks entirely different than what happened in 2020 and 2021. It's far more of a like the difference between a hype cycle and something that is probably sustainable at least over the midterm.

If I were to be a bit of a pessimist in say, 2026, superhit test intensity, '26, we may see a digestion. Your kind of rebate will be there's a much broadening of the customer's AI revenue, seems to have a little bit more durability and longevity. These kind of CapEx plans are kind of spread through.

So I feel like there's waves in the AI space. So in 2025 and into 2026, most of the growth in memory is really revenue growth driven by ASP increase. Like more people are chasing after the same number of parts. We are not seeing -- we're seeing very good healthy TAMs, but we haven't seen the impact of significant capacity increases in memory. That's more of a '27 and beyond thing. We are seeing significant growth in AI accelerator and in networking, but none of that incorporates the complexity and the TAM expansion that we would expect to see from the introduction of co-packaged optics for scale out and scale up. Those are things that are likely to like be the next waves falling on the shore.

And beyond that, we are still in a world where Edge AI is not yet a market driver. But in terms of delivering a compelling user experience, there's a future where smartphones are running large language models. And in order to do that, the complexity of the silicon in those phones, the amount of memory in those phones is going to need to be significantly increasing. And so that would be like the next wave crashing on the shore. And because of Teradyne's exposure beyond just wafer to package, we go all the way from wafer to data center, the changes in data center architecture are a tailwind for our product test business. The need to build and maintain these data centers and to build the equipment is a tailwind around our physical AI and robotics business.

So we think that there's a fundamental transformation of the global economy that is happening because of AI. And the thing that drives that whole transformation is the silicon.

Great. Great. I guess then let's jump right into compute. So 40% of your SSD revenue, just largely driven by networking and VIP. Just can you discuss what you're seeing across those kind of 2 customer cohorts?

Yes. So in networking, Teradyne has a very good market position in the networking part of the business. The most remarkable thing in the networking business has been how quickly NVIDIA has grown in that space. And so if you look back a few years, the network part of the market was it wasn't very fragmented, but there were a lot of players that mattered.

Now essentially, it's kind of 3 players that matter, and we have a good position with all of them. And so as data center architectures are becoming increasingly dense from a network connection perspective and the network switch technology is increasing, we're seeing increased test intensity and increased unit volume for this business. So we think that, that showed significant growth from '24 to '25, and we think that it's positioned for growth out in the future. So that's networking.

Now VIP compute, the thing about VIP compute is there are a lot of hyperscaler ASIC programs, different device types, different hyperscalers that have their own ASIC development programs that are going on. At this point in time, there are only really 2 that are at scale. And then if you include Edge AI, automotive applications, it's 3.

It's possible and even probable that other hyperscaler ASIC programs are going to scale, but they have a really high bar to try to produce more tokens per watt than what you can do with merchant GPU. And networking is a -- like it's a very difficult challenge. And so for particular hyperscalers, if a lot of their workload really comes down to something that can be reduced in silicon, video encoding or something like that, I would expect those things to ramp but at a much lower complexity than the merchant GPU or the ASIC programs associated with training.

So right now, that market is very concentrated. There's kind of 2 major programs that you want to be a part of. And as of the end of 2025, Teradyne has kind of fought our way to about a 50-50 share market, which is much higher than our position in merchant compute.

I want to go back to the networking portion as customers are sort of pushing the envelope with increasingly complex looking photonics or co-packaged optics and you're sort of enabling that. Just how high do you think that sort of -- what are you seeing in terms of test intensity for networking and kind of your demand for that end market?

So there's a lot of preparation work going on right now for silicon photonics and co-packaged optics. It is not a meaningful part of the end market at this point in time. It's certainly optical networking is a big thing and optical connections at the sort of data center level and long haul totally in the optical domain. But for scale out rack to rack and scale up, it is not yet really had an impact on the market. That being said, there are many, many development programs that are going on as we speak. And those are driving a lot of activity in the market, and Teradyne is a very good position. We believe that this is going to begin [indiscernible] the second half of this year. And in 2027, we it will be meaningful but not a huge deal.

The best way to put this into context -- and by the way, there might be upside from this view. But if you look at our long-term model, we have this long-term model that is around like what are the conditions for a $6 billion revenue at Teradyne? In order for Teradyne to be $6 billion of revenue, we expect the ATE TAM to be in the $12 billion to $14 billion range. Inside of that $12 billion to $14 billion TAM, we think that the total computing TAM, so that would be merchant computing, VIP computing and networking, that that's probably about $8 billion of that $13 billion. So big. But as you divide that, the merchant computing part of that is probably more than half of that. And then the rest is divided between VIP compute and networking.

In the time frame that we're talking about for this model, where you're talking about this $12 billion to $14 billion TAM, we would expect that the silicon photonics and co-packaged optics would be 1/3 -- 1/4 to 1/3 of that networking TAM, so hundreds of millions of dollars but not like a game changer for the whole ATE marketplace.

I guess I want to go back to that $8 billion compute. So when we think about it, you said half of it comes from the merchant GPU opportunity where you're starting to make inroads. Half of it comes from networking VIP, networking where you have dominant share of VIP where you're splitting it. When I think about the kind of merchant GPU opportunity where you haven't historically had exposure, could you kind of remind us what's the progress on kind of penetrating into that market and what's the kind of next steps in that progress?

Yes. So what we've talked about before, and we are on track with what we've said before, is that we believe that we will achieve a tester qualification in the first half of 2026 and that we will start to see some revenue -- some production revenue associated with this in the second half of 2026. But it would be like single-digit -- low single-digit percentage of share for merchant GPU in 2026. So it's going to take a while to take off from where it is.

That program is proceeding on schedule. We are still on track to achieve that. And let me describe a little bit about how this works. When you're trying to qualify a new tester, what you want to do is you want to make the new tester the only variable. So you pick a stable high-volume device as your vehicle to do that. And so as you go through it, you can be pretty sure that if there's a difference in the results, that the difference in the results is due to a difference in the tester platforms, not anything related to the device itself.

Once you've -- and that project is pretty extensive because you have to convert all the libraries. You have to develop all the test algorithms to work across 2 platforms. Once all of that's done, then you've shown that, that tester is an equivalent and you've qualified the tester and the libraries. Now you can go into a mode. So the first thing that can happen once you achieve that is you can test that part. And so that would be the first production usage, is testing the specific part that you've converted.

The second thing that happens is you can go into what I would call a fast follower mode for other devices. And you would be focusing on devices that are earlier in their life cycle, and you would be trying to develop a test solution for those devices so that the customer would have the choice of loading the incumbent platform or the challenger platform. So that's the second phase, is this fast follower phase.

We believe that between this initial qualification in the fast follower phase that, that is enough for us to get up towards like meaningful share towards 30% share of GPUs. To get higher than that, there are 2 things that need to happen. One is that most customers in this space, including merchant GPU customers are exploring a DevOps approach called tester-agnostic development. So what they're trying to do is they're trying to develop the test solution without -- with abstracting the characteristics of the tester that they're going to put it on. And we've done this in the VIP compute space. We are applying it in the merchant GPU space. And so what that means is that the customer can defer the decision about which tester to put it on until later in the process.

The other thing that happens is as you do more parts in a fast-follower mode, the customer gets more familiarity with your platform and you have the opportunity to compete for to be the sort of the primary platform for a new device. And that's something that's probably 3 years out in the future. It takes a while to get to that point. But at the point in time when you have a qualified platform, then it becomes more around the characteristics of the tester. So if, for example, we have a better solution for a new technology like PCIe 6, then we might be selected for a part like that as the primary; or if we're able to handle higher device power requirements, that would be a way that we could -- like once you're qualified, you can then win those kinds of opportunities. And it's like -- it's not that I would expect to run the table and own the account, but it gives you more opportunities for share gain once you've achieved qualification.

I guess to kind of sum up what you said there, so I guess the initial step is getting into a high-volume manufacturing product. And then the next step would be getting to a fast follower status and starting to get kind of expanding your share from there. And if your customer goes towards a tester agnostic platform, then you're kind of on a level playing field.

Yes. So like a test agnostic platform, you can almost think of that as faster follower. So it makes the supporting multiple platforms even easier than if you have to do a part-by-part conversion because They would use that tester agnostic development approach even in new part development, but they would like, by default, go to their incumbent platform because they know it so well. And so it just -- it reduces the level of effort to support multiple platforms by a lot.

Got it. Got it. Got it. I want to shift over a little bit towards memory as DRAM kind of -- you've broken it into compute. DRAM is also an area where you've sort of gone back to that level playing field. Just could you begin by just discussing the changes you're observing in DRAM, particularly as just wafer test evolves from a commoditized step. So you got some more performance-oriented one and is where this DRAM test intensity could get to? So loaded question.

Well, no. It's -- there's a lot of DRAM wafer sort that is still commoditized. So let's talk about an HBM memory for a minute. So to build an HBM memory, you have a stack of DRAM die. Those DRAM die get tested at the wafer level and that test at the wafer level, before anything gets stacked, that is a relatively low-tech test insertion. You're validating the core of the memory. You're testing that every bit works. You're validating various timing parameters. That is a tough test step to differentiate on because the test time is largely set by the architecture of the device.

Once the DRAM die gets stacked on the base wafer, now you're into the performance test. And that performance test is quite differentiated between tester suppliers that you have differences in capital lifetime in signal integrity, yield, throughput, all of that is there's a lot more differentiation potential at that test step. And then after the wafer gets diced up, each of those singulated HBM stacks, in some cases -- in most cases, that stack is not tested again before it's put onto a substrate with an accelerator, but there is an increasing trend towards testing each one of those stacks by itself. Okay? So singulated stack test.

Our gains have been primarily in the area of stack -- wafer stack test. And in that space, we've been able to get better than 50% share against the commercial part of the market. There is one major memory manufacturer that has their own test solution, and so neither us or Advantest is getting that part. But for the part that's up for grabs, we're kind of splitting that. Advent test share is higher in the wafer than nondifferentiated wafer sort space. Our share is very good in the stack test and also in like traditional DRAM, DDR5, LPDDR6. We have good share in that final test as well.

Now the way to think about this is for HBM, as you look at HBM, there are kind of 2 parameters around what describing HBM stack. One is how high and the other is the HBM standard. So if you go from 8 high to 12 high, that is a driver for the TAM of wafer sort. But if you go from HBM3 to 4 to 4E, that is primarily a driver for the post stack wafer test because you have new data rates. You have new test coverage that you need. That's the part of the market that gets driven by -- where we are strongest.

So across all 3 steps of DRAM test, we're sort of increasing intensity. So you guys are all kind of -- it's like a high tide lift all boats situation with DRAM test.

With -- for DRAM test, yes. I mean there is -- it is unlikely that there is going to be something that takes the wind out of those sails. If you look way back in time to like 2008, there was a fundamental change in like flash memory, where flash memory controllers could begin to work with imperfect flash memories. And more and more built-in self-test was being incorporated into devices and the site counts went way high.

We're now at a point for like memory wafer sort, it's one touch down and getting to higher throughput than one touch down for wafer is very high. And the rate of technological change for the final test is also very high. So the kinds of things that depress the TAM are absent and the kinds of things that increase the TAM like unit volume and complexity are going up.

Got it. Got it. I want to move over to a market where you've historically had really high share, which is SSDs and HDs just storage. What do you see in that market? Just now that we're seeing kind of the end markets for that market gets stronger, are you starting to see signs of that tester purchases pick up?

So from a HDD rotating storage, we are definitely seeing signs of life that -- and this is a part of the market that works against pretty long lead times. And they had -- even when demand in that market was quite low, we were still shipping at low rates and increasing the amount of capacity in the market because, essentially, the players there wanted to keep our -- they wanted to keep us in the market so that we -- like when they needed more capacity that we would be there to do it. So we went through that whole period of time. We have now gotten to the point where all of that installed capacity has been filled up. And the market for that is growing significantly from '25 to '26, and that's. In HDD, we also are winning new customers. So we've historically had 2 big customers in HDD. In -- we've now won a third customer in that space. And so we expect to grow associated with that from a share gain perspective.

And that will be in the system test.

So this is all part of the integrated system test group. And those results -- we aggregate those results financially into our semi test group, but we don't count that as part of the semi revenue or the semi TAM. So -- but because what we discovered is that as that integrated system test group was doing more and more system-level test of semiconductor devices, it really made sense for us to aggregate the go to market and the technology between the semi test group and that group. So we combine that. We combine those results, but we do break it out in terms of the SOC and memory revenue and the TAMs that we serve.

Now on the flash side, there's a disconnect that we haven't figured out yet. We see that from a data center perspective, data center architecture, server architecture that the amount of flash storage like the SSD that are required going from generation to generation is significantly like a triple in flash capacity. At the same time, we don't see the investment in capacity from the flash manufacturers.

On the front-end side.

On the -- well, from the manufacturing side, so in terms of the foundry investments that they're making or the test forecast that they have. So I don't think that, that mismatch is going to last forever. So it's either going to turn into additional upside in HDD or we're going to see sort of more investment allocation from the memory manufacturers in the flash space, and that's probably going to drive TAM increases for us there.

Got it. So DRAM pretty strong, HDD and SSD sort of waiting for demand to pick up.

No, no. So DRAM really strong HDD has inflected, will grow. SSD, we're waiting for a shoe to drop. Like we haven't seen that inflect yet.

Got it. Got it. Got it. I want to move over to mobile. So historically, really strong end market for you. We're seeing packaging transition towards more chip architecture, but unit demand is probably feels a little bit lackluster this year. Just how are you thinking about this mobility? What are your kind of outlook for '26 there?

So if you look at the -- there's sort of 3 things that affect the size of the mobile market. One is device complexity. The X is unit volume, so handset volume; and the third is tester fleet utilization. We think that 2026 is going to be like one of the best years we've had in a long time in terms of device complexity. And that's both transition -- node transition and also for the premier tier a transition in memory technology. So that is a tailwind.

The unit volume right now, it looks like that is likely to be a headwind. And that is kind of price elasticity with demand that if phones cost more because memory costs more, then people are going to buy fewer phones is the equation that people are working through in their minds right now. So we'll see. We'll see how much margin phone manufacturers are willing to eat in terms of higher memory costs in order to maintain unit volumes.

The third thing is tester utilization, and we have a huge fleet of testers to test mobile devices. And that gives our customers a lot of ability to optimize the way that tester fleet is used, and they've been able to -- they've been able to increase utilization in a pretty significant way over the past few years. We always think that they're just about out of cards to play there, but they continue to surprise us. So we believe that we're looking at a mobile market that is as strong or a little bit stronger than it was in 2025 in 2026, but there's some upside potential around complexity and there's some headwinds associated with the unit volume.

Got it. Got it. And I guess kind of wrapping up this talk on semiconductor test. Great things happening in [indiscernible], DRAM is great, mobile, we'll see. How are you getting your supply ready to meet this idle wave of demand?

Well, I think the same thing -- the same reason that we're seeing opportunity in the customer space. The main reason we're seeing opportunity and like merchant GPU is really around supply chain resilience, that, in essence, dual -- like competitive dual sourcing is a way to provide significant resilience in your supply chain. So we know that from the customers that we serve. We apply that same approach to our procurement. So we are primarily outsourced manufacturing. We have enabled our highest volume platform to be produced by multiple CMs. And we are working with both of those CMs to increase their capacity. And behind them, all of the component suppliers, we've established an envelope that is very aggressive against our 2026 demand picture because we are on a platform that legs. We're going to be -- that this platform is going to last a very long time. We want to make sure that we have enough material to be able to respond to upside demand. But the key difference between us and our competitor is this idea of using competitive dual sourcing in our supply chain.

If your customers come over to you and tell you, Greg, we need more testers, you will be ready.

Yes. And we'll be ready with really competitive lead times.

Great. Great. I guess last question. We've spoken about a really strong demand environment for '26. But at your latest earnings, you did talk about a first half weighted demand environment. Can you just kind of talk about that just what do you see between the first half and the second half? Is there anything different?

Sure. I mean, one of the things -- we talked about a first half weighting, and part of that is limited visibility into the second half of the year. Economic conditions, programs that have not yet ramped, share gain that we're not sure of. So part of it is uncertainty. But part of it is also that me of our most important customers are leaning hard into capacity adds in the first half year. So if that's the case, those -- it's like you have a certain number of bullets in the gun. If you shoot a bunch of bullets in the first half, you don't have that many bullets to shoot in the second. So that's part of it, is that we don't see the same kind of like untapped potential for the second half that we do in the first.

That doesn't mean that like we think that the market is calming down. It's more that the market doesn't follow pure calendar year cycles anymore. That -- and the -- like one way to look at it is that 2025 was back half weighted. 2026 is front half weighted. The other way to look at it is we're in the middle of a 4-quarter boom, right, that started in Q3 of last year, continues through Q2 of this year. And then there'll be a digestion, and then we'll go into the next. And I don't know whether we're working in a 5-quarter, 6-quarter 7-quarter new world based on what we're seeing right now.

Got it. Well, I think that takes us to time. Thank you very much, Greg.

Awesome. Thank you.

Ladies and gentlemen, good morning, and welcome to the Teradyne Fourth Quarter and Full Year 2025 Earnings Conference Call. [Operator Instructions] As a reminder, today's call is being recorded. I would now like to turn the call over to Amy McAndrews, Vice President of Corporate Operations for Teradyne. Please go ahead.

Thank you, operator. Good morning, everyone, and welcome to our discussion of Teradyne's most recent financial results. I'm joined this morning by our CEO, Greg Smith; and our CFO, Michelle Turner.

Following our opening remarks, we'll provide details of our performance for the fourth quarter and full year of 2025, for the first quarter of 2026, and our new target earnings model. The press release containing our fourth quarter results was issued last evening. We are providing slides as well as a copy of this earnings script on the Teradyne Investor website that may be helpful in following the discussion.

Replays of this call will be available via the same page after the call ends. The matter that we discuss today will include forward-looking statements that involve risks that could cause results to differ materially from management's current expectations. We caution listeners not to place undue reliance on any forward-looking statements included in this presentation. We encourage you to review the safe harbor statement contained in the slides accompanying this presentation as well as the risk factors described in our annual report on Form 10-K for the fiscal year ended December 31, 2024, on file with the SEC.

Additionally, these forward-looking statements are made only as of today. During today's call, we will refer to non-GAAP financial measures. We have put additional information concerning these non-GAAP financial measures, including reconciliation to the most directly comparable GAAP financial measures, were available on the Investor page of our website.

Looking ahead between now and our next earnings call, Teradyne expects to participate in technology or industrial-focused investor conferences hosted by Citi, Susquehanna, Morgan Stanley and Cantor. Our quiet period will begin at the close of business on March 13, 2026.

Following Greg and Michelle's comments this morning, we'll open up the call for questions. This call is scheduled for 1 hour. Greg?

Thanks, Amy, and thank you all for joining us today. I'll start off by summarizing our fourth quarter and full year 2025 results and provide some context for our initial view of 2026 and our new target earnings model.

Teradyne had a strong fourth quarter with 41% sequential revenue growth and more than 100% non-GAAP earnings growth. Both revenue and EPS were above our high guidance as trends we noted previously continued through the end of the year. Semiconductor Test, Product Test and Robotics, all delivered double-digit sequential growth. A striking trend was the increase in AI-driven revenue in the second half of 2025. This is obvious in compute and memory. However, the rapid build-out of cloud and edge AI is also driving demand for power management, SLT, HDD, ICT and optical test. This aligns with the themes of AI, verticalization and electrification that we have highlighted in prior calls.

When you roll it up, AI demand drove 40% to 50% of our revenue in Q3. In Q4, AI drove more than 60% of our revenue. Looking forward to Q1 of 2026, we expect that upwards of 70% of our revenue will be driven by AI applications. Now Michelle will go into a lot more detail about the quarterly results and trends, I'd like to give you a little full year color for each of Teradyne's businesses.

Starting first with the Product Test Group. Overall, we grew revenue 8% in 2025 driven by strength in defense and aerospace. We have successfully integrated Quantifi Photonics into this group, including training the sales team for Lifepoint and Production Board Test on the Quantifi Photonics line. We expect all of our business lines in this group to grow in 2026.

Turning now to robotics. In 2025, we saw consecutive quarters in growth starting in Q2. As we've discussed previously, we are optimistic about the value-creating opportunity in Physical AI and Advanced Robotics. And our strategy has been to focus the organization on the segments customers technologies with the highest growth potential. For all of 2025, the Semiconductor Test Group delivered 19% year-on-year growth. SoC test revenue grew 23% year-over-year, driven mainly by networking and VIP Compute. Memory Test revenue was up slightly in a roughly flat memory test market on continued share gains in HBM and DRAM final test.

With strong VIP revenue, we believe that we maintained about 50% market share in the VIP compute market in 2025. This entire segment remains very concentrated with only a few players driving significant ATE purchases. This contributed to revenue lumpiness in 2025 and complicates forecasting VIP share in the future.

Our full year financial results reflect a successful pivot to AI-driven demand in high-performance computing. Back in 2020 and 2021, our business was dominated by Mobile. We were highly exposed to Mobile in SoC, Memory and Wireless test.

Now in 2025, Compute is the largest component of our revenue and grew 90% year-over-year. This growth can be attributed to the decisions and investments we have made over the past few years that are now yielding. Our historically strong networking business has been growing because of high-density network connections in AI data centers and the increasing complexity of networking components. The work that we have done to align our product roadmap and customer-facing teams to VIP and merchant computing customers has enabled us to capture valuable design wins.

While we are gaining in Compute and Memory, we believe that diverse revenue mix is Teradyne's long-term strength. Using round numbers, in 2023, only about 10% of our SoC product revenue was in Compute, 50% was in Auto and Industrial and 40% was in Mobile. Now in 2025, nearly 50% was in Compute and Auto/Industrial and Mobile were roughly balanced at a quarter each. This balance derisks our target earnings model.

The SoC TAM reached record levels in 2025, nearly 60% larger than 2024. Looking forward, we expect that TAM to grow robustly over the midterm, driven by continued data center build-out and the growth of Edge AI. Predicting this growth rate from year-to-year is going to be difficult because of the high concentration and less predictable product ramps. One big socket sliding across your boundaries could have a significant positive or negative effect on year-to-year growth.

Although this uncertainty makes it challenging to predict the 2026 SoC TAM we are expecting robust year-on-year TAM growth. At a segment level, we expect Compute to grow significantly from a very high base driven by AI. We expect to see moderate recovery in Auto/Industrial, but we are uncertain about the Mobile TAM. Although we are expecting to see a significant jump in device complexity, there are questions about unit volume, product mix and capital efficiency improvements.

All in all, we believe that we are positioned to gain share in the single digits in SoC test and a significantly larger market.

Now shifting gears to Memory. In 2025, the overall Memory TAM was down about 4% from 2024, and we were able to gain a little share. A bright spot in the test, the memory test market was AI Compute demand for both HBM and DRAM.

Again, it is useful to take a longer-term look at the changes in memory test. Back in 2020 and 2021, the memory test market was split more or less evenly between flash and DRAM. In 2025, DRAM and HBM comprise nearly 90% of the memory TAM, a trend we expect to continue into 2026.

Overall, we expect a resurgent memory market in 2026 with low double-digit TAM growth over 2025 driven by continued strength in HBM and DRAM, and we expect to continue our incremental share gains.

Our IST business delivered over 50% growth from 2024 to 2025. Historically, IST has had very high segment and customer concentration. In 2024 and before, we served the HDD and mobile SLT markets, and our revenue was mostly driven by a large single customer in each segment. In 2025, this began to change. We won a new customer in mobile SLT in 2024, and that ramped strongly in 2025. Also in 2025, we entered compute SLT and won business from 2 customers in that segment. Finally, in late 2025, we received orders from a new customer in HDD, which will be ramping in 2026.

All of this is setting us up for continued strong revenue growth from IST in 2026 and beyond. Michelle will be going over our target earnings model in some detail. I'd like to comment on the underlying drivers of that model. In looking at the future, we had to answer 2 questions. The first question is whether the markets we are in are poised for growth. In our mind, the answer to that is unequivocally true.

Right now, the prime mover of the market is AI Data Center. Our product lines cover this market from beginning to end from testing compute devices to complete server trays all the way to robot-assisted operations in AI Data Centers. Looking beyond the AI Data Center segments of the market where Teradyne has high share are poised for a recovery.

Auto/Industrial will have long-term growth tied to the transition to AI, EVs and 800-volt data center power. Mobile is positioned for steep complexity increases as the compute power required to run inference on LLMs is crammed into phones. Physical AI is already expanding the applications of Advanced Robotics, and we believe that trend will continue to strengthen.

The second question is whether we're positioned to gain share in the markets where we play. Again, I think the evidence from 2025 is clear. We are. We have gained share in HBM and DRAM. We have maintained high share in networking -- we have ramped significant new VIP sockets. We have a leadership position in silicon photonics, and we are in play for a share of merchant GPU. We have won new segments and customers in our IST Group in both storage test and System Level Test of compute devices, but Teradyne's exposure to the growing AI data center market extends beyond device test. Our production board test business tests the server trays that devices go into. Our Quantify Photonics instruments test silicon photonics from device to rack.

In alignment with our strategy to go from wafer to data center, last Thursday, Teradyne announced an agreement with MultiLane to form a joint venture. MultiLane is a global leader in High-Speed I/O and data center interconnect test solutions. This joint venture will be called multilane test products and is being formed to serve the growing AI data center demand. Upon the close of this transaction, which we expect in the first half of this year, we will be the majority owner of the JV and Multilane will maintain a minority position. In robotics, we have built a world-class platform for physical AI applications that is being applied in multiple industry verticals, and we have embedded AI capabilities into our AMR products.

Most importantly, we have begun to ramp an important worldwide AI-driven application in e-commerce. So to sum up, Teradyne is positioned to deliver better-than-market growth in markets that are going to be growing robustly over the next few years. We foresee a future where the ATE TAM will be $12 billion to $14 billion, up from about $9 billion in 2025. In that market, our long-term model illustrates our expectation that Teradyne would deliver nearly 2x 2025's revenue and 2.5x the earnings per share.

With that, I'll turn the call over to Michelle Turner, our Chief Financial Officer, and welcome her to her very first Teradyne earnings call. Michelle, over to you.

Earnings call. Michelle, over to you.

Thank you, Greg, and good morning, everyone. I'm thrilled to have joined the Teradyne team and look forward to the value-creating opportunities ahead.

Today, I will cover our fourth quarter and full year 2025 financial results, then I will share our Q1 2026 outlook. And then finally, I will discuss our new target earnings model. Now on to Q4. Fourth quarter sales were $1.83 billion with non-GAAP EPS of $1.80, both above the high end of our guidance range. Fourth quarter sales were the highest revenue quarter of 2025 and our second highest quarter in history, only $3 million below our record during the mobile boom of 2021. Semi Test revenue was $883 million, fueled by AI compute and memory demand. Within Semi Test, SOC revenue was $647 million, up 47% quarter-on-quarter.

And Memory revenue was $206 million, up 61% quarter-on-quarter, marking a record sales quarter for our memory business. The Product Test Group at $110 million grew double digits sequentially and year-on-year, driven by strong defense and aerospace demand. Robotics revenue of $89 million grew for the third consecutive quarter and was up 19% from Q3. In Q4, greater than 5% of our robotics revenue was driven by a large e-commerce customer.

Moving on to bottom line. Non-GAAP gross margins were 57.2%, aligned with our guidance range, driven by Semi Test AI demand strength, offset primarily by lower Product Test Group margins and robotics mix and an inventory write-down on legacy products. Non-GAAP operating expenses were $306 million, and the non-GAAP operating profit rate was 29% in the quarter. Non-GAAP operating profit dollars in the quarter roughly doubled to $314 million in comparison to both prior quarter and prior year.

We generated $219 million in free cash flow and returned $204 million to our shareholders through share repurchases and dividends. Our tax rate for the quarter, excluding discrete items, was 10.6% and 10.3% on a non-GAAP and GAAP basis, respectively. Overall, fourth quarter results were strong across the portfolio.

Now turning to full year results. Our revenue was $3.2 billion, up 13% from prior year. At the beginning of the year, our SoC revenue was equally divided across our major end markets of compute, mobility and auto and industrial. Exiting the year, fueled by strong AI-driven demand, compute is now the largest part of our SoC portfolio, eclipsing our historical stronghold of mobile. From an overall portfolio perspective, Semi Test now represents close to 80% of our enterprise sales, an increase from the low 70s over the last few years. From a customer perspective, I'd like to remind you about a characteristic of our business model.

We typically have a specifying customer who chooses platforms and drives demand and a purchasing customer who actually places the order and receives the equipment. In different cases, the specifying and purchasing customers have more influence in the purchase decision. In 2025, we had 2 greater than 10% specifying customers and 1 greater than 10% purchasing customer.

Gross margin for the year was 58.3%, OpEx was $1.2 billion and operating profit was 22%. Non-GAAP EPS was $3.96. We generated $450 million in free cash flow and returned $785 million or 174% of free cash flow to our shareholders through share repurchases and dividends. We ended 2025 with $448 million of cash and marketable securities. Our tax rate for the full year, excluding discrete items, was 12.8% and 12.6% on a non-GAAP and GAAP basis, respectively.

Now to our outlook for Q1. Since our October call, we've continued to see demand across our group strengthen. Q1 sales are expected to be between $1.15 billion and $1.25 billion, which would be a new quarterly record, driven by all things AI. The midpoint of this revenue range is 11% growth from an already strong Q4 and 75% growth from the same period in 2025.

Non-GAAP EPS is in the range of $1.89 to $2.25 on 158 million diluted shares. From a margin perspective, we expect first quarter gross margins to be in the range of 58.5% to 59.5%, up 180 basis points at the midpoint of the guidance quarter-over-quarter. OpEx is expected to increase 6% from Q4 and run at approximately 26% to 28% of first quarter sales.

The non-GAAP operating profit rate at the midpoint of our first quarter guidance is 32% -- with the strong start to the year, I want to take a minute to talk about our historical sales patterns and how this is a classic example of history is not necessarily indicative of the future. Many of you who have been following us for a while know historically, we've experienced what we call lumpy Q2 or Q3 revenue trends tied to Mobile demand and product life cycles.

From 2020 to 2024, we consistently delivered the majority of our sales in second and third quarter. 2025 broke this pattern. Q4 represented our largest quarter of the year. As our Compute and Memory portfolios continue to grow, our revenue will continue to be lumpy, yet follow a less predictable pattern. While 2025 sales were 40% in the first half and 60% in the second half, based on what we know today, we [indiscernible] the inverse.

Before I walk through our new target earnings model, a few comments on our recently announced MultiLane joint venture. As Greg mentioned, we expect to close the joint venture in Q2 26. For your modeling purposes, the results of this business will be consolidated into the results of our Product Test Group, and our EPS will reflect our share of the results of this business. We will disclose the net income attributable to the noncontrolling interest as a new line item on our income statement. We expect this deal to be accretive in 2026 with a de minimis impact to EPS.

Now moving on to our new target earnings model. Rather than incurring our earnings model to a specific future year as we've done historically, this year, we are framing it around what our P&L looks like at an ATE TAM of $12 billion to $14 billion, which we believe is [indiscernible] evolve within this new term. This approach better reflects the inherent lumpiness both compute and memory demand, where program timing and customer buying patterns can shift revenue across the quarter and year bounders.

So at an ATE TAM of $12 billion to $14 billion, our target model assumes roughly $6 billion of revenue. At this scale, we expect gross margins between 59% and 61%, a point higher at the high end versus our prior model. We anticipate OpEx of 27% to 29% of revenue, reflecting operating leverage and the benefits of scale.

This results in an operating profit of 30% to 34% and non-GAAP EPS of $9.50 to $11. We expect this growth over the midterm to be proportional across each of our groups. From a SemiTest group perspective, we expect to grow our revenue greater than the overall ATE market growth rate, reflecting our expectations of share gains, which is driven by continued strength in AI Compute and Memory as well as anticipated recovery in Auto and Industrial and mobile. Our Mobile assumptions reflect recovery but not a return to the 2021 peak. We also expect growth in IST tied to wins in SLT for compute as well as HDD.

From a Product Test Group perspective, we expect growth across the portfolio tied to Compute, Defense, Photonics, High-Speed Internet Data and Data Centers. From a robotics group perspective, we expect growth tied to physical AI expanding SAM, reducing implementation complexity and continued persistent labor shortages.

Our strategic pivot towards large accounts, along with a sharper focus on e-commerce, logistics, semiconductor and electronic verticals is expected to further support growth. This new target earnings model is reflective of our conviction in the growth potential of the ATE TAM driven by all things AI even at today's unprecedented levels. Moving from a date-driven earnings model to an evergreen one reflects this conviction while also recognizing a lack of precision in terms of which year this comes to fruition. Now turning to capital allocation. Our strategy remains consistent to maintain cash reserves that enable us to run the business and have dry powder for M&A. For reference, from 2015 to 2025, we returned over $5.4 billion to shareholders through share repurchases and dividends, which is roughly 100% of free cash flow.

We will remain opportunistic around value-creating inorganic opportunities as well as share buybacks. So summing up, exiting 2025, we are encouraged by the strength of the business. Our overall company revenues grew 13% year-on-year, and our SoC and memory contributed 17% year-over-year, helping to achieve a 23% increase in our EPS to $3.96. We are making strategic investments to drive competitive advantages and gain market share in the Semi Test and Product test groups.

We remain focused on large accounts and attractive verticals to drive sustainable growth in robotics. We enter 2026 feeling good about the year ahead. With that, I'll turn the call back to the operator for questions. Operator?

[Operator Instructions]

We will take our first question from C.J. Muse with Cantor Fitzgerald.

I guess, wanted to focus near term and then a longer-term question. On the near term, can you kind of help us understand how to better think about calendar '26. I heard you talk about kind of the inverse of that 58%, 42% we saw in calendar '25, but curious how to think about perhaps the overall revenue new growth rate or thinking about June so we can size it. Will you gest above the high end of kind of the revenue target range of 25%? Or any help would be great.

Thanks for the question. This is Michelle. So I'll give some color commentary. I know this is going to be aventuous to everyone who's listening in. So a couple of things I would reference in terms of this year versus past years. One, we are entering the year with a healthy backlog. And so we're excited by that. That's a positive when you think about our positioning for 2026. So we have better fidelity than we did say the same period last year. And then the other thing I would highlight is we typically for those that follow the Teradyne story, you know we talk about having my 13 weeks of demand kind of insights from a forecast perspective. I would say we have better insights this year to first half. And so that's a positive from an overall 2026 perspective. I do want to balance this, however, with kind of what I talked about in my opening remarks and really emphasize the lumpiness of this new sales pattern. So I want to caution us against kind of a linearity trend assumptions with the recognition that that we could see things move between quarters and between years as we recognize some of these ordering patterns in this kind of new AI infrastructure build-out environment.

And C.J., I'll add 1 thought here. This is Greg. Because the run rate that we have in Q1 we have a fair amount of strength in Q1. We don't have great visibility into the second half. So we're a little bit cautious that we don't want people to sort of take that and run with it for the full year. we expect that we're in kind of a 2-, 3-quarter surge that may lead to a shorter period of digestion afterwards.

Great. Very helpful. And then, Greg, longer-term question. implicit in your new target model is a vision for your share of AAT to grow from about 25% to 46%. So would love to hear kind of your high-level thoughts on what the key drivers are behind that?

Right now, in the -- our model, our sort of $12 billion to $14 billion TAM model with us at $6 billion. That actually is moderated from that 46% level just a little bit. And that reflects, in that model, we expect that the compute TAM is going to continue to grow. We're going to be gaining share in the compute space, but we're coming from a much lower share position. So I think we are -- like thinking about it from a long-term model perspective, we expect to gain share in compute. We expect the mobile market to probably get to maybe 1.5x the size that it is now, and we've maintained the share that we have auto and industrial, probably the same kind of proportional gain in terms of the TAM size and we'd maintained. And then in memory, it's going to have incremental growth through this midterm. And right now, we feel like if you look back a few years, there were multiple parts of the memory market where we were not even present.

Now we feel like we are in most of the segments for most of the customers, and so we are in a position to sort of split the share and [ ride ] them. Does that help?

Our next question comes from Timothy Arcuri with UBS.

Greg, I wanted to come at the last question from a different angle. So if I take your new model, it seems -- if I take what you've already said for Robotics, how much it will grow, and I grow systems test at a pretty good clip, product test at a good clip. It seems to me like the SemiTest share only gets back to the like high-series which is really only where it was from 2022 to 2025. So it doesn't really seem to imply that much share gain? I mean it does off of where it was in 2025 because it was sub-30, but it doesn't seem to imply very much share from where it was in 2022 to 2024. So maybe you can provide a little more color, like is that calculation wrong? Is the SemiTest number assumed having share higher than that? Because it seems like it's not that much higher than it's been in the past few years.

I think in order to get to the numbers that you have that you probably have slightly more aggressive growth expectations for the robotics and the product test group. So we're kind of thinking across this midterm that the proportion -- sort of the 80-10-10 proportions are going to be roughly the same. And so -- and by my math, we are in the low 40s for share in ATE. And remember that like the IST stuff is in our revenue, but it's not in the ATE TAM. That's in a separate segment.

Yes. Okay. All right. And then can you break either your -- Michelle, can you break down the SOC TAM in 2025, the $7.2 billion. Can you break it down between compute, mobile, auto? And maybe also, you've talked before about how large the VIP TAM is within compute. Can you give us a sense of how big that was in 2025?

So in 2025, the TAM broke down roughly -- we think it's -- and this is subject to us sort of totaling up the final numbers, which will come in over the next couple of months from third-party sources. But our expectation was that compute was in the neighborhood of $5 billion for the year. Mobility in about $1 billion; auto industrial just under $1 billion and service was in the $700 million range. For memory, overall, we think that the TAM was just under $1.4 billion and $1.2 billion of that was DRAM. The rest was flash.

And Greg, of the compute portion, how much is VIP, sorry?

I think just over 600.

We will move Next with Mehdi Hosseini with SIG.

Just one additional follow-up on the SoC TAM. Greg, you highlighted market share of around 50% for custom ASIC back in 2025. How do you see that evolving, especially given the increased new products coming out? It is my expectation that the concentration is actually going to get broaden out and more custom ASIC coming out? And would that enable you to increase market share above 50%? And I have a follow-up.

Sure. So yes, so in 2025, we think share was roughly 50% of VIP compute. When you look into the future, we expect that, that share split is going to -- it's -- the thing about this space is that no share is safe. We are challenging for sockets that we don't have. We're being challenged for sockets that we do have. And that's true for stuff that is in high volume right now and programs that have yet to ramp.

The thing that I'm a little bit cautious in terms of trying to size the TAM for ASIC programs that are not yet in volume because what tends to happen is that the hyperscalers will benchmark the performance of their own ASICs against what they can get commercially, and they will only take their ASICs to full volume if they see an advantage in like tokens per watt or what other metric they are trying to focus on.

So I think long term, it's likely that we'll be able to maintain 50%, but I am expecting that this is going to be a really noisy number, especially at the quarter-by-quarter level, but even yearly, depending on when different ramps happen, it's going to slosh the share around a fair amount.

Okay. And then for the entire team. As we look at $6 billion near-term revenue target, given your revised ATE market, it was only 6, 9 months ago that we were contemplating if your revenues would increase above a couple of billion. And now there's a new target. And what I wanted to figure out, what the question is, what is the sensitivity to that TE and the $6 billion revenue target?

Is that a baseline assumption? Is that a kind of average of awards and a best case scenario? And any thoughts around how you came up with the ATE and the $6 billion revenue target would be very helpful.

So, I think-- first, I'll say that it's a balanced number that there are potential balloons and anchors around our $6 billion. Probably the most important uncertainty is the speed at which the market grows? Like how long does it take to get to this kind of a TAM size? And I say that because, like I've been in this business for a long time. And I've been in situations where we look into the future and we see like up until the right TAM forcasts and then external conditions change things. So I think it's 1 of the reasons that we wanted to look at an evergreen model is because we can't predict the external factors that are going to drive the TAM.

So that's probably the biggest X factor in all of this. The next is our share in the ATE market. And that is really related to the compute space. We have -- we believe we're positioned to gain share in the Compute space on an incremental basis, but it will take time, but we think that we have a good product and good position with the customers in this space to be able to increase ourselves from a relatively low position.

The next part of this and the thing that gives me a fair amount of confidence around our $6 billion number is the other stuff beyond the core ATE, the Compute space, if you will. I think the Mobile space, we are going to ride whatever the TAM does, and I think that TAM is going to recover on the basis of complexity. We are in a position to gain share in the Industrial and Automotive space, because of the acquisition of the Power group that we got from Infineon last year to cover the wideband gap power market. And our IST business has a much broader customer base to help drive healthy revenue growth through this midterm and then we have this extra large customer in the Robotics space on top of the core business that we think is a catalyst for growth there.

So I think we have a balanced plan going on to the future, and I think that helps to derisk that number.

Our next question comes from Krish Sankar with TD Cowen.

Greg, congrats on the great results and guidance. I know you can't get into specifics. I'm just kind of curious on the GPU test side, which is a growth opportunity for you. What is the realistic market share expectation for this year? And how high can that go over the next 2 years or so? And can it partly into basic market share to? And then I had a quick follow-up.

Okay. So let me give you sort of a quick update of where we are in this project. So we're making great project -- progress, and we expect to achieve production qualification. The call process for these things is very complex. It's very extensive, and the exact date for our release production is tough to pin down, but we're really confident of our success here.

Once we achieve qualification, I believe we'll incrementally gain share for these devices over the course of a couple of years. I just want to be clear that our guidance for Q1 does not include any merchant GPU revenue. We see that as more of a factor in the second half of 2026, and we believe it would be a material amount of revenue for us, but it would not be a heck of a lot of share inside of the account where we want. So with single-digit kind of share numbers to start.

And then over time, what we've seen in other situations where we have a competitive dual-source situation, is that we eventually get to a situation where the vendors are balanced between 30% and 70% share. And that is not saying that like our share top is 30%, but we'll take time to get up to that 30% and then we'll essentially be going head-to-head on a competitive basis around who gets how much?

Got it. Very helpful. And then just a quick follow-up. I understand you don't want to give a full year outlook in first half weighted for this year. Is that because of I'm just curious because given that Mobile is less, I would expect no seasonality anymore. So I'm just curious, why do you think it's still first half weighted besides visibility on [indiscernible].

So I'll give Michelle a chance to comment. I would say that part of it is because major programs that we're a part of are first half loaded, and so the demand that we can map for the full year is definitely more concentrated in the first half than the second half, but there's also an element of, we don't know about the second half that there are a lot of irons in the fire that could result in second half growth, but we can't pin that down enough to sort of give a confident forecast for where the full year would be. I don't know, Michelle, do you want to add anything on there?

I think you summed it up well, Greg. I think the only other thing I would add is in comparison to previous years, coming into '26, we have a really strong backlog, which is giving us better fidelity and insights into the first half.

Our next question comes from Jim Schneider with Goldman Sachs.

Just a bit of a clarification, if I could, and I may have missed it, so I apologize. But if you -- as we think about 2026, can you maybe help give us a sense, given everything you said about the first half and 2-quarter visibility and the potential unknowns in the back half, maybe give us any sense about how we should be thinking about the Q2, relative to Q1 rough numbers? I mean, is flattish, something we should expect as a reasonable jumping up point on a sequential basis and/or can you help us on where you think, roughly, even if it's a large range, we could land in terms of the ATE TAM in 2026?

Yes, so from a Q2 perspective, we're not going to give the second quarter guidance, just going to go back to, we have better visibility within the first half and I would expect '26 to be the inverse of 2025 by roughly 60% of our sales in the first half. We'll give you an update when we get into Q2. Thank you, Jim.

On the ATE TAM for 2026, what we talked about in our prepared remarks was a robust growth from 2025, and the reason that we're using adjectives versus numbers is that it comes down to a really wide range essentially based on the uncertainty in the second half. So like if you wanted to put a big wide range around it, it would be like 20% to 40% growth for the year, but we don't know where in that range it would land.

Understood. That is helpful. And then maybe just as a follow-up, it was referred to before, and I think you talked about -- I think, to paraphrase many ways to get to the target model, assuming that not everything happens perfectly, even if the endpoint is uncertain, but as you think about that model, is that what you think could be a mid-cycle model in a couple or 3 years' time, whereas that would kind of incorporate a lot of cyclical ups and downs once we get past this kind of period of very explosive growth.

Yes. So the -- when we when we were doing a model that was like fixed to a particular year -- we always had all of these caveats around sort of we're trying to average out the cycle, looking at long-term growth trends and all of that was, in the end, not particularly helpful. So what we decided to do was to give people an idea of sort of what Teradyne would look like at $6 billion. And at $6 billion, we think that we need an ATE TAM between $12 billion and $14 billion to be at that kind of revenue level? And then the rest of the business model soREIT of drops down from there in terms of our expectations of the investments we need to make and the kind of margin we get. So I do believe that the model is achievable over the next few years.

But whether it's at the -- whether you interpret a few as a small number or a few as a larger number really depends on how quickly the ATE TAM grows and that has to do with whether the current pace of data center build-out continues kind of at this rate. So the numbers that are out in terms of how much silicon is going into data centers are kind of mind-boggling. It's from '24 to '25, it's like 60% more silicon revenue in data centers. Looking out to 2026, it's way more than 100% year-on-year growth. So whether that can persist from '26 to '27 or if it moderates, that's the thing that's going to determine how fast it takes to get to that $6 billion.

We will move next with Brian Chin with Stifel.

Maybe to start with, Greg, I was wondering if you could outline maybe a few catalysts for GPU share gain over the next few years in terms of Teradyne's platform differentiation, higher device power and complexity and maybe the addition of new test insertions.

Yes. So the addition of new test insertions, I think, is a catalyst for TAM growth more than a catalyst for share growth. So as these new insertions come in, we will have an opportunity to compete for them. And the same thing is as the merchant GPU market has more specialized chiplets per device, I think there's more shots on goal, more higher quality requirements for the test at the chiplet level. So there's a bunch of things that I think are accelerating the compute TAM. Now in terms of compute share, there are a number of things that I think our customers like about our product. The first and most obvious is that they believe that we have a more resilient supply chain that we're able to respond to demand with generally shorter lead times. And that's an important thing when their demands are somewhat unpredictable.

The second is that it's actually a better tester. We have very good reliability in production circumstances. We have good uptime. The OSATs like it a lot. So they are helping to advocate for that as a choice. And we also have a next generation of instruments that is in beta test now, which will significantly increase the amount of power available to the devices and very importantly, the amount of memory for the test programs and the test patterns that these devices are going to need.

The last is that I think our tester has better capabilities to allow these devices to be tested in the same way that they're used in the server in a mission mode. And that requires a pretty sophisticated, almost like building a server into the tester itself. So I think we have some advantages that allow us to achieve higher coverage, essentially moving defect detection as far to the left as possible.

Great. That's really helpful. And then I think in the prepared remarks, you mentioned a new HDD customer. Is that an example of your test platform outperforming their internal tester? I guess, how much growth do you expect from HDD test in '26 and also what revenue base in '25. And then kind of last part of that. More broadly, are there other historical instances of semiconductor logic and DRAM IDMs using captive test platforms? And way at the point there where complexity in semi test really compels those companies also to use external platforms.

Yes. So in HDD of commercial test replacing in-house test -- the right way to think about it is complementing internal tests. I don't think that this is like a way for a customer to affect -- so -- but we're really excited about the change because we've been and we've been working to try and achieve it for a number of years. From a revenue perspective, we don't break out the HDD versus other revenue inside of the IST Group. But I will tell you that like our HDD revenue is going to like double between 2025 and 2026.

Sorry, you had a part B to your question, which is other captive in the rest of the semiconductor ATV space. So right now, -- there are really -- there's 1 big player in SoC, and there's 1 big player in memory that have captive ATE strategies. I think the like long term, I think the memory 1 is probably more persistent. The SoC one, I think, is probably going to change over the next couple of years as the there are a broader range of customers for foundry that want commercial platforms.

We'll walk next with Samik Chatterjee with JPMorgan.

Greg, maybe if I can just change gears here and ask you about the TAM. And in your prepared remarks, I think you did say you're expecting it to be about 1.5x in the current TAM in your target model. I mean, is that sort of all driven by the complexity or are you thinking about some sort of volume tailwinds as well -- and then you did mention near term there being sort of a capital efficiency of customers that may be making you a bit more cautious, if you can explain that as like what you're seeing on that front that we quick follow-up.

Sure. So yes, the thing that we want to try and emphasize is that in like a $12 billion to $14 billion TAM model we are not expecting the mobile TAM to get back to prior peak that it's like a half decent guess is somewhere like halfway between where it is now and the prior peak. So -- and you're asking in terms of like what would drive that. I think that it is primarily around complexity, that's not units. That smartphone units have been sort of hovering in a relatively narrow range there's a potential that if there's a compelling new form factor or really compelling AI-based features that it would drive a higher refresh rate, but that certainly hasn't been the case for the last 4 or 5 years. So we're modeling kind of relatively consistent unit volume but increased complexity across the broad product line.

Now the reason that we are are cautious about that, I think like we're pretty certain that there's going to be a lot of complexity growth and more complexity means more testers are required. However, there is a really large fleet of testers that are installed for mobile, and there are a lot of different parts across a number of different vendors that could use very similar tester configurations. And so by carefully arranging the use of that fleet they can optimize the utilization on a year-round basis, and it can help to moderate the amount of additional capacity that they need to add.

In the old days, like back in 2020, 2021, there was a smaller installed base and there were fewer SKUs that were being tested, -- more of them were being introduced and ramped very quickly. That was the kind of thing that really piled up the demand to drive much higher TAMs.

And maybe just for my follow-up, going back to the AI compute side. I mean you did mention that the VIP ASC sort of what's not launched already in production in the volumes are a bit tough to Quantifi at this point. But in terms of broadening of the customer base, given it's a very concentrated sort of purchasing from a few customers right now, as you look out to the medium term, particularly in terms of your earnings target earnings model, do we see -- do you see a broadening out of the customer base? Does that sort of reduce when you get to that target model does that reduce the lumpiness in the business, just given higher visibility from a broader set of customers?

Yes. So in a $12 billion to $14 billion TAM, our expectation is that we would add additional logos in terms of VIP compute wins. But it's not like it's going to go from a very small number to dozens. It's more like 4 or 5 different programs -- and what I expect to see the steady state in this market is that Teradyne and Advantest are going to be competing on a generational basis for new design wins -- and those decisions are going to be made on the basis of the features of the tester, the reliability of the tester more than sort of incumbency as the thing that drives the selection.

Our next question comes from David Duley with Steelhead.

I guess the first 1 is, I think you mentioned 30% customers. Could you talk about which segments they might be in or how large they might be? I know you probably don't want to give us the names, but if you could give us things that would be great as well.

In the 3, 10% customers, as Michelle said, 2 of them were specifiers, one was a purchasing customer. The specifying customers, 1 was in the Mobile space, 1 was in the Compute space. And the purchasing customer does it all.

Okay. And relative size of how much above 10%, I guess I'm just trying to figure out customer concentration.

Roughly 10%, it's not substantially higher than that.

So each 1 around 10%, is that what you just said? I'm sorry.

Yes. Yes.

Okay. All right. Final question, I guess, is, Greg, I think you kind of mentioned, when you look at all the pieces for 2026, the overall TAM growth, I guess, is going to be around 30% I'm guessing that the SoC TAM goes faster than that and the memory TAM goes slower than that in 2026. If you could just comment on roughly the growth in each piece. And then if you said you're going to gain share in 2026, and that means, obviously, you're going to grow faster than 30%? Is that a fair assumption?

No. The -- yes, the -- like in a range between 20 and 40, you may arithmetically put that at the mean of 30. We are not trying to communicate that at all. We are trying to communicate that we have -- we don't have sufficient visibility into the second half to give a good TAM estimate for 2026. Your assumption around SoC growing faster in memory going slower, I think it's fair. I think that we are expecting that kind of a market where the Compute TAM is already big, and it's going to grow a lot.

The Memory TAM is going to grow more incrementally. The -- we believe that we are positioned for share gain, and that really depends to a certain extent around whether -- like which segments of the TAM grow the most. So even if we gain share in compute, since our share position in compute is relatively lower if the compute TAM grows a ton, then that could be dilutive for our overall share position even though we're getting better in every segment that we serve. So that's the reason that I want to be cautious about that.

And we have time for one more question. We will move next with Vedvati Shrotre with Evercore.

One clarification I had is so the GPU win, the GPU merchant win -- does that, in any way, dictate your second half versus first half dynamic? And then even in the new target model, are you assuming contributions from CP wins?

So yes, a significant ramp associated with merchant GPU would have an impact in the second half. I'm not sure I caught the second part of your question.

Is that a part of your new target model as well?

Oh, yes, yes. So -- and it's -- but the thing I want to emphasize is -- as I said, the merchant like share in merchant GPU is going to be an incremental gain over years -- and so it is a part of that $6 billion model, but we don't assume a radically high share in the merchant GPU space.

Understood. And then the second question I had was on the robotics, you have the large e-commerce program starting to ramp. So does that mean that you -- is there a possibility that your revenues grow like the robotics piece grows to higher than the breakeven revenues that you have for that business?

Yes. So we're aiming at breakeven for Robotics this year -- and we expect -- so just in terms of the large e-commerce customer, we think that, that revenue is kind of going to triple ish between 2025 and 2026 and then grow substantially post '26 as the deployments go to a larger number of facilities. So that's a pretty good tailwind. And it -- so I think we're looking to have that business at breakeven in '26 and then contributing positively beyond.

And this concludes our Q&A session as well as the Teradyne Fourth Quarter and Full Year 2025 Earnings Call and Webcast. You may disconnect your line at this time. Have a wonderful day.

Okay. We're going to get started. I'm Tim Arcuri. I'm the semi and semi equipment analyst at UBS. And very pleased to have Teradyne here later in the afternoon and very pleased to have Greg Smith, who is the President and the CEO. So thanks, Greg.

Thanks.

Great. Well, let's just start from a macro perspective. And I wanted to ask you about the big picture opportunity in Semi Test. And we hear all these massive deals. We hear OpenAI signing all these deals. We hear how fast data center is growing. We see these great numbers from NVIDIA. And yet the semiconductor test TAM, the SOC TAM is still going to be $6 billion to $6.5 billion this year. It seems very low relative to how much semiconductor revenue is growing. So I guess my question is sort of how do you think about the opportunity and like how much the TAM can grow relative to how we're hearing all these semiconductor revenue growth opportunities?

So I think we should probably talk about things we know and things we don't know. So things that we know is that there is a certain amount of committed data center capacity that's going in over the next few years. And that is going to drive significant silicon growth year-on-year. The other thing that we know is that the silicon complexity is growing at a much faster rate in the data center than it is -- than it has in other parts of the market.

So -- and then the last thing that we know is that the time frame between like a deal being announced and it actually having an impact in the market is a significant period of time. So people are very excited about new deals being inked around new hyperscaler ASIC programs, next-generation ASIC programs. Those are all things that are going to have an impact in '27 and '28. That's when it would really impact the test market. So I think you're talking about like, oh, the TAM is only $6.5 billion. I'm kind of psyched that it's $6.5 billion now, and there is room to go up from here that I think all of the equations are leading us towards this kind of sustained growth.

And the last thing that's really exciting about the data center opportunity is the pursuit of high-performance compute, sort of tokens per watt or whatever metric you want to use to measure sort of how much revenue you can get out of the data center, that is such a potent draw that people are pulling advanced technologies into the process very, very quickly. And it is also a situation where you can end up with technologies coming into production that are at much lower yields than they have been in other device technologies.

So you're in this situation where people are spending a lot of money to build out this capacity, but they're also really trying to see what they can do to maximize the output that they have of good parts. That's kind of the perfect recipe for higher spend on test because we can help people sort out the good ones from the bad ones when processes are immature.

Got it. And so I hear quite often, I mean, obviously, we know it's a duopoly. And the theory is that the duopoly has favored your competitor because they're more exposed to memory and compute. You're more exposed to mobile. I think that's really changing, but I hear that a lot. So can you just talk about like what does that theory miss?

Well, I think the theory doesn't miss much from a historical perspective. And if you look in the rearview mirror, incumbency was everything that if you had a possession of a customer, the amount of energy that it would take to enter or disrupt that customer was sort of beyond what any customer would invest in. But now the difference is that for the main providers of compute for data centers, it's such a major part of their overall corporate strategy, and that's such a major part of the overall semiconductor economy that supply assurance now becomes a much more important factor than when you're talking about gaming chips.

If you're building gaming chips and that's a small single-digit percentage of total TSMC capacity, that's sort of one set of circumstances around your overall test strategy. If like economic growth in the entire planet is dependent on your success in AI, that's a whole different story in terms of what you need in terms of supply assurance. So I think we've entered this era that for the large players in the data center that supply chain resilience is increasingly important, and Teradyne is positioned to benefit from that.

I think that Advantest has done a great job aligning to the compute market for a long time, and they've reaped the benefit of that in the short term. But I think this importance of the compute market overall is such that these customers are going to be seeking alternative solutions.

So you mentioned the risk mitigation and dual sourcing. And obviously, the big GPU customer is trying to do that across everything it buys. Do you think from a custom ASIC perspective, it's historically like TPU, for example, those decisions have been made via the custom ASIC producer. You, however, are going more to the direct customer and saying, listen, you need to get more involved in this, and we have a better tester. So is the procurement decision on the ASIC side also changing as well?

I think over time, it is migrating more towards the hyperscaler, this vertically integrated platform that's driving the design and the power of the aggregator is reducing. That's certainly -- like I don't know if it's a trend, but that's the evidence that we've seen in the large-scale deployments that we've seen that -- and by the way, it's not just around tester selection. What you'll see is that the hyperscaler, even if they have worked with a particular aggregator for multiple generations of parts, now they are looking to other aggregators as a way for them to provide assurance.

So we believe that the ultimate decision-making power will be migrating to these hyperscalers. And so we think that's the most important place to prove our advantages. And that doesn't mean that we aren't focused on spending time with those aggregators and to try and show them that we have a great way to ensure that they have the capacity that they need, but I think it's a more powerful argument to the hyperscalers because they're the ones that are making these billion dollar commits on data centers that they need to get up.

So you think there are these pools of moats that they've had in the past where your competitor has had a 100% share for the big GPU supplier. They've had a 100% share for TPU. The bulk of what's growing, they've had 100% share. But would you say, I mean, obviously, on the GPU side, they are trying to diversify. But also for TPU, I mean, do you think all these customers are going to want dual source and so you're the share gainer?

So I think the difference between the merchant GPU opportunity that we have and these other things that you're talking about is kind of how far down the road they are. At the end of the day, the same challenge that led a merchant GPU manufacturer to invite us to compete, which is wanting to be able to be sure that they could get the test capacity when they need the test capacity. I think that becomes even more important if you are not the most important customer for our competitor.

So if you are going to be the third or fourth priority, then you always have to be worried about who's going to get in line in front of you. So I think there's one reason why a leading manufacturer of GPUs would do it. There's another reason why a VIP might do it. But in general, it's all around making sure that you have multiple shots on goal to get the capacity that you need.

Great. So let's talk about that GPU opportunity you have. It seems like you're in the correlation phase now. From what I know about that, that's a win to even get to that point. And I would think that this would be a first for your program and maybe a little less volume at first, but then you sort of get your foot in the door and the bigger data center stuff comes in the not-too-distant future. Is that fair?

Yes. So I think to like lay it out, we've been working on this sort of qualification project for a little bit less than the last 6 months. And that project has 3 phases. First phase is conversion and offline development. The second phase is debug. The third phase is correlation and qualification. And we are late in that third phase now. And to your point, nothing terrible has happened so far. Like it's been -- we've made good progress. We've been on schedule. We've had a great level of engagement with the customer. So we expect to succeed.

We are unable to predict exactly the timing of when that will occur because the qualification process is iterative. You're trying to prove equivalence between 2 solutions. And if there is an equivalence, it could be a problem with our solution. It could be a problem with the incumbent solution. And frankly, it could be something that doesn't matter. And so each one of those has to be investigated as an engineering challenge and resolved. And once they're resolved, then the platform is qualified.

What happens beyond that point is an interesting like there's a sort of a long-range set of events that will occur. The first is, as soon as we're qualified on a part, we would be able to capture incremental capacity against that specific device. We could also potentially, if there is a need for capacity on the incumbent platform on another part, we could backfill against that platform so that they could move capacity over. But it's going to be a small incremental start because we just have this one part.

The other thing that happens when you're qualified is they give you more to do. So we'll begin other projects to do more conversions of parts that are much earlier in their life cycle. And those would be the things that could ramp to more significant volume, and that's probably back half of '26.

Longer term than that, the vision that we have is to help them get to what we're calling a tester-agnostic development flow. And that would mean that they're able to flick a switch and decide whether they put it on one platform or the other or even both and then plan capacity against that very late in the process. And we really like that outcome because then, we're competing on a level playing field. It's a matter of differentiating on tester performance and reliability, customer service, time to market, yield, things where in other head-to-head competitions, we felt like we've done quite well.

Well, they're $2 billion of the $6 billion to $6.5 billion SOC TAM this year. So I mean even if you got 25% share, that's a big number for you. So if I told you that in 2027 -- let's say, just to pick a year, if I told you that you're going to be 25% share of whatever their share of the TAM is in that year, would that be a disappointing number to you? Would you be okay with that? Or would you say, oh, I could do better than that?

So I -- the two adjectives that you should probably take away is inevitable and incremental, right, that I think we are going to gain share, but I think gaining share is going to take time because there is a broad range of parts, there is a broad ecosystem to bring those parts to market on the incumbent platform, and we are building that up. So if you were to tell me in 2025, I had 25% share of that, I'd be delighted.

I think that ultimately, long term, what we've seen in other customers that do dual sourcing is that your share ends up in a 30% to 70% range based on differentiation and customer satisfaction. It's probably going to take us 3 or 4 years to get into that range to be competing on an even keel, and 2027 is still on the earlier side of that.

Got it. Okay. Let's talk about another growth driver, which is Apple business for you. And how much visibility do you have there? Typically, not much. They don't give you very much visibility. But the drivers are units, transistor density and test times. And it seems like -- and not just for that customer, but generally, it seems like test times are going up because some of these wafer-level packages. Finally, I think there's more aggressive shrink going on in mobile. So it seems like mobile could be a little bit better outlook for you next year than it's been the past couple of years. And would you agree with that? And how much -- does like a wafer-level package? For example, how much does that increase test time?

Yes. So I'll comment more sort of generally about the mobile ecosystem than around any specific customer. But I agree with your assessment that we're positioned for a more significant complexity increase next year, and that's both through adoption of 2-nanometer process, enabling more functions in the die, especially more AI processing tops in the die and then also a change in the memory -- sort of the packaging technology around memory for at least the premium tier.

And the interesting thing about the change in memory technology is the reason that, that is being done is to make the bus width bigger between the processor and the memory to open up memory bandwidth. Every single connection between the mobile processing die and the memory actually requires a tester resource to be connected to it. And we have this large installed base to test mobile processors.

The fact that each one of those processors is going to consume more tester resources means that you'd be able to test fewer of them at the same time. And it's not like a huge difference, but even an incremental decrease like from 6 to 5 at the same time is a much larger jump in complexity than even the increase in transistors going from 3-nanometer to 2-nanometer. So that's sort of two tailwinds.

The things that we don't know is we don't know how pervasive this technology will be across a mobile phone product line. So if it's just at the super premium tier and very low volume, then that's not going to be enough to move the needle. If it's across a broader range of products, then it could be a very big impact.

And then the other thing that has been -- like forecasting has been sort of historically pretty bad is around unit volume. And the hard part of the unit volume for any of the non-Chinese smartphone makers is predicting what their share in China is going to end up being. So I think that that's a potential anchor against it. But when you add it all together, I agree we're more optimistic about mobile in 2026 than we have been for the past couple of years. But it's kind of like we're a little afraid to dream the dream because there are these factors that we can't really measure.

Great. And then let me ask you about the third, at least, I mean, there's more than just 3, but the third thing that I'm focused on is your opportunity in cobots. And you are building a big manufacturing facility in the U.S. to support a single customer. You -- there's been some blogs. You've -- you're being qualified there. I mean they have a lot of warehouses that could be quite big depending on how quickly it gets adopted. So can you talk about that and sort of just the general outlook for your robotics business?

So first of all, the manufacturing facility that we are going to be building out in 2026 for cobots in the U.S., that isn't -- like that's certainly going to help us serve this new major customer, but there are many large customers in North America that want to procure from U.S. manufacturing.

So there's a more general trend towards reshoring and procuring equipment that has been manufactured in the U.S. And so we wanted to do this not only for the major customer, for other big customers and also because there's significant tariffs against products that are being imported into the U.S. So there's a bunch of reasons why it made sense for us to do our next capacity expansion for cobots in the U.S. versus in Europe where we do our other manufacturing.

Now against the large opportunity, the large customer that we have, it's in the e-commerce space. I agree with you that there is a large opportunity set that there are tons of distribution centers and warehouses that are slated to be automated through this technology. The thing that is uncertain in my mind is the rate at which that automation will be installed.

In the past, most of the automation projects that this customer have done have been associated with like greenfield build-out. So build a new distribution center and introduce a new kind of automation with it. This is a bit of a new thing for them where they're taking an existing process in the distribution center and changing it from a manual process to an automated process.

They've planned an incremental path to get that done. And it sort of the first wave happens in 2026. And then subsequent waves in North America and Europe and beyond are happening in '27 and '28. So we think that this has legs to grow in '26 and '27 and beyond.

So if I took those 3 opportunities, the mobile opportunity, the robotics opportunity and the GPU opportunity. And I said over the summation of '26 and '27 -- let's not take 1 year or the other. Over that period, which one of those 3 would you say would be the largest?

And what are my choices again? We got GPU...

GPU, robotics and the mobile opportunity.

Robotics, mobile. So GPU.

GPU.

GPU. Yes. Because in terms of percentage above baseline, the biggest opportunity is probably in robotics, but that is coming from a smaller base. And also in terms of an impact to earnings, our semiconductor business is much more efficient at turning revenue into [ EPS ] than our robotics business. So there's a lot of room to run in terms of robotics before it's going to have a significant impact on results.

The other thing is that GPU, the sort of the compute TAM is so huge that -- like a, you were saying just one customer is kind of $2 billion this year. If that one customer grows by 20%, that's like bigger than our entire robotics business. So a relatively small change or an incremental share gain in compute will have a really disproportionate positive effect on our results.

Makes sense. Great. Let's -- we haven't talked about memory yet. So historically, you've not had much exposure for wafer test and DRAM. But now we have these new test insertions like stack test for HBM. And the memory market seems to be really changing. So can you talk about how the market is changing? And maybe you can extend the answer to include HBM4E, you've taken a little bit of a different approach where you can reuse your tester and your competitor, you have to buy a new tester. So that's allowed you to take some share. So can you just talk about all that?

Sure. So if you look into history, we have tended to focus our investments towards the final test markets. And the reason that we did that is -- the way that the world had worked historically was at the wafer level, it was kind of a commoditized test that you were just validating that the core of the memory worked. There wasn't a lot of ability to differentiate on the architecture of the tester or the performance of the tester. You were kind of just providing a commodity.

In the final test Yield was dependent on signal integrity, throughput was dependent on the efficiency of the architecture. So we were able to differentiate. So our investment priority was against the final packaged parts. HBM changed that because now there is actually a performance-oriented insertion at the wafer level. And that was the reason that we developed the Magnum 7H to participate in that performance test. And that's been the part of the market that has been driving most of the growth in memory.

And over the past -- from the middle of 2024 to the end of 2025, we've gone from like 0% share of HBM performance test at the wafer level to like 50-50. And so that's been a great story of that we know how to do performance testing. We applied that at the wafer level, and we were able to achieve a lot of share gain.

Looking into the future, as HBM units expand, I would expect that, that is going to be a growing market and something that we would expect to be able to compete quite well against Advantest, not only because of the asset life being able to be used for multiple generations, but also, we also have -- it's more scalable, and we believe it's more performant in terms of delivering yield.

And just from a memory TAM point of view, how do you assess sort of the puts and takes on the memory TAM next year? I mean it's been great the past few years, but could we see some digestion next year?

So I think 2025 was a digestive year that a lot of HBM capacity was put in place in 2024 that was occupied in 2025. So I think when you look at the TAM for memory going forward, there are more puts than takes that I think the HBM market is going to be increasing year-on-year.

And right now, you have a memory market that's like 70% to 80% DRAM and 20% to 30% flash. And there's certainly a tightening in the flash market around SSD that could result in additional flash capacity coming online, and that would need test as well. So I think from a baseline perspective, the core of growth in next year is really around HBM and DRAM, DDR, LPDDR, but there is an upside opportunity against flash depending on how desperate people get for SSDs in data centers.

Great. I wanted to ask you about the financial model. You're going to give a new model next quarter. Your typical pattern would be to keep the same 2028 base year and to just refine the $4.5 billion to $5.5 billion revenue and $7 to $9.50 EPS range to just tweak that range. Is that what we should expect? Or do you want to take a step back and take a different approach?

So I think we'll give you the full story in January. The things that we're wrestling with as we're trying to do this is our practices around providing a business model were really formed when the semiconductor market was far better behaved than it is right now that if you look at that period from like 2010 out through 2019, it was just like there was -- there were little bumps along the way, but it was pretty smooth growth. And that -- and the other thing is that predictability from quarter-to-quarter was actually pretty strong because the market was mobile dominated and that was following a calendar period.

What we're facing right now and what we're wrestling with is that the revenue is much lumpier. There are a few incredibly important customers. They are putting in capacity in huge tranches and their capacity needs do not follow a regular calendar. They follow the sort of the device maturity and release schedule. So while we have a pretty good idea of how efficient a business model we would have at different revenues, it's much more difficult for us to predict how quickly we would achieve a particular revenue point. So I think we'll probably be talking in terms of expected aggregate growth rates. We will talk a lot about inherent uncertainty in the market and try to make sure that people understand just how resilient our business model is across a range of revenue outcomes.

Got it. And maybe I'll ask about Technoprobe. I know you've been through -- you took the position there. How much of a differentiator has that been so far in winning business with these VIPs?

So one of the parts of this -- the deal that we had with TPI is our -- we had an internal group that developed device interfaces. And we actually sold that group to TPI because the trend in the market was to integrate the device interface and the probe technology. And that has been great. They have been a fantastic partner for business acquisition and to provide these device interfaces going forward.

There's another part of that partnership, which was really around the direction that the market was going. And the direction that the market was going was around HBM performance test and around hybrid bonding and very dense connections and things that we thought would be opening up opportunities for us because we had a more scalable test platform. The work on those projects is ongoing, but it has yet to have an impact on share acquisition.

So what we've gotten so far, we've kind of gotten the old-fashioned way from demonstrating the superiority of our tester from a reliability and availability and performance perspective. This idea of like 1 plus 1 equals 3 in terms of a partnership, those are projects that are ongoing, and we would expect to start showing some benefits in '26.

'26. Okay. And then maybe just last question. So the guidance for Q4 was obviously much better. It sounds like you think that what's strong in Q4 will persist into Q1. And I'm not asking you to guide next year, but it seems like next year, maybe the sort of normal pattern of June being the strong quarter, it might be a bit different next year than it's been in the past.

Yes. Like I feel confident saying it's going to be different than the past. I'm not as confident saying what it will be. So I am -- like at this point, I'm confident that 2026 is going to be a good year. I believe that we are seeing demand solidifying throughout the year, but there is so much lumpiness in the demand. We were -- in the -- our last earnings call, we talked about uncertainty between Q4 and Q1 and even between Q1 and Q2. That kind of capacity on the edge is something that's going to make sort of percentage quarter-on-quarter comparisons really tough to do.

Got it. Well, thank you for the time, Greg.

Thank you.

Appreciate it.

Ladies and gentlemen, good afternoon, and welcome to the Teradyne Third Quarter 2025 Earnings Conference Call. [Operator Instructions] As a reminder, today's call is being recorded.

I would now like to turn the call over to Amy McAndrews, VP of Corporate Relations for Teradyne. Please go ahead.

Thank you, operator. Good morning, everyone, and welcome to our discussion of Teradyne's most recent financial results. I'm joined this morning by our CEO, Greg Smith; and our CFO, Sanjay Mehta. Following our opening remarks, we'll provide details of our performance for the third quarter of 2025 and our outlook for the fourth quarter of 2025. The press release containing our third quarter results was issued last evening. The slides as well as a copy of this earnings script are available on the Investor page of the Teradyne website. Replays of this call will be available via the same page after the call ends.

The matters that we discuss today will include forward-looking statements that involve risks that could cause Teradyne's results to differ materially from management's current expectations. We caution listeners not to place undue reliance on any forward-looking statements included in this presentation. We encourage you to review the safe harbor statement contained in the slides accompanying this presentation as well as the risk factors described in our annual report on Form 10-K for the fiscal year ended December 31, 2024, on file with the SEC. Additionally, these forward-looking statements are made only as of today, and we do not undertake any obligation to update forward-looking statements to reflect subsequent events or circumstances, except to the extent required by law.

During today's call, we will refer to non-GAAP financial measures. We have posted additional information concerning these non-GAAP financial measures, including reconciliation to the most directly comparable GAAP financial measures were available on the Investor page of our website. Looking ahead between now and our next earnings call, Teradyne expects to participate in the UBS Technology Investor Conference. Our quiet period will begin at the close of business on December 24, 2025. Following Greg and Sanjay's comments this morning, we'll open up the call for questions. This call is scheduled for 1 hour.

Greg?

Thanks, Amy. Good morning, everyone, and thanks for joining us. Today, I'll discuss our third quarter results, talk a bit about what is driving the business in Q4 and provide a general update on conditions across our businesses. Sanjay will then provide more detail on our third quarter results and fourth quarter guidance.

As you saw in the earnings release, we grew sequential revenue 18% and non-GAAP EPS by 49% in the third quarter. This growth was driven by AI demand in semiconductor tests. Our other test businesses delivered on plan in the quarter. In Robotics, we continue a slow crawl up from our Q1 revenue trough in a challenging environment. The huge investments in cloud AI build-out drove our Q3 performance to the high end of our guidance range as our customers ramped production of a wide range of AI accelerator, networking, memory and power devices. An example of this AI strength is in compute, where our view of the second half of 2025 revenue is more than 50% higher than our expectations just 3 months ago. Some of this increase comes from us responding to customer pulling requests and some is demand increases. As design, process and packaging technologies for AI compute rapidly advanced, we expect that our growth will continue. Our UltraFLEXplus system has been architected from the ground up for high-performance processors and networking devices, which have demanding power, pin count and test data requirements. As AI devices become more complex, the UltraFLEXplus architectural advantages become more valuable to potential customers by enabling fast test development times and high-efficiency volume production.

Our focused investment in R&D is also yielding new differentiated capabilities for compute tests, some of which have been already been announced in Q3. In memory, our Q3 memory test sales more than doubled from Q2 to $128 million, with the majority of those shipments supporting AI applications. In Q3, 75% of our memory revenue was driven by DRAM, nearly all of it from final test of DRAM and HBM performance test. 25% of revenue was from flash mainly for cloud SSD, another segment being driven by AI data centers. Our Magnum [ 7H ] product is differentiated in HBM performance test because it is a multi-generational product. It can cover the test needs of HBM3E and HBM4, and it provides upgrade headroom for HBM4E and HBM5. The Magnum 7H also supports HBM singulated stack performance test. In Q2, we won a design in for this insertion. And in Q3, we began volume shipments. So at this point, Teradyne participates in all major test insertions for HBM, memory die wafer sort, post stack wafer test and singulated stack test. Our results in memory test this year are especially satisfying in light of the composition and size of the memory [ TAM ] for 2025. Our best guess is that the total memory TAM for 2025 will be down low double digits, and the weakest part of this market is flash, our traditional strongest segment. Despite this, we expect our memory revenue will sustain at 2024 levels.

AI-driven applications for power ICs were a bright spot in the auto industrial market segment. The Eagle Test platform has a leading position in the test of high-performance power conversion devices for data center applications. Volumes of these devices are forecast to grow over 50% between now and 2027. We expect the demand for VIP compute and networking to continue to grow significantly, and we have been investing R&D, applications, sales, support and manufacturing capacity for this expansion. This includes investments to win new VIP and merchant GPU customers. We're making good progress on new design and opportunities and are cautiously optimistic about our potential success. But I would like to make it clear that our Q3 results and our Q4 guidance do not include any revenue from these types of new opportunities. For the deployed fleet of UltraFLEX and UltraFLEXplus testers, we see higher utilization and fewer system upgrades than in past quarters, which we believe means that customers are exhausting their inventory of underutilized systems. As a result, we now expect a more direct connection between inflection in end market demand and new system sales.

Looking beyond AI and Semi Test, conditions in mobile and auto industrial remained somewhat weak. Now in our Integrated Systems division, our Q3 shipments were above plan as SLT customers accelerated deliveries for mobile processors and compute applications. We also saw increases in orders for both HDD and SLT systems. Now recall, lead times are generally measured in quarters for this business, so most of that order strength will translate into revenue in 2026 and beyond.

In robotics, we are growing slowly from our trough quarter in Q1 2025. If you go down 1 level of detail, we continue to see persistent weakness in our core indirect distribution channel as we expand our large customer and OEM channels. An important element of our robotics strategy is to establish UR cobots as the preferred platform for AI-driven work cell applications and to deliver superior performance for our AMRs by leveraging AI features. In the third quarter, over 8% of robotics sales were for AI-related products, up from 6% in Q2.

Another element of our robotics strategy is to deliver value-added service to our installed base of over 100,000 robots. Service represented 14% of sales in Q3, up from 12% in Q2. As we noted in our July call, our Semi Test business has evolved to where the largest demand driver is AI data center investments rather than consumer end markets. We have aligned our R&D and go-to-market investments to capture the tremendous opportunities in test driven by this AI related demand. Our investments are focused on extending our product performance advantages with innovative R&D, while also expanding our engineering teams to help customers develop and ramp production of these fantastically complex devices on Teradyne platforms. Sanjay will describe how these investments translate into OpEx, but as we're seeing, the returns are well worth the investment.

Looking at Q4, we expect AI-related demand for compute, networking and memory to be the primary engine of our growth, which reflects both industry trends and the result of our investments to align with those trends. Looking to the future, the long-term themes that we've highlighted in the past, AI, verticalization and electrification remain firmly intact. As we enter 2026, we expect AI and verticalization will be the primary growth drivers. We've said before that the AI market is both highly concentrated and highly dynamic. The timing of any one project can affect the delivery schedule for hundreds of testers. This can swing quarterly results significantly.

So with that understanding, let me offer a few high-level comments about how we're looking at 2026. At the company level, 2026 looks stronger today than it did 6 months ago, and all indications suggest solid growth from 2025. We anticipate that business conditions for mobile, auto industrial and robotics will improve, but the timing and the intensity of that recovery is uncertain. But the real story in 2026 is AI and the investments that we have made to develop differentiated solutions in that space will drive our growth plan. I'd like to share a few specific examples.

Massive investments in building data centers are translating into strong demand for UltraFLEXplus in VIP, compute, merchant compute and networking. In the memory market, AI will drive growth in HPM, DRAM and flash for SSD applications served by Magnum. Accelerated big growth in HDD is driving the demand for more HDD test. The deployment of AI-capable processors for mobile, client computing and cloud AI is driving the demand for more system-level test. We plan to give you a more detailed view as part of our model update in the January call.

Now before I hand the call over to Sanjay, I would like to say a few words about the CFO transition that we announced last night. Michelle Turner will be our Chief Financial Officer effective November 3, 2025. She brings 30 years of financial and strategic leadership experience in the technology and manufacturing sectors, and she has a strong track record of driving growth, disciplined capital allocation and operational efficiency. She is looking forward to getting to know all of you in the upcoming quarter. I'm excited to welcome Michelle to the Teradyne team. Now Sanjay has been Teradyne's CFO since 2019, and he has offered to stay on as an executive adviser to operations as we expand capacity in 2026. I want to thank Sanjay for his excellent leadership and contributions over the past 6 years, and I'm grateful that we will have the benefit of this guidance.

With that, I'll turn the call over to Sanjay.

Thank you, Greg. Good morning, everyone. Today, I'll cover the financial summary of Q3 and provide our Q4 outlook. Now to Q3. Third quarter sales were $769 million and non-GAAP EPS was $0.85, both near the high end of our guidance ranges. Non-GAAP gross margin was 58.5%, above our guidance range due to favorable mix. Non-GAAP operating expenses were $293 million, up sequentially and year-over-year on higher R&D, sales and marketing investments tied to AI as well as increases in our variable compensation. Non-GAAP operating profit was 20.4%.

Turning to our revenue breakdown in Q3. Semi Test revenue for the quarter was $606 million, with SOC revenue contributing $440 million, which was up 11% sequentially and 12% year-over-year. Memory revenue was $128 million, up 110% sequentially and down 15% year-over-year. Strength in [ SLC ] was driven by AI compute and AI-related power test. Memory revenue more than doubled from Q2 on HBM and AI-related LPDDR demand. IST revenue was $38 million, up 9% sequentially, 46% year-over-year driven by strength in SLT shipments. In product test, Q3 revenue was $88 million, up 4% sequentially and 10% year-over-year, driven by growth in defense and aerospace.

Now to robotics. Revenue was $75 million, flat quarter-on-quarter and down year-over-year. In the quarter, UR contributed $62 million and MiR contributed $13 million of revenue. As we noted in July, volume shipments to our large e-commerce customers are not expected to have a material impact on robotics revenue in 2025.

Some other financial information in Q3. We had 2 customers that directly or indirectly, which drove more than 10% of our revenue in the third quarter. The tax rate, excluding discrete items for the quarter was 16% on a GAAP and non-GAAP basis. Our free cash flow was $2 million. Our net income was offset by our net working capital increases tied to accounts receivable and inventory, which reduced our free cash flow. Receivables growth was tied to increased sales, which were weighted to the second half of the quarter. Inventory growth was tied to the ramp in compute and memory driven by upcoming AI demand. CapEx of $47 million was reasonably consistent with Q2. We repurchased $244 million of shares in the quarter and paid $19 million in dividends. Through the end of the third quarter, we've returned $575 million or approximately 2.5x our free cash flow through dividends and buybacks to shareholders during the year. We ended the quarter with $427 million in cash and marketable securities.

Now a little more detail on OpEx and our balance sheet strategy to help you with [ your modeling ]. In the second half of 2025, we're continuing to lean into R&D and go-to-market investments for AI opportunities that we expect will drive revenue in 2026 and beyond. OpEx in the second half of 2025 is also increasing tied to our variable compensation linked to increasing financial performance. At the midpoint of our Q4 guidance, we'll have full year revenue growth of 9% and OpEx growth of 7%. Long term, we target OpEx growth at approximately half the rate of our revenue growth. In 2026 and longer term, as AI revenue blossomed, we expect to meet our OpEx target.

Regarding the balance sheet, we expect to keep our cash and marketable securities at roughly $400 million while also continuing our balanced capital allocation strategy. In 2025, we saw an opportunity to accelerate buybacks in the short term to further enable shareholder value. At the operational level, we expect to exercise our credit lines more frequently as we did in Q3 and expect to in Q4. From a modeling perspective, this means the interest and other line of the P&L will reflect higher interest expense. You should expect to see a couple of million dollars of net interest expense per quarter while we utilize our revolver.

Now turning to our outlook for Q4. Before discussing the details of Q4 guidance, I'd like to remind you of some of the commentary from our July call. Specifically, we noted that we had large projects expected to ramp, which straddled Q3, Q4 or Q4, Q1. As we move through the second half of 2025, we saw projects accelerate into Q3 and are now seeing projects accelerate into Q4. These projects are [ AI drop ]. In Q3, we were able to meet early ramp demands. In Q4, we are seeing demand ramp significantly. We continue to expedite our supply chain, and we are accelerating production capacity growth at factories in multiple geographies to meet the demand.

Now the details. Q4 sales are expected to be between $920 million and $1 billion. Fourth quarter gross margins are estimated at 57% to 58%. This includes some onetime supply costs in the quarter to meet accelerated demand.

Turning to OpEx. Q4 OpEx is expected to run at 31% to 33% of fourth quarter sales. The non-GAAP operating profit rate at the midpoint of our fourth quarter guidance is 25.5%. The Q4 GAAP and non-GAAP tax rate is expected to be 14.5%. Q4 non-GAAP EPS is expected to be in the range of $1.20 to $1.46 on 157 million diluted shares. GAAP EPS is expected to be in the range of $1.12 to $1.39.

Summing up on Q3 results and Q4 guidance. AI is growing across the economy, driving exceptionally strong semiconductor test demand in the second half of 2025. This is evident in our Q3 sales, profit performance and our outlook for Q4. The acceleration of test demand in Q4 reflects customers' drive to pull AI projects in from Q1. We're optimistic about the AI-related market 2026, but we also know shipments can be lumpy.

Now to my final remarks. After 6-plus years at Teradyne, it's clear that Teradyne is well positioned for significant growth over the midterm. Many environmental challenges have occurred during my tenure such as significant government regulations, COVID, tariffs, CEO transition, along with the strategic pivot of investments to AI in 2022. Through all of these opportunities, we have strengthened the company's infrastructure and processes. Our operational resilience is significantly stronger as we have derisked our supply chain, started the journey of multiple factories and multiple geographies to enable significant growth rooted in AI, strong management leads our diversified portfolio enabled through our variable business model, which has consistently delivered tremendous free cash flow through all of the changes and volatility we've experienced. Our balance sheet is strong with firepower to enable strategic investments, continue to deliver a balanced capital allocation and strong returns for our shareholders. I've had the opportunity to make [ New England ] my home and built many lasting relationships here internally and externally. I've enjoyed working with our shareholders and all of you in the investment community.

With that, I'll turn the call back to the operator to open up the line for questions and soon hand the keys over to Michelle. Operator?

We will now be taking questions from Teradyne's research analysts. [Operator Instructions] We'll take our first question from CJ Muse with Cantor Fitzgerald.

Sanjay, big congrats to you. I guess when you look at -- so short-term question, long-term question. So short term, roughly $150 million upside versus consensus for December. And I would be curious if you could kind of share how much of that upside is versus what you thought maybe 3 months ago is driven by HBM, VIP, networking, SLT or perhaps other?

So CJ, it's Greg. When you look at Q4 it's really all in compute and memory is where the upside is coming from. If you look across the rest of the company, it's kind of not too different quarter-on-quarter, maybe a little bit stronger in our product test division, a little bit stronger in robotics. But the real story is in compute and memory. And I'd say it's kind of 2/3, 1/3 in terms of the like compute is kind of 2/3 of it, memory is about 1/3 of it, and HBM is really strongly represented in that memory up.

Very helpful. And then I guess, longer-term question on the compute side. I would be curious, as you think about high-performance compute, leapfrogging mobility reports suggest that NVIDIA is going to be the lead customer for -- with [indiscernible] How are you thinking about compute intensity? How are you thinking about increased test insertions? And really, how are you thinking about kind of test time in a world where compute is driving leading edge?

So the -- we're pretty bullish about it in general, that as the [ die ] sizes get bigger and the performance required from the devices sort of gets -- the performance goes up, the test intensity also has to go up.

The other thing that makes us pretty optimistic in terms of sort of how this will affect the TAM for compute devices is that chiplet-based designs are becoming more and more of a thing. And when you get to the later stages of building up these complex multi-chip packages, the cost of -- sort of the cost of scrap really, really escalates. So that drives this sort of shift left adding tests intensity upstream.

And then there's also the downstream effect, which is these chips are going into data centers. And as NVIDIA likes to say, they're being used as if they're like 1 gigantic GPU. And so like 10,000, 50,000, 100,000 nodes have to work perfectly for an entire training run. So there are tolerance for latent defects coming out in the middle of those kinds of training runs is very, very low. So those sort of environmental factors really make us think that the test intensity for the compute segment is going to continue to grow over the next few years.

The other thing that's happened in compute is that because it's now the primary driver for the semiconductor industry that many of the strategies that we've seen in the mobile space for years and years around things like dual sourcing, are becoming much more important to the producers in this space, like the strategy that you need when you're a small portion of your capital equipment providers overall shipments is different when you actually dominate those shipments. You start to feel a little bit more vulnerable. And so customers in this space are increasingly turning to this notion of dual sourcing their supply chain at every step. And since we're coming from a lower share position trying to gain share, that dual sourcing is actually a very good thing for us.

Thank you, Greg. Appreciate it.

I'll take our next question from Mehdi Hosseini with SIG.

[ The 1 ]Double click on these structural changes that are happening at various test insertion. And I want to focus on wafer level test. How do you see your activity and design wins manifesting into increased penetration in this specific segment. And with the burn in on a wafer level be part of the [ those design is ]? And I have a follow-up.

Yes. So we definitely believe that SLT is a critical part for this sort of late-stage, ensuring that there aren't latent defects going into the data centers. So we think that there's positive effect there. The other thing is that there are new technologies that are coming along, like [ COW OP ], where more complex modules are being built up and they need to go through a relatively extensive system test and burn-in. So we believe that -- so we look at this as sort of a contiguous market between burn-in and SLT because the burn-in is generally done with the devices fully operating, not in a sort of -- in the test state.

Okay. Maybe we could take this off [indiscernible] technology. But when I look at your commentary that in your prepared remarks, one of the design wins have been embedded in your guide. I want to go back to your 2028 EPS target of [ $7 to $9.50 ]. I imagine these design wins [ were a factor in ] when you provided the target at this year. Would that be a fair statement? And how we should think about those target in addition to the design wins that you highlighted in your prepared remarks?

Yes. So we'll update everyone in January in terms of our long-term model. The thing that I -- like as we're thinking about it, the thing that is apparent to us is that the long-term destination is not all that different, but the composition of the market is. So we believe that we're well positioned to achieve the long-term model that we have published prior, but we believe that the composition of that business is going to be much more heavily dependent on the things that are being driven by the data center build-out. And that's across compute, networking, memory, even power that, that is far more important in the mix than the way we were looking at the long term before.

We'll move next to Timothy Arcuri with UBS.

Greg, so I assume Semi Test is up something like $200 million in the guidance for December. So I'm just wondering if you can give us a sense. I know that this stuff is all pretty lumpy. But can you give us a sense like it seems like it's maybe evenly split between memory and SSD. Is that a fair just general number to kind of think about in terms of the composition of the growth in calendar Q4?

Yes. So -- the -- it's not quite -- it's not half and half. It's more 2/3 compute and networking and 1/3 memory in the up.

Okay. Okay. Great. And then, Sanjay, can you talk a little bit about revenue shaping next year? I know memory tends to be pretty lumpy. I mean, in particular, memory. And I know that there's this big stuff shipping in Q4. So can you just maybe give us a little bit of a sense on Q1? I mean, is it -- should we expect it to be down a little bit? And how do you see next year from a loading point of view?

Sure. I think Greg and Greg's prepared remarks, we talked about the key drivers. And in my prepared -- key drivers going into 2026 and overall, we thought that the revenue would be up relative to 2025 tied to those drivers. And in my remarks, as well as Greg, we talked about the acceleration of key projects that straddled Q3, Q4 and Q4, Q1 and the ones in kind of Q1 kind of accelerating into Q4. And so overall, we do see demand accelerating.

I will share that from a seasonality, maybe if that's what you're getting to, is that the revenue mix has changed as we've noted in the second half, really tied to Semi Test, driven by AI tied to compute and memory revenue. Historically, our business has been driven by the mobile launches where we've had significant demand in kind of Q2 and Q3, business is no longer driven by that. If it comes in, sure, we'll have we'll have tailwinds on that front. It's more driven by compute projects. And those are lumpy and they're really tied to customer launches. So I think overall, what you'll see is a little different shaping in the way of seasonality for our business going forward, and we'll give you an update in the call in January.

We'll take our next question from Krish Sankar with TD Cowen.

This is Steven calling on behalf of Krish. Greg, first question for you related to the [ VIP ] customer demand as well as you mentioned of merchant GPU opportunities in the future. I guess first thing is in terms of the VIP customers, how much more expansion do you see in terms of the customer base from the larger [ CSPs ] and similarly for Tier 2 CSPs. Is there a direct relationship that you have also with those customers potentially? Or is that more of a foundry type relationship? And similar question for merchant [ GPs ], is that direct or more of a foundry type of testing relationship?

So in terms of the VIP customer base, it is incredibly concentrated that there are a lot of design starts. There are a lot of chips that are being developed, but the vast majority of the tester demand in the VIP space is really being driven by 2 customers. And that's how that market is playing out right now.

What seems to be happening is that each of the hyperscalers has their own chip development program, and they benchmark that against what they can do with merchant silicon. And if the merchant silicon provides a better sort of tokens per watt, then they don't tend to ramp their internal silicon to the same extent. So like right now, we see that market as concentrated and it's going to expand -- that base is going to expand pretty slowly because it's a very competitive environment.

The thing that we see with VIPs is as those VIPs are growing there, the actual specifier, the chip developer, the hyperscaler itself is exerting more control over the whole supply chain. They are moving from aggregator to aggregator. They are forming different partnerships and they're taking more control over their supply chain. So sort of the way that people look at it in terms of the aggregator, the design partner as the one influencing the decisions, I think that, that is something that will fade over time, and we've seen it that occur in some of the VIP customers.

In the merchant space, it is all at the specifier, not at the foundry. So the merchant GPU or CPU player is the one that is going to decide on the test platform. They're the ones that are going to control the test programs and all the test IP. So our efforts around gaining share in merchant GPUs is directed at the specifier themselves, not at anyone in the supply chain.

Great. And just for my quick follow-up, for hard disk drives, just some of the strength that you mentioned there from the cloud demand. Just curious like for [ System Test ], like, are you expecting strong double-digit growth in that segment for Q4 as well? Or it could be higher than that?

So the process of adding capacity in HDD is it takes time. The manufacturing process for HDD is a complex one, highly automated. It's heavily capital-intensive, and so we're definitely seeing an uptick in the orders associated with that. There's a lot of optimism in the space, but we would expect that to have a greater effect in 2026, then we would expect to see anything in 2025. 2025, I think, SLT is -- we're not expecting to see significant growth in the IST Group in Q4.

We'll move next to Shane Brett with Morgan Stanley.

Let me ask a question in a bit more of a direct way. As of this moment, do you sort of expect SOC test to accelerate from this really strong December quarter into the first half where do you really kind of bake in a bit of seasonal decline or kind of a bit of conservatism into the March quarter?

I'll just reiterate some comments and maybe, Greg, if you want to add to it. The second half of the year '25, we talked about straddling and we talked about the projects accelerating, and we're seeing that continue to accelerate. Of course, there's projects in the pipeline in Q1 and Q2 of 2026. It's just going to depend on how those projects go. Generally, we're seeing projects accelerate though.

Yes. I think as we talked about, there are things that are perched between quarters. The thing that I will say is Q4 was sort of a new high watermark for us in terms of capacity and shipping against our memory test and our SOC products. We expect demand to continue to be robust going into 2026. But like as an analyst, it would probably be a mistake to like look at the growth from Q3 to Q4 and draw a line straight up from there because there -- it's -- we're at a relatively high level, and we expect continued strength, but it is really lumpy and the timing continues to be uncertain, even between Q1 and Q2 of next year. .

Got it. That's helpful. And for my follow-up, it's on memory. At a conference since the quarter, you mentioned that on a go-forward basis, DRAM and NAND will grow but NAND will grow faster because of how low it is right now. Just how low has NAND been this year relative to prior years? And what sort of growth are you expecting for the NAND portion of memory going forward?

Yes. It's Sanjay. So NAND is really low, really from a percentage standpoint. I think it's really going to tie to the growth driven in the mobile industry. [indiscernible] and rough numbers. But the flash view of TAM looking backward was more than double than what it is now. And so it's really contracted. It's a much smaller component of the [indiscernible] TAM.

We'll move next to Jim Schneider with Goldman Sachs.

I was wondering if you could maybe kind of return to your expectations for mobile SOC heading into next year, realizing that Q2 and Q3 are seasonally stronger quarters. Do you have any sense what you might expect sort of directionally in terms of improvement next year? And maybe just remind us of the relative sort of either test time or content increase you would expect moving to the [ N2 ] generation?

Mobile SOC has been at a pretty low level for the past couple of years. Looking forward to next year, we don't know. I think the honest answer is we don't know exactly how big it would be. We're optimistic that it should be bigger than it is this year, but we're unsure of the magnitude of that. And let me tell you the why. So there are 3 factors in terms of how big the mobile TAM is really going to be.

One is the complexity of the part and with N2 and with new packaging technologies like WMCM, we expect that the test intensity per part is going to be higher like double digits kind of higher. The next factor is like yield. For new technologies, sometimes yield is lower. We do not count on that because there's been sort of a pattern of execution at very high yields for these kinds of products. So we are not expecting that to be a particular tailwind. The one that's really the most important factor right now is in unit volume that if there is a significant upward inflection in handset sales, people are refreshing phones because new phones are doing something interesting, then that not only drives the whole mobile processor space but it also drives [ RF and PMIC ] and everything else.

So I think the big X factor for us is whether we see an inflection in unit sales in 2026. If we do, it could be a strong year. If we don't, it would probably be just a modest improvement from where it is.

That's helpful, Greg. And then maybe from a sort of financial perspective, obviously, 2026 is saving up to be a relatively solid growth year for you, but maybe you can remind us of the OpEx leverage you expect to get in the model, in other words, for every dollar of revenue increase [ sort of every ] 10% of revenue increase, how much OpEx increase would you expect to flow through?

It's Sanjay. So as I said in my prepared remarks, our growth in OpEx was a little bit higher relative to the leverage we wanted to have -- we have in our overall earnings model, and that's going to happen from year-to-year. But in 2026, and our operating principle is basically for every dollar of revenue, we want to have of growth, we'd see roughly 50% of that in OpEx growth. So kind of like a ensuring that we are driving OpEx leverage. We expect to be at roughly that rule going into next year or being at that target roughly in 2026.

We'll take our next question from Samik Chatterjee with JPMorgan.

Sanjay, congrats on the retirement. And Michelle, congrats on the role as well. Maybe for the first one, Greg, I'm curious, I mean, you're calling out the memory increase overall in 4Q [indiscernible] step-up for next year as well, how much of the improvement here is related to market share, [indiscernible] you've talked about market share [ wins that you talked about ] earlier in the year, relating to sort of general industry purchasing [indiscernible]. And when you look at the overall portfolio, what are the end markets you would expect to sort of gain share [ in which the area ] you would sort of highlight as share opportunities as you look to 2026? And then I have a quick follow-up.

Yes. So the way that we look at the memory market is primarily -- think of it as like a 2x2 grid. We think of DRAM and we think of flash memory and then the other axis is wafer sort and then final or performance test. And if you look at like within each segment, our share is like very high like flash final test and DRAM final test HBM performance test. We have healthy share in the final test section of the grid. Our share overall in the wafer sort, excluding HBM performance test is significantly lower. And so when the market is dominated by purchases for final test, our share tends to go up. And when the market is dominated by capacity adds for wafer sort, our share tends to go down.

Looking into 2026, we're expecting that it's going to be an expansion year for memory in general. And so we think it's going to be a good year for us, but we also think that there's going to be a significant expansion in the [ SAM ] for the wafer sort part of the market. So I don't know if we're going to see significant share increase but I'm pretty sure we're going to see revenue growth.

Okay. Great. And then a quick 1 for Sanjay here. Sanjay, and I apologize if this has been addressed before I jumped on a bit late. But the gross margin guide for 4Q [ rate to where you ] ended 3Q given the volume leverage that you [indiscernible] seems a bit more sort -- can you just walk through the gross margin driver for the fourth quarter?

So obviously, volumes are going up, and that's a tailwind. And -- at the midpoint of our guide is 57.5%. And the headwind is really driven by 2 factors. First, we are investing in factory expansion in multiple geographies. That's a bit of a headwind. But I think the larger headwind is really tied to the significant acceleration tied to projects and end market demand. And to meet our customer delivery requirements, we've gone out and we've sourced some -- think of it as onetime kind of supply that's had kind of onetime, a little bit of elevated cost tied to meeting kind of customer requirements. And so that's what's occurring in Q4. So 2 headwinds offsetting the volume tailwind is really expansion of the factory and some supply chain kind of cost increase that's somewhat transitory or onetime.

We'll move next to Brian Chin with Stifel.

Sanjay, definitely wish you the best. For the first question, on the AI accelerator part of the business, can you give us a sense of how significantly weighted that was the second half and maybe 4Q this year. And obviously, a lot of networking strength this year in SOC. For the full year, is AI accelerated revenue significantly above what you expected entering the year?

That's -- it's an interesting question because our view of the year changed pretty dramatically between January and March. So in January, we were pretty optimistic about how 2025 would come out. By March, we were far more pessimistic because there was a lot of uncertainty in the market by the -- but leaving the year, like looking at it from the -- towards the end of 2025, I would say that our -- like sort of maybe this is at a higher level than you asked the question, but definitely significantly higher revenue in compute, both VIP compute and networking have worked out to be stronger than we expected coming into the year. I would say that mobile is weaker than we expected coming into the year and auto and industrial is a little bit weaker as well. But the up in the compute space is the thing that really kind of brought us back up to that level. Memory, I think, has strengthened from our perspective at the beginning of the year, but not as significantly as the growth in the compute space.

Okay. Great. And then for maybe just kind of a clarification in the question. In industrial robotics, do you expect Q4 to shift some positive seasonality Q-on-Q or because even approach flat on a year-over-year basis? I know [ there's some headwinds ] for the year and probably into the back half of the year. And then kind of maybe not that side, but just very curious on the [ Titan SLT ]. Just for any given accelerated [ ship ], is it common to just win in that insertion kind of towards the back end of that test queue? Or is it more common do you think to win multiple subsequent insertions as well?

So in terms of SLT, once you have been designed in for a particular AI accelerator, then our -- the Titan system for these AI accelerators was designed with sort of significant upgradability. So going from part generation to part generation as long as it fits within the general power envelope that we can provide, and there's some headroom there, then you're able to do change kits and upgrades. So there is an incumbency advantage in SLT. Kind of it's similar to the incumbency advantage that you have in ATE, not quite as strong, but pretty strong. .

And then on the robotic. So there is some seasonality that we are expecting. We do expect an increase from Q4 to Q3 and [indiscernible] a weaker kind of automation market. And as you know, we're still working through the strategic shift to large accounts and OEMs, but we do expect a seasonal outlook.

One comment on robotics is that we've seen through this year that our ability to predict our revenue is somewhat limited. We are -- it's a very high turns business. We see demand shifting and demand responding to sort of current events depending on where you are. So we're trying to be as cautious as we can in terms of predicting growth. We do expect that Q4 would be stronger, but we're not predicting a gigantic hockey stick or anything like that.

We'll move next to Vedvati Shrotre with Evercore.

The first 1 I have is a follow-up to the HBM questions you had. So on HBM, you talked about a new test insertion, like a singulated die test and you're shipping volumes in 3Q. So I wanted to understand if that is the norm now? Has that increased the TAM and are all suppliers expected to do this, [indiscernible] factors?

Right now, only 1 major manufacturer is routinely doing this singulated stack testing and I wouldn't say that it is even pervasive across all memory types that we're not sure if there is a significant improvement in downstream yield. The device that the HBM gets put into, if yield of that device goes up and HBM related faults go down as causes of problems downstream, then that will proliferate across multiple manufacturers. But right now, it's really only 1 of the 3 major HBM manufacturers is doing that for a high volume of devices.

Understood. Okay. And then my follow-up was on SOC. Have you -- can you provide any color on how much compute is as a part of SOC in second half '25? And then even as we think about 2028, I think your -- during the Analyst Day, the idea was the composition would be 1/3 mobility, 1/3 compute and 1/3 auto industrial market. So do you have an update to that now that compute is a strong driver?

So do you want to take the compute second half?

Sure. Well, I won't break out the specific numbers. It is a significant component, just as we talked about. And in the second -- in mobile was more first half dominated by the supply chain shift and think of the second half of our business is very strongly driven by compute, really tied to VIP and networking. So a very significant component.

So I did a quick math and like. So this is trying to understand AI driven. So it includes all of the memory-driven stuff and the SOC compute stuff. Just like from Q3 to Q4, like 50% of our total revenue in Q3 was coming from AI-driven stuff in those segments. It's up to like 60% in Q4. So this is -- it's a very different composition of business in 2025 and especially the second half of 2025 from where we were before. We'll update you in January in terms of what our long-term model is. But it's safe to assume that, that model is going to have a much heavier weight on the compute and the compute part of the market and the AI-driven parts of the memory market.

We'll move next to David Duley with Steelhead Securities.

I guess, first, I had a clarification. You talked about strength in AI in Q4 coming from networking and hyperscalers and HBM. I'm assuming that you have not won any business on stand-alone GPUs yet, and that is not included in any of the guidance statements.

Yes, that's correct. Look, we're making good progress, but we have not included that in our guide and there wasn't revenue for that in Q3.

Okay. And then as my follow-on, do you think you could update us on the size of the SOC TAM and then perhaps some of the major pieces like the high-performance computing piece? And then just one last question is, as far as your HBM ramp going into Q4 -- excuse me, HBM ramp going into Q4, is that mainly driven by HBM4 ramping up? Or is it driven by new test insertions or is it driven by something else?

So let me take the second 1 first. So the HBM ramp that we're seeing in Q4 is probably half and half between new test insertion and additional capacity for -- and it's really like new test insertion singulated stack test, additional capacities for stack die wafer level test. So both of those are increasing, but it's all around capacity adds for HBM4 or like it's all driven by HBM4. Now we are not providing an update to the SOC TAM mainly because the SOC TAM is all over the place. And for both Teradyne and our competition, there is a lot of dynamic action between Q4 and Q1, that's going to have a significant impact on the ultimate size of the market and especially the size of the high-performance compute market. So we're going to stick with our overall guide, and we're going to watch how that turns out.

This concludes the Q&A portion of today's conference. I would now like to turn the call back over to Greg Smith for closing remarks.

Thank you, operator. I'd like to offer a quick final thought. I mentioned in closing the July call that AI was having a profound and positive impact on Teradyne's business. I'm encouraged by how quickly we're seeing returns on our investments to pivot to AI. AI is the dominant driver of our business for the foreseeable future and will continue to align ourselves to the outsized opportunities that it offers. We've made great strides so far in 2025 and while our progress is not expected to be entirely linear, we're more excited than ever about our prospects for continued profitable growth in the years ahead.

Thanks for joining us today, and I look forward to updating you on our progress in January. Thank you.

This concludes today's Teradyne third quarter 2025 earnings call and webcast. You may disconnect your line at this time. Have a wonderful day.