Arm Stock

Good day, and thank you for standing by. Welcome to the Arm First Quarter Fiscal Year 2027 webcast and Conference Call. [Operator Instructions] Please be advised that today's conference is being recorded.

I would now like to hand the conference over to your first speaker today, Ian Thornton, Vice President of Investor Relations. Please go ahead.

Thank you, and welcome to our first quarter fiscal 2027 earnings call. On the call are Rene Haas, Arm's Chief Executive Officer; and Jason Child, Arm's Chief Financial Officer.

Today's call contains forward-looking information about the company and its financial results. While these statements represent our best current judgment, our business is subject to many risks and uncertainties that could cause actual results to differ materially. Important risk factors that may affect our business and future financial results are described in our annual report on Form 20-F filed with the SEC. I assume no obligation to update any forward-looking statements. We will also refer to non-GAAP financial measures. Reconciliations of these non-GAAP financial measures to the most directly comparable GAAP measures can be found in our shareholder letter as can a discussion of certain projected non-GAAP financial measures that we are not able to reconcile without unreasonable effort and supplement to financial information. Our earnings material are available at investors.arm.com.

And with that, I'll turn the call over to Rene.

Thank you, Ian, and welcome, everyone. Arm delivered a record first quarter and a strong start to fiscal 2027. -- our results reflect growing demand for the Arm compute platform as AI expands across cloud infrastructure, edge devices and the physical world. Revenue reached $1 billion [indiscernible] up 22% year-over-year, driven by record first quarter licensing and royalty revenue. Royalty revenue grew 22% to $715 million Licensing revenue grew 23% to $574 million, and non-GAAP EPS increased 29% to $0.45, above the high end of our guidance.

AI is changing where and how compute happens. We're seeing that in the data center where the transition to ARM continues to accelerate. And we're seeing it beyond the data center as AI expands into PCs, smartphones and physical AI applications. Across each of these markets, customers are increasingly standardizing on the Arm compute platform. These trends are the driving demand for the ARM AGI CPU. We introduced the Arm GI CPU in March to give customers another way to deploy the ARM compute platform. Since that time, we've made significant progress. Initial product has now been delivered to multiple customers, and we have secured the manufacturing capacity needed to support the $1 billion opportunity we outlined last quarter across fiscal 2027 and fiscal 2028. Demand now exceeds $2 billion as we continue to add new customers, including multiple customers in the U.S. and China, while the overall value of our pipeline has continued to strengthen.

We're also working closely with our manufacturing and supply chain partners to expand capacity. Our confidence in achieving upside to our $1 billion opportunity for the AMAG ICP business has increased in the past 90 days. This momentum is part of a much broader shift already taking place across their [indiscernible] business. Data Center royalties more than doubled year-over-year, once again as adoption of Arm Neoverse continues to expand. This pace at which Arm is becoming the CPU foundation for AI infrastructure is accelerating.

Arm [indiscernible] shipments have now surpassed 1.5 billion cores with most recent $500 million shipping in just the last 9 months, where the first $1 million took 6 years. The world's leading AI infrastructure providers continue to validate that trend. NVIDIA has now brought Vera into production. Built on Arm Vera delivers up to 50% higher CPU performance and 2x greater energy efficiency than comparable x86 systems and will serve as the CPU Foundation for NVIDIA's next-generation AI infrastructure.

Google has stated that its ARM-based Axeon CPU is a core component of its AI infrastructure strategy and the host CPU for its latest TPU AI systems. AWS announced plans to deploy tens of millions of Graviton 5 cores to power agent AI workloads. Microsoft expanded Azure Cobalt 200 virtual machines built on ARM never CSS. Additionally, Qualcomm has also [indiscernible] plans to enter the AI data center CPU market with ARM-based Dragonfly C1000. Each of these companies is approaching AI infrastructure differently, but they're all moving in the same direction. ARM-based CPUs are becoming central to next-generation AI infrastructure.

IDC reported that spending on ARM-based accelerated server platforms has nearly doubled in the past 2 quarters and has now surpassed x86 platforms. We're witnessing both the rapid expansion of AI infrastructure and ARM's growing role within it. Our opportunity extends well beyond the data center. As AI moves into production, customers are increasingly focused on the economics of deploying AI at scale. Whether AI runs across the cloud, the edge or ultimately the physical world, efficient compute is becoming just as important as model capability, bringing AI closer to where data is created improves performance, reduces latency, enhances privacy and lower infrastructure costs. This is where Arm has always differentiated itself. The result is a new generation of computing devices spanning into 2 distinct categories: efficient AI PCs designed for mobility, and more powerful Agenetic systems capable of running sophisticated models locally.

NVIDIA introduced RTX Spark, the first AgenticPC built on Arm compute subsystems, enabling sophisticated AI agents and larger AI models, which can run locally. For on-the-go AIPC, those same OEMs continue expanding the Windison-arm ecosystem with new Snapdragon-powered APCs -- while Google's continued investment in AI-enabled Chromebooks is broadening access to on-device AI. As AI drives a transition to the next generation of personal computing, Arm's opportunity continues to grow across an expanding range of AI-enabled devices.

The same economics extended into the physical world. vehicles, robots, industrial systems and autonomous machines increasingly rely on efficient, secure and real-time arm-based compute to sense, reason and act autonomously. NVIDIA recently expanded its physical AI platform with Kosmos 3 and the Isaac Group humanoid robotics platform powered by Jetson Thor, which combines an Arm-based CPU with NVIDIA black well GPU. ARM's software ecosystem continues to expand and now supports more than 22 million developers worldwide. During the quarter, Arm introduced performance with support from Microsoft, MongoDB, Redis and SAP, helping developers and AI agents, analyze and optimize workloads running on ARM-based infrastructure.

We also expanded our AI developer tools, including the ARM MCP server, which has surpassed 10,000 docker downloads and integrates Arm's expertise into leading AI developer environments. From cloud infrastructure to PCs and physical AI, developers can build on the same ARM architecture and software ecosystem. Customers can deploy ARM through IP, compute subsystems or silicon, depending on what best fits your business. But in every case, they're building on the same Arm compute platform, software ecosystem and developer community.

As AI becomes part of every cloud, every device and every sector, the industry is increasingly converging on a common compute platform. We believe that convergence will define the next day in a computing. AI is changing where and how it [indiscernible] happens and Arm is at the center of it. With that, I'll turn it over to Jason.

Thank you, Rene. We have started fiscal year '27 with another strong quarter, delivering the highest first quarter revenue in our history. Total revenue grew 22% year-on-year to $1.29 billion. Royalty revenue grew 22% year-on-year to $715 million, also our highest-ever figure for Q1. Once again, the largest driver of royalty growth was cloud AI. Data center royalty revenue continues to more than double year-on-year, reflecting our sustained momentum across the market. This is being driven by the continued ramp of ARM-based server chips at all the major hyperscalers alongside increasing deployments of data center networking chips, particularly DPUs and SmartNIC, where Arm technology is deployed in nearly all leading products.

Edge AI royalty revenue continued to grow despite a soft end market in [indiscernible]. We continue to benefit from higher royalty rates as RMB 9 and compute subsystems continue to increase their penetration in the smartphones, tablets and other consumer electronic devices. These drivers more than offset the decline in smartphone sales due to higher memory prices. Physical AI also made a strong contribution to royalty growth, supported by the continued secular expansion of ADAS and autonomous systems built on ARM technology.

Turning now to licensing. License and other revenue was $574 million, up 23% year-on-year, also a record for Q1. Growth was driven by strong demand for next-generation architectures and deeper strategic engagements with key customers. This quarter, we signed multiple high-value agreements as existing customers renewed long-term licenses. Some of the world's largest hyperscalers, automotive and robotics companies and handset OEMs all secured access to Arm's future road map for their next generation of products.

Of the $574 million of license revenue, our agreement with SoftBank for technology licensing and design services contributed $193 million. We expect the quarterly run rate for the rest of the year to be around $200 million. As always, licensing revenue varies quarter-to-quarter due to timing and size of high-value deals. So we continue to focus on annualized contract value, or ACV, as a key indicator of the underlying licensing trend. ACV grew 13% year-on-year, maintaining strong momentum. This continues to be above our long-term expectations for license revenue growth.

As Renee mentioned, customer demand for Arm AG ICP remains very strong. And this quarter, even more customers have wanted to place orders with us. We've secured the manufacturing capacity needed to support the initial $1 billion opportunity and we have made progress to securely additional -- to secure additional supply as well as optimizing our customer mix and commercial terms.

Our confidence in achieving more than $1 billion has increased in the past 90 days. We'll provide an update at our Q3 results, which is when we will have better visibility of Q4 2027 and fiscal '28.

Turning to operating expenses and profits. Non-GAAP operating expense was $733 million, up 18% year-on-year due to the ongoing R&D investment. This was about $27 million below our guidance due to timing as our spending plan for the year remains largely unchanged. We are expanding our engineering teams to support increasing customer demand while advancing the technologies that will underpin future growth, including next-generation architectures, compute subsystems and RMB [indiscernible] product family. Non-GAAP operating income was $531 million, resulting in a non-GAAP operating margin of about 41%, up 200 basis points year-on-year. Non-GAAP EPS was $0.45 and driven by both higher revenue and slightly lower OpEx than expected.

More importantly, even at these elevated levels of investment, we generated free cash flow of $665 million for the quarter and $1.4 billion over the trailing 12 months, giving us the flexibility to continue to invest long-term growth.

Turning now to guidance. For Q2, we expect revenue of $1.38 billion, plus or minus $50 million. At the midpoint, this represents revenue growth of about 22% year-on-year. We expect license and other revenue to be up about 30% year-on-year and royalty revenue to be up in the low teens year-on-year. We expect our non-GAAP operating expense to be approximately $780 million and our non-GAAP EPS to be $0.47, plus or minus [indiscernible].

Looking ahead, we continue to see strong customer demand across our business, combined with the expansion of our technology portfolio and deepening strategic customer relationships, this gives us confidence in our long-term growth outlook. With that, I'll turn the call back to the operator for Q&A.

[Operator Instructions] And our first question today is on the line of Joseph Quatrochi from Wells Fargo.

Yes. Maybe just first, can you give us a little bit more detail on just the confidence that's increased in terms of the upside to the $1 billion for AI revenue. Is there any clarity you can provide just on is it better way for supply customers getting access [indiscernible] any help there.

Yes, I'll take that question. I think in general, it's all of the above. If we -- if I look back to the call we had 90 days ago, we had committed to supply of $1 billion -- we talked about demand pipeline of $2 billion, and we were working to secure supply for that delta between 1 and 2. 90 days later, the demand picture has even gotten better as Jason mentioned, north of $2 billion. But our ability to secure that supply for the north of $1 billion, our confidence has increased in the last 90 days. And I would say it's across all of those areas. Making a chip is complex relative to supply chain. You have wafers, you have substrates, you have test capacity, you have memory and all of those areas our confidence level and being able to secure the supply necessary has gotten better, and that's the commentary we're stating at this time.

And maybe just as a follow-up to that, I mean, on top of the the extra above $1 billion of AGI revenue that you now think you can do. How do we think about the gross margin structure relative to kind of what you were thinking about for the original $1 billion?

Yes. No change at this time. We said that -- I think last quarter, we said there was probably going to be somewhere in the high 30% range, maybe low 40s for the first generation for really in Q4 this year as well as for next year. As I say, over the next couple of years, we do expect to get to 50%, but that's going to basically just entail us actually bringing more of some of the work that maybe an [indiscernible] has helped us with the past and bring some of that work in-house. So that will probably take a couple of years -- those are initial expectations and no real change from last quarter.

Obviously, with some of the price increases and things that have happened which you're kind of working through how to digest those and what needs to be kind of -- what needs to affect our pricing. So those are some of the things that we're going to work through. And as I said in my prepared remarks, we'll provide a much more detailed update for both the revenue and margin profile before we go into Q4. So we'll be at the end of Q3.

We will now take the next question. And the question comes from the line of Sebastien Naji from William Blair.

My question is on the smartphone market because of the much higher memory and cost, handset OEMs are absorbing meaningfully more BOM inflation. They're starting to raise prices. I think given the smartphone royalties remain a significant portion of your overall royalty base, can you maybe just talk a little bit about what you're seeing in terms of unit demand and mix and how this impacts your royalty revenue outlook for the fiscal year?

Yes. So I'll take the first part of the question, I'll let Jason talk about some of the details around the numbers. Generally speaking, we have been somewhat isolated from the negative growth of the smartphone market because of the fact that we have moved the vast majority of the customers to [indiscernible], in some cases, [indiscernible], in some cases, the second version of CSS. And what that's all contributed to has been an increase in royalty growth. So whereas the smartphone market has been going -- projected to be down by some level of double digits. We are projecting double-digit growth in royalties and smartphone market vis-a-vis exactly how that translates to the forward guidance and forecast. I'll let Jason do in some more color.

Yes. I think there's obviously others that are reporting this week and key partners that we'll learn more from. But when we look at our forecast, we mostly try to look at the industry projections across all entire industry, IDC is, of course, 1 of the things that we look at. And then, of course, we update for partner mix, where we have slightly different royalties based on whether it's CSS or B or behavior, et cetera. So -- so when we add all that together, we have seen some incremental slowdowns versus what was expected at the beginning of the year. And I think the new piece was initially the expectation was just going to affect the lower end of the market and now we are seeing all parts of the market, even some upper and mid-tier being affected. And so that's maybe an incremental for us.

And so as a result, I would say, I think going into the year, we probably thought we were -- I think we said last quarter that we're expecting somewhere around 20% year-on-year for the next few years in royalties included this year. Right now, if I had to guess, that's probably somewhere closer to the high teens right now. But hard to say until we learn more about the next couple of quarters. I say right now, the -- we do think royalties will come down a bit in this next quarter, we guided to kind of the low to mid-teens for Q2, and we'll give you a better update next quarter.

But I think the one thing I would make sure you don't sit of is, the good news is the overperformance that we've been seeing at Cloud AI continues to help offset that. And so while there is weakness on the smartphone side, the overperformance on the cloud AI business, continues to accelerate. And that's the piece that gives us confidence in full year and next year's numbers. But again, we'll give you more updates over the next couple of quarters.

[Operator Instructions] And the next question today comes from the line of Gary Mobley from StoneX Group.

You highlighted maybe a little bit of upside to the demand profile for the next 2 years, $2 billion plus, I think to paraphrase correctly, but what about the demand pipeline for the out years fiscal year 29 through 31 in -- and I assume the $15 billion in projected AGI revenue expected for 31 contemplates all the supply chain-related headwinds. That's a supply served revenue, correct?

Yes. I'll let Jason comment further in terms of any forward-looking comments he wants to make. But I'll say what has changed since March? Is that, to your point, clearly, supply is an issue across a number of different factors. The outlook that we gave at that time was based upon our view of the supply picture. On the flip side, we did say at that event that the CPU TAM, we thought was about $100 billion in those outer years. And going back, you may recall that prior to that, most folks were talking about a number of around 50 to 60. So when we talked about the $1 billion number in March, there was some surprise around it, and there was a lot of back and forth in terms of justifying that number.

Since that time, a number of my peers have talked about numbers quite a bit bigger than that, in some cases, up to $200 billion. I think all of that is really being driven by the increase of overall compute capacity which will be inference as -- and all of that for space compute is going to largely be running genetic workloads and the Agentic workloads are essentially capacity constraints in terms of throughput by the number of CPUs you have. So as inference demand goes up, which it clearly is [indiscernible] and as a genetic demand goes up as it clearly is, that means [indiscernible] obviously, it's gone up to, I think, now the most recent estimate is [indiscernible] sitting on -- from a variety of sources.

So I think our expectation is if you flow through the same kind of market share that we were expecting at 100, could it be much higher if in fact the market is at 200-plus well certainly, it could be. The constraints over the next couple of years, I would say, that for us, FY '27 or calendar [indiscernible] which is, I guess, calendar 27, supply chain is pretty tight. And so we're maybe there's upside to get to this $2 billion or $2 billion plus, and we'll provide updates on that over the next couple of quarters.

But to go beyond that, the supply chain is going to loosen or at least expand quite a bit as we get it into calendar '28 and '29 or for us, FY '29 and '30. Right now, if you look at wafer capacity and memory capacity, you probably know as well as I do. I think there's estimates that the capacity is going up somewhere between 70% to 100% depending on your assumptions on wafer and some of the different some of the different partners that can help there and even more so on the memory side.

So our expectations are -- there certainly is potential to to beat the number that we give externally. Internally, we do have higher targets. But until we start to see the supply chain capacity come online. We're mostly focused on just trying to get to the $2 billion number next year, and we'll provide updates, of course, later as we learn more.

Next question today comes from the line of Tom O'Malley from Barclays.

Last week, A&D hosted at Analyst Day and talked about a $220 billion [indiscernible] by the end of the decade. And kind of underneath that, the largest contributor was agenetic AI that kind of laid out traditional CPU. They get out head nodes and then agent applications. So I was curious, when you look at your silicon business longer term, -- where do you see that fitting in? Is that going to cover all 3 of those sub-buckets in your early wins in your early pipeline? Is there 1 way that that's leaning or another? Just would like to get a flavor of what you foresee for that silicon business.

Yes. And I'm sorry, could you repeat those 3 categories, so I make sure I've got the definitions right.

Yes, traditional server and then you have head nodes and then you have Agentic [indiscernible].

Got it. Yes, Arm AGI CPU is going to play in all 3. When we talked about the customers that we had signed up back at the ARM ever event, we had folks like Cerebras and OpenAI, which were largely around head [indiscernible] type applications. We also talked about meta and Cloudflare, which are around Agentic applications, but also general-purpose server. We've also just talked about [indiscernible] Oracle. So the Arm AG ICP is a very good fit for all of those to be quite candid, which is why we are very, very optimistic about the demand. And we did talk about quite a number of customers that day, and we were very specific about inside the cloud, the Agentic workloads, it's a pretty broad term because the agents are running through the head mode, but they're also running through the general purpose, racks and sit inside the data center.

So short answer is those 3 categories that you defined as described earlier, the RMA GI CPU is a great fit for all 3, and we have customers in all 3.

Your next question comes from the line of Vivek Arya from Bank of America.

I had a near and long-term question on your AGI CPU. Near term, I'm very curious what is preventing on from securing supply for just $1 more billion? I mean, it's a [indiscernible] billion market, right? And there are a number of foundries who can make it. So I'm just curious what is preventing on from getting that extra $1 billion a year from now? And then longer term, Rene, when I look at the 3 players that have Arm based CPUs in AI, whether it's NVIDIA or Amazon or Google, they each have their proprietary accelerators also which are often codesigned with the CPU. So don't you think that restricts Arm's opportunity given that you have a CPU only offering? Or is your intention to add other things to that CPU only offering over time?

Yes. So I'll take both parts of that question. Jason, you can add on. I'm glad you think $1 billion is not a big number. There are a lot of folks that's trying to secure extra supply up to $50 million. It is a very, very tight market across everything, whether it's memory, whether it's test equipment, whether it's substrates, whether it's TSMC wafers, it's a very, very, very tight world. I think even my compatriots who are in the CPU business for living and have their own fabs have not been able to supply the demand. So -- we are more optimistic than we were 90 days ago, which is great. Demand has increased, but we feel better about the supply side.

On the proprietary accelerators, if I understood your question correctly, a couple of proof points that, that is an area that we can certainly play in. First off, at Google, they have used Axon as an interface into their TPUs away from x86. So clearly, there's a space for Arm to exist with custom accelerators. If your question was around the Arm AGI CPU specifically, there has been some announcements made by NVIDIA about NVLink fusion, and it's a very interesting platform approach where they're offering a mix and match where you could take a vera CPU and plug it into a proprietary accelerator or you could take a CPU made by someone else in the Arm family and connect it into a Ruben accelerator.

So we don't have anything to announce around that today, but there are absolutely paths for Arm to connect to custom accelerators, whether it's through a sell posted design done internally and/or connecting to something through something like [indiscernible].

Next question today comes from the line of Vijay Rakesh from Mizuho.

Good to hear you guys got some capacity on the Agentic AI I CPU side. I was wondering, when you look at your agentic AI CPUs are they mostly in the 128 core? I know core count is important. So are you looking at -- when do you expect the next 192 or [indiscernible] core CPUs coming out? And any color on the ASIC side as well? And for Jason, do you expect to like start breaking out the silicon side of revenues, the aging CPU side because that seems to be becoming bigger and bigger, I guess.

Yes. So thank you for the question. On the product side, nothing to tell you specifically today about the road map and where we're going with that. So the Arm [indiscernible] CPU today is is 128 core-based design. There are instances in the market using ARM that are better -- or actually a greater number of cores, and that's Graviton 5, which is 192 cores. -- certainly, the direction of travel is more cores. And the reason for that is for running a genetic workloads, more cores is a better outcome because the software overhead is simpler. You can run brutal machines on virtual jobs on single cores, and they're going to be much more efficient in terms of throughput.

So while I don't have anything to talk about today in terms of the road map and the quire count, it's not a stretch to think that the next-generation designs are going to have more and more course because that's clearly the direction of travel. And I'll let Jason comment on the second part of the question.

Yes. On the revenue breakout, well, First, as soon as we start shipping, which will be at the end of this year, that's the first milestone. And then once it becomes at least 10% of revenue, we'll break it out separately as a third line. So separate from license, separate from royalties, last day we be solid revenue. So I would expect that based on our forecast, that should be broken out then in FY [indiscernible].

Next question today comes from the line of Charles Shi from Needham and Company.

Jason, I think if I hear correctly, you are talking about maybe royalty revenue growth. This year instead of 20%, probably going to be high teens because of all the things happening in the smartphone incremental weakness, et cetera. anything we should be looking at at the moment, let's say, on the licensing side that can show up with a little bit of upside to offset the incremental weakness on the royalty side. And maybe on that topic, what's the early view on maybe FY '29, which is largely FY '28? And do you think the 10% royalty is still a good number and I want to get some thoughts there.

Yes. In terms of guidance for Q2 we did lower, I think, maybe what our expectations were a couple of quarters ago. We didn't guide to it, but I think our expectations were that we're going to be a little higher -- and so we're now guiding to 13% royalty growth in Q2, but then we also increased our license growth. So overall, on a combined basis, we're actually ahead of where we'd expect to be. I would expect similar trends for the rest of the year, and that is any softness or weakness that we experienced in royalties. -- will be at, I would say, at least as much license revenue to overcome that. So I don't really expect overall to be any sort of a slowdown.

Now the thing that could change is we continue to see the cloud AI business overperformed -- and in particular, we're really seeing strong deployments from some of our partners like Google with our Axion as it's growing with all the new TPU deployments. -- certainly, Amazon and then certainly with Bira at NVIDIA, Amazon with Graviton and then we at the hyperscalers that are also exceeding their plans. So -- and increasing their their trajectory and their velocity.

So there is certainly some possible upsides on royalties throughout the year. But if I had to say right now, I would not expect there to be weakness across the overall business. It just might be some weakness in the smartphone side of the business, and we'll just have to wait and see how the cloud AI business if it's able to overcome all of it or just some of it we'll have to update you as we go later into the year.

And then in terms of '29, sorry, '28 and '29, yes, I think the 20% plus royalty growth that we provided, I think that's very much still intact. The things that are happening right now, the only real weakness in royalties is really on the smartphone side in particular to some of the memory issues there. And while they're certainly having a tough year this year, I think most of the partners that we talk to seem to think that there's going to be stronger recovery in the back half of the year. And so I would expect that the year-on-year growth rate by next year should be kind of back in line as we start to lap some of these challenges.

And then, of course, as we continue to see the AI deployments continue to grow at the level they have been that we should be able to, again, be in that 20% range next year and beyond.

Our next question today comes from the line of Krish Sankar from TB Cowen.

This is Steven calling on behalf of Krish. Rene, I had a question on low-cost AI models. Just with the recent cycle of headlines around low-cost whether from China or elsewhere, do you see the proliferation of low-cost AI being a net positive or negative for your AGI CP franchise given your market positioning? And also, if you could also walk us through how low-cost AI might be a protocon for your IP business, like for both data center and the edge side of things.

Yes. Thank you for the question. I'm going to presume that when you say low-cost AI, you're referring to open source models, open weights, et cetera. if the world moves to something that's more open source-based AKA open weights, -- for us, I think it's somewhere between a net neutral to a net positive. And I start with net neutral from the perspective of we're somewhat agnostic relative to whether it's a closed frontier model or an open source model those models all require CPUs. They all require the IP that we deliver for people to build custom chips. So whether it's an opportunity for Arm AGI CPU or the IP or compute subsystems, we're going to be required independent of what those models are, just frankly, just given the nature of the workloads that need to get run, they're kind of independent of the hardware underneath it. You need CPUs to run agents, whether that's open source or close source.

There's a very interesting argument about to be made relative to the potential of those open-source models being much more differentiated, smaller, more efficient that could run in different edge footprints. If that were to happen, it would be a great thing for Arm, just given the footprint that we have in smaller edge devices. Right now, the most sophisticated models that are the closed models, all are literally 100% cloud-based. The RTX Spark that was announced is a very interesting product where you could potentially want some smaller models locally. I think open source has traditionally shown that whether it's in this area or any other software domain, widely frightens the choice relative to how you run different things. In other words, open source leads to higher innovation and differentiation. So I think it's either neutral to a positive depending on where it goes at the edges. But probably the most important thing to remember about Arm and it applies here as well is we're pretty agnostic to the top layer. All of that will need to run on our CPUs and [indiscernible].

We will now take our final question for today. And our final question comes from the line of Timm Schulze-Melander from Rothschild & Co.

Yes. Great. I had two, one was a more kind of operational question, maybe maybe for Jason. Just talking about the timing and the sequential quarter-to-quarter cadence of R&D, you talked about how it had come in below your expectations. And I just wondered maybe if you could talk about how that kind of comes to pass. Is it kind of a milestone-based thing and that maybe they're not met. And so that the cost slips a quarter or two? Is it a capitalization issue? Just how does -- what are the things that dictate the timing of R&D cost in the P&L? And then I had a quick follow-up on the CPU.

Sure. So with OpEx, we did come in a little bit lower. When it comes to OpEx, you have to break out R&D, you'll break it down into really 2 buckets. So you have the people or developer costs and then you have the tools that they're using, which is typically emulation, cloud spend, et cetera. And so in this last quarter, we did come in a little bit below and that would be more not on the people side because that was pretty much right in line. I would say it was really more on kind of the tools that were utilized and necessary versus what our forecast was.

So I think for the most part, you should think about kind of banking that savings that we delivered in Q1. And then going forward, our estimates, I think, are pretty similar, maybe a little bit lower. So overall, again, you can bank the savings from Q1 and then assume that for us the year just at a high level, it's going to grow by kind of mid-single-digit percentage quarter-on-quarter. The overall -- and the things that can move is less around the total number of engineers because that number is one where we can forecast and we're pretty consistent with. It's really just about what is the tool of utilization that's necessary. And so certainly, with things like inlation and cloud spend, that's going to flex a little bit based on what are we learning on the latest developments? And do we need to do you need to increase some of our testing or are we kind of done.

And so as a result, we're getting better and better at forecasting this, but it does move around a little bit because we've really only been doing this now for the last year or two. And so I would expect our brands probably to get smaller going forward.

Got it. Got it. And then just to circle back on the AGI CPU conversion of customer engagement and interest and as you've described, being able to kind of bring that across the Rubicon and book it as a firm order. Just wanted to understand and just kind of circle back to just confirm what we're talking about here. So we're talking about maybe early calendar '27 as you come into the December quarter report that you should have more visibility. And I guess one of the things I'm just thinking about is, is there a concern is it possible or probable that you get into a kind of a golden screw type situation that you've got 99% covered, but because there's tightness, you're missing 1 small component that could kind of [indiscernible] that? Or is it really not quite that vulnerable to one single sort of small component to be missing?

Yes. I make sure I answer your question is as we get closer to fulfilling the demand for the product, is there still some risk at the [ $11 ] that one critical component that we haven't secured could tip the whole thing over. No, there's not a concern of that. We are very confident in terms of partners we're working with. We have a very good understanding of their capabilities and their upside capabilities. So there isn't one single point of failure to your point, of the 1% that could cause this [indiscernible].

This concludes today's Q&A. I will now hand back to Rene for closing remarks.

Thank you. The quarter for us is really a tale of 2 stories. One is the core business itself, the IP business continues to perform extremely well, a record first quarter, and we are guiding ahead of expectations, as Jason had mentioned, and our results were beyond expectations. The growth is really being driven by the data center. Arm continues to increase its market share. and we have many key customers now shipping Neoverse cores which, as I mentioned before, have now surpassed $1.5 billion. But we also now have our new business, the Arm AGI CPU business, which we just announced back in March. The confidence level we have in the success of that product only continues to increase. And as we talked about, the demand is greater than what we shared 90 days ago. And at the same time, our confidence to fulfill that demand has also increased. So all in all, very proud of the results that we've achieved, a great quarter for the company and thank you all for your interest and questions.

Thank you. This concludes today's conference call. Thank you for participating. You may now disconnect.

Good day, and thank you for standing by. Welcome to the ARM Fourth Quarter Fiscal Year 2026 webcast and conference call. [Operator Instructions] Please be advised that today's conference is being recorded. I would now like to hand the conference over to your first speaker today, Jeff Koval, Head of Investor Relations. Please go ahead.

Thank you, Sharon, and welcome everyone to our fourth quarter fiscal '26 earnings call. On the call are Renee Hass, Arm's Chief Executive Officer; and Jason Child, ARM's Chief Financial Officer. Today's call contains forward-looking information about the company and its financial results.

While these statements represent our best current judgment, our business is subject to many risks and uncertainties that could cause actual results to differ materially. Important risk factors that may affect our business and future financial results are described in our annual report on Form 20-F filed with the SEC. ARM assumes no obligation to update any forward-looking statements.

We will also refer to non-GAAP financial measures during the call. Reconciliations of these non-GAAP financial measures to the most directly comparable GAAP measures can be found in our shareholder letter as can a discussion of certain projected non-GAAP financial measures that we are not able to reconcile without unreasonable effort and supplemental financial information.

Our earnings materials are available at investors.arm.com. With that, I'll turn the call to Rene.

Thank you, Jeff, and welcome, everyone. ARM delivered a record quarter and record fiscal year. Revenue this quarter was $1.49 billion, up 20% and our highest quarterly revenue quarter ever and above the midpoint of our guidance. Licensing revenue grew 29% year-over-year to $819 million, driven by strong demand for the ARM platform. Royalty revenue grew 11% to $671 million with growth across Edge AI, physical AI and cloud AI, where our data center royalty has more than doubled year-over-year.

That drove record non-GAAP EPS of $0.60, even while we continue to increase investment in R&D. For the full year, revenue reached a record $4.92 billion, up 23% year-on-year. Royalty revenue was up $2.61 billion, up 21% and licensing revenue was $2.31 billion, up 25%.

Non-GAAP EPS was also a record at $1.77. Fiscal 2026 was our third consecutive year since going public of more than 20% revenue growth, demonstrating the strength of our business and the increasing relevance of ARM in the highest growth areas of compute. This quarter was driven by key highlights, including the expansion of Arm's product strategy and our continued momentum in cloud AI.

As AI is moving from human-based queries to continuous agent-driven workloads, this shift is expanding the role of the CPU. These agentic workloads require CPUs to coordinate tasks, move data, manage memory, enforce security and orchestrate work around accelerators.

As Agentic AI scales, data centers will require more than 4x today's CPU capacity, creating a data center CPU market opportunity of more than $100 billion by 2030. The ARM AGI CPU, which we launched at our ARM Everywhere event last quarter and is purpose-built for Agentic AI addresses this need directly.

Our first production silicon product to the data center will deliver more than 2x the performance per rack compared with X86 platforms with the potential to reduce AI data center capital expenditure by up to $10 billion per gigawatt. Meda is our lead partner and co-developer and is working with us on a multi-generation road map to support personal super intelligence for more than 3 billion users.

The ARM AGI CPU expands our customers can work with or Customers can now deploy ARM compute through IP, compute subsystems or silicon, 1 compute platform, 1 software ecosystem. That is unique to ARM. IP and CSS remain the foundation of our royalty growth. Silicon extends the ARM platform and gives customers another way to build AI infrastructure. Ecosystem support has been significant. More than 50 leading companies are supporting the expansion of the ARM compute platform into silicon, including the very biggest names in the industry.

Customer response to the ARM AGI CPU has been very strong. We now have more than $2 billion of customer demand across fiscal 2027 and fiscal 2028. This is more than double what we stated at launch. We are on track towards our forecast of $15 billion as this business, as stated at our ARM ever event and soon, the data center will be arm's largest business.

The direction is clear. Customers want ARM at the center of the AI data center. Customers need ARM where Agentic applications run and they need ARM or accelerators scale. For example, SAP will move their core database and business application workloads to ARM, starting with AWS Graviton and expanding the ARM AGI CPU.

This represents a significant strategic shift. Cloudflare will deploy ARM across its global network to support traffic management, security and AI inference closer to users. We have also secured design wins with key network infrastructure providers, including F5 and SK Telecom. AI infrastructure needs CPUs and accelerators working together efficiently at scale. NVIDIA, Amazon and Google are already using ARM-based CPUs as head nodes along the accelerator-based systems.

Cerebos, OpenAI, Rebellion and Positron are doing the same with the ARM AGI CPU. This momentum builds on our existing scale in the cloud. That scale is increasingly driven by ARM never CSS and ARM-based compute which now represents about 50% market share with top hyperscalers.

Recent announcements from key customers show that AI infrastructure is being built around ARM-based custom silicon. At Google Cloud Next, Google announced TPU for training and TPU 8 for inference, in both cases, replacing x86 host processors with custom arm Axion CPUs. The increased performance at 50% less power enables an 80% improvement over the previous X86 solution.

AWS continues to scale its custom silicon strategy with ARM-based Graviton alongside Trinium and Nitro while Microsoft is advancing its ARM-based strategy with Cobalt, designed to deliver high-performance and energy-efficient compute for Azure workloads.

And at NVIDIA GTC NVIDIA and Severa, the next-generation ARM-based CPU built for Agentic AI and building a stand-alone rack integrating 256 Beras. Across the largest AI platforms, ARM-based CPUs are becoming central to performance, efficiency and cloud economics. Our opportunity does not stop the data center.

AI is moving to every device and every physical system. -- phones, PCs, vehicles, factories, robots, cameras, sensors and connected devices all need efficient, secure compute with software that scales. These AI workloads will all run on ARM. With over 350 billion chips shipped and over 22 million developers, the ARM compute platform is the most comprehensive in history, and we are positioned to bring AI from cloud infrastructure to the edge and to the physical world through a common compute platform and ecosystem.

We enter fiscal 2027 with record results, strong customer demand and a larger opportunity ahead of us. Our strategy is clear: grow royalties through IP and CSS and add silicon as a new growth vector and scale the ARM platform across the next generation of AI workloads.

With that foundation in place, our focus is execution and continuing to build the future of AI on ARM. And with that, I will hand it over to Jason.

Thank you, Renee. We have delivered another record quarter. Total revenue grew 20% year-on-year to $1.49 billion, which is nearly $250 million higher than our previous record. Our strong Q4 revenue also capped a record year.

Revenue growth of 23% exceeded 20% for the third straight year and lifted fiscal '26 revenue to $4.9 billion. Our revenue strength translated into record EPS for both Q4 and fiscal '26. Royalty revenue grew 11% year-on-year to $671 million, our highest ever figure for Q4 royalty revenue.

The biggest contribution to royalty revenue growth was from cloud AI. Data center royalty revenue continues to more than double year-on-year, and we see no break in this momentum. This is primarily driven by the accelerating ramp of ARM-based server chips by all major hyperscalers as well as increased deployment of data center networking chips, particularly DPUs and SmartNIC, where ARM has close to 100% market share.

In Edge AI, our smartphone revenues continue to deliver growth despite end market weakness. This is driven by our higher royalty rates from increasing penetration of RMB 9 and compute subsystems into higher-end smartphones. Physical AI has also contributed to our strong royalty performance driven by secular growth of ADAS and autonomous systems based on ARM technology. Turning now to licensing. License and other revenue was $819 million, up 29% year-on-year. Growth was driven by strong demand for next-generation architectures and deeper strategic engagements with key customers.

For example, we signed a long-term strategic partnership with the Indonesian government to strengthen Indonesia's capability and the development of AI technology.

We also signed 2 next-generation CSS licenses one to be used to develop chips for smartphones and the other for data center networking chips. These agreements reflect the continued investment by our customers in next-generation ARM technology.

Of the $819 million of license revenue, our agreement with SoftBank for technology licensing and design services contributed $200 million, flat with the prior quarter. As always, licensing revenue varies quarter-to-quarter due to timing and size of high-value deals.

So as we continue to focus on annualized contract value, or ACV, as is a key indicator of the underlying licensing trend.

ACV grew 22% year-on-year, maintaining strong momentum. This continues to be above our long-term expectations for license revenue growth. As Rene noted or as Renee mentioned, customer demand for the ARM AGI CP was very strong. We now have line of sight to more than $2 billion of demand across fiscal '27 and '28.

However, we are maintaining our outlook of $1 billion while we pursue supply chain capacity, and we still expect the first revenues from production ship sales to land in the fourth quarter of this fiscal year. Turning to operating expenses and profits. Non-GAAP operating expenses were $734 million.

This was about $10 million below our guidance and up 30% year-on-year due to strong R&D investment. These investments in R&D reflect ongoing engineering expansion to support customer demand for more ARM technology, including continued innovation in next-generation architectures, compute subsystems, and the recently announced ARM AGI CPU product family.

Non-GAAP operating income was $731 million, resulting in a non-GAAP operating margin of about 49%. Non-GAAP EPS was $0.60, driven by both higher revenue and slightly lower OpEx than expected.

Turning now to guidance. For Q1 we expect revenue of $1.26 billion, plus or minus $50 million. At the midpoint, this represents revenue growth of about 20% year-on-year. We expect royalty revenue and license and other revenue to both be up around 20% year-on-year. We expect our non-GAAP operating expense to be approximately $760 million and our non-GAAP EPS to be $0.40 plus or minus $0.04.

The strength of demand we're seeing, combined with our expanding portfolio and deepening customer engagement gives us confidence in our ability to deliver sustained long-term growth.

For those of you that missed our March Arm EveryWare AGI CPU event, we expect that by FY '31, we will be generating $15 billion in AGI CPU revenue and $10 billion in IP revenue for a total of $25 billion. We expect this to translate to more than $9 in EPS. With that, I'll turn the call back to the operator for Q&A.

[Operator Instructions] and the first question comes from the line of Andrew [indiscernible] from Citi.

It's only been 6 weeks since you held beyond everywhere event and announced the new AGI CPU, somehow it feels a lot longer given the rate of change in the chip industry and in particular, the AI space, you can certainly see that as well in the demand you just highlighted. You're now saying that you've got over EUR 2 billion of demand for the next 2 fiscal years.

I was just hoping that you could give us a bit more detail on how that additional demand has transpired over the last 6 weeks or perhaps type of customer is some of it greater demand with the launch customers you highlighted to us in San Francisco with is it all new interest, new customers applications they've got in mind? And also, Jason, I just wanted to touch on the point you made there where you're not changing the guidance, the demand has doubled and you're now sort of seeking incremental supply from your foundry and I presume memory partners as well.

What do you think you'll be able to do in terms of accessing that supply? How are you going about that?

Yes. Thank you for the question, Andrew. So I'll tackle the first part in terms of where the demand is coming from such a little bit upon the supply situation and then let Jason address sort of the members aspect to it. So 1 of the beauties of the product that we announced at the Arm Everywhere event is the fact that the CPU can come in a number of different flavors, but 1 of the more attractive options is actually buying finished racks from our partners, such as Super Micro Lenovo, ASRock, that enables customers to order and deploy quite quickly.

Many customers that we've talked about are already using Arm, so whether that's through internal designs or running designs at the cloud. So a lot of the software work has been completed -- so the work required in terms of bringing on new compute capacity that's ARM-based, there isn't a lot of friction to that.

So you have a situation where a software is done; and b, you've got availability of a rack design that you can put in the data hall pretty quickly. So to your question, it's a combination of both. It's a combination of some of the customers that we talked about during the day, increasing their forecast.

And there are also customers that we didn't talk about on the day who have said, hey, we are very, very interested, and we're ready to deploy, how quickly can we get units? So the number that we talked about at the end of March was supply in place to support $1 billion of demand.

And that includes memory that includes wafers, that includes packaging, that includes access to test equipment. So for the $2 billion, we are now in the process of securing supply to support that. And the teams are working around the clock to make sure we can find the right answers for our customers. Relative to how that impacts guidance going forward, let me turn that to Jason.

Yes. Thanks, Andrew. So in terms of the expectation, I think we said back at the Armor event that you should assume probably $90-ish million or so of revenue, just $100 million in -- at this point, we're not changing that target. As we get deeper into the year, we will provide some kind of indications on how things go in the supply chain.

And then, of course, in Q3, we'll give you much firmer estimate of what we expect to deliver in Q4 and then maybe some indications of what we think FY [ '28 ] will look like as well.

And the question from the line of track well Go ahead.

Yes. Thanks for taking the question. Maybe first just as we think about, I guess, like the royalty rate growth for 1Q and then just thinking about the full year, is there any help that you can provide just kind of the puts and takes there? Clearly, data center very strong, accelerating but then how do you kind of think about consumer electronics, smartphones, et cetera?

I'll go -- thanks for the question, Joe. So in terms of for, as we said before the quarter, we had a bit of a tough comp in that. We had particularly strong ramp of MediaTek 400 a year ago, more so than what we expected this year.

And so as a result, you saw a bit of a slowdown in royalty revenue. As indicated by our guidance, we're expecting that to get back to the kind of 20% range by Q1. So I would say within the assumptions within our expectations are we will probably continue to see unit growth, I think, actually flip to negative for the mobile market in this last quarter. We're going to continue to see very flattish, maybe slightly negative numbers for the overall market.

The lower end of the market is probably going to be where most of that impact is. So it doesn't have too much impact on us. And then, of course, any sort of negative impact we do see. We expect that to be more than offset by demand in cloud AI or specifically in the data center. I would say on the data center, there's certainly been a lot of positive announcements from some of our partners, certainly, AWS and Google, in particular, about kind of the deployments and either accelerating some of those deployments.

And now you could say that the 3 largest GPU providers across with Vera or Grace. And then, of course, with now Google pairing TPUs with AxION 2, the latest ARM-based chip and then, of course, training all paired with the latest version of arm-based Graviton.

All of those partners are now all in arm, and we're expecting that, that should provide continued growth and upside throughout the year. hard to say exactly what that pace of growth will look like since we're not going to announce their numbers that they haven't announced, but we do expect to see continued upside.

So any of the weakness that we see in memory and mobile, we do expect that to be largely offset plus, I would say, on the cloud side. And then outside of that, the other category of automotive continues to be growing strong and we don't see any changes there, we continue to gain share and continue to grow kind of in that double-digit growth rate. And so overall, we feel pretty good about the royalty outlook for the balance of the year.

The next question comes from Vivek I from Bank of America Securities.

Renee, yesterday when AMD spoke about the CPU market 2030 [indiscernible] $120 billion, slightly more than the $100 million that you had given before. But then they also suggested that they expect to maintain 50% share, I imagine interrelevant in that time frame also.

And then there are all these captive programs, right, Graviton Axion and Vanda trying to understand what is ARM's natural kind of niche in the market? Like which of these parties conceptually can you take share from to your business...

Yes. Thank you for the question. Yes. When we talked about $100 billion TAM on March 24 at the ARM Everywhere day, I think we were the first company to talk about numbers in that magnitude, and we had a bunch of questions about it. Now it's sort of nice to see the rest of the market catching up and going higher than the number.

Could the number be $120 billion out in that time frame, certainly, we are seeing literally not only an explosion of CPU demand, but 1 of the areas that we're seeing growth in terms of CPU is number of cores per CPU.

Many of these agents want to run independent jobs or flows or a batch on a specific CPU core. So again, the ARM AGI CPU is 136 cores, which is much larger than many of the competitive offerings in the go-forward space, could I see a world of 256 cores, 512 cores absolutely. It's a great place for ARM because in a very, very high core count design, what really matters is efficiency per core, and that's where we're world class.

As far as the market share numbers, AMD has 50 -- Intel has 50, and we have 50, so you add up to some crazy number. All I would say is as follows: we see a very, very strong direction with all of the arm hyperscalers who use our technology today. As Jason mentioned, whether it's NVIDIA, whether it's Amazon, whether it's Google, the very largest and most prevalent accelerators by volume is the TPU, it's trainium and it's Ruben today, Blackwall going to Ruben, those all connect to arm and increasingly, they are going to be a 100% arm.

So we feel very, very good about the market share there. We also talked about a lot of customers such as Cloudflare such as Meta, such as SAP SK Telecom, OpenAI, customers who simply aren't going to design their own ARM-based CPU, they either for CapEx reasons or engineering reasons for whatever the reasons are, they're not going to go off and make the spend there.

So we think it's a market that we can play in, in a very large way. And I think even indicators of AWS selling Graviton to outside partners -- it's kind of an indication that there's just huge, huge demand for ARM-based capacity. So we think we're going to play alongside our partners in this space. And we also think the opportunity is very, very large for both. And I'm actually confident that by the end of the decade, I believe the largest market share by CPU type will be ARM.

Our next question today comes from the line of Timm Schulze-Melander from Rosharon.

so Rene, maybe just to start with you and to key off that CPU TAM commentaries you just made there. I just wanted to check that I heard you right, that you anticipate 100% attach rate of Arm CPU with those accelerators you mentioned?

And then maybe just looking forward, from an OpEx perspective, as you get into that merchant market as your products attach to some of your partners' products. Do you have any undertakings in terms of operating expenses in terms of in-market customer support? And then I had a quick follow-up for Jason.

Yes. Thank you for the question, Tim. Yes. So to clarify my comment, my expectation is that for the training platform over time, TPUs over time, and NVIDIA's accelerated over time, I believe that the vast majority of the market share there will be ARM. NVIDIA is there essentially and we are starting to see that happen with Graviton already over the last number of quarters.

And the announcement that Google made at Google -- next with the and AI, the training and inference chip. So that trend is well underway. And the reason for it, as stated is that by getting much better performance in the same power envelope, the overall performance of the platform has greatly improved.

Google is talking about an 80% improvement in terms of the overall performance. So it's really numbers like that and the advantages that customers see in terms of embracing the platform. That gives us very, very high confidence that, that trend should continue. If I understood your question correctly on the OpEx side or customer support issues, et cetera, et cetera, relative to ARM-based server racks being bought by customers.

What I would expect it would be a relationship that we would typically have with our partners with the ODMs where they're building finished racks and design, the customer is responsible for the software. We're not delivering the application software, but anything regarding the low-level code, the firmware, the boot on, that's all on us. So we'll have customer support ready for that.

And then, of course, if there's any issues regarding the hardware, we will manage that all ourselves. Relative to the OpEx associated with that inside the company, I'll let Jason bridge on to that. But that's all baked in, in terms of when we talk about the growth rate of headcount. We've taken all of that into consideration relative to what customer support platform solution looks like?

Yes. In terms of the OpEx for support, that was baked in the numbers that we shared at Armour and embedded within our long-term guidance and so already accounted for.

Got it. And maybe the quick follow-up, just looking at FY '27 for the year. Could Jason, could you just give us some help in terms of the shape of the of how you expect royalty revenues to kind of flow through the year? And maybe some commentary around the OpEx as well would be really helpful.

Yes. So I expect that the royalty growth is going to be in the roughly 2% for the year. And so across each of the quarters, plus or minus spending on the quarter, but we'll all be relatively close to the 20s percent range.

And that's really for both license and royalty. The license will be just like the last -- I guess, the last 3 years, will be a little bit back-end weighted. I think it's about 60%, probably second half versus 40% points. And then on OpEx I know we initially had thought maybe the OpEx would step up a little more into -- from Q4 to Q1 as we just guided to, we're expecting that to be a little less than we thought before.

And now I would expect that the license -- or I'm sorry, that the OpEx growth is going to is going to grow sequentially every quarter by, call it, a few percent. And the overall expense will still kind of be -- we'll show incremental margin improvement throughout the year. and we will be growing expenses less than revenue by the time the end of the year, and it will be building kind of slowly throughout the year.

But we will be getting back to delivering incremental margin by end of the year, like where we're at, we were at a couple of years ago before we started the investments in the --

Our next question today comes from the line of Chris Sankar from TD Cowan.

Any other question on the CP to GPU ratio. It seems like with inference as a compelling case for AGI, CPU, I'm wondering what is your assumption on when we get to 1:1? And do you expect the CPAP ratio across one-to-one and also, if you can differentiate between the opportunity and the head node versus the host note, that would be very helpful.

Yes. So I think it's a kind of a complex question in terms of thinking about today's world and statically how things look. The way I think to think about it is while the ratios may not go to more CPUs than GPUs from a chip standpoint, they probably will from a core count standpoint.

And what do I mean by that? The way to think about Blackwell and Ruben and some of these large accelerators is that they're pretty much reticle limited, meaning that the size of the chip is already limited by the amount of area that a mass can print -- so it's not like you're going to get many, many more GPUs and then 1 can argue how efficient those GPUs are as they consume all that silicon.

On the flip side, today, the RMA GI CPU, for example, it's 136 CPU cores, Vera, that's 88. As I mentioned earlier, could I see those core counts doubling or quadrupling over the next number of years? Absolutely. Does that mean that, oh, the ratio of chips stay the same, if 1 chip has 500 cores and it used to have 136 cores. So clearly, the ratios are going are going to change from a CPU core count, maybe not a chip count.

Where we'll see the growth, in my opinion, is not so much in the head node to a GPU architecture because it's a little bit fixed given the way the GPU is architected and how it feeds to CPU -- but will you see many, many more CPUs inside of data hall, dedicated racks of CPUs that are doing agenetic orchestration and scheduling and management 100%.

And you simply just have to look at Infinia announcing a dedicated Virirec, 256 Vera CPU chips, 88 cores per chip in a 200-kilowatt liquid cooled rack, that is designed to sit in a data center adjacent to a verarubin system. And that's simply because of the size of the system. It's liquid cooled.

So imagine a world where you had scores of vararubin racks, now you may actually have a veriracin between or 2 varies -- so that changes the ratios completely. I think 1 thing we know for sure is that we probably have undercalled the CPU demand in terms of the transition here.

We talked about a 4x increase we could get our heads around a bigger number than that. But ratios are a tough way to look at it, just given the math that I just described.

Now take our next question. And the next question comes from the line of Sebastien Naji from William Blair.

It's great to see the strong demand for your AI CPU. One of the questions that is top of mind for many investors is how this shift in your business model is impacting some of your existing IP customers who also sell ARM-based CPUs in some form -- could you maybe just give us a sense of how your larger customers have reacted since your announcement in March? And how you think about managing that potential tension between your product and IP businesses.

Yes. Thank you for the question. It's a super important question. One of the things that we wanted to ensure relative to the overall strategy of selling silicon, is that we had the support of the ecosystem.

Now for us, the ecosystem represents a lot of different partners. The ecosystem are the people who build Arm-based chips like Samsung or TSMC that ecosystem is EDA partners like Synopsys and Cadence, the huge amount of people who work in the software space in the Linux world and everything around Kubernetes containers, for example, and then, of course, our licensees, AWS or Microsoft or Google or NVIDIA in this case, all we have products, so we went to them early on this, and we explained to them what we were doing. We explained to them why we were doing it, and we explained to them why it was beneficial to the ARM ecosystem that we do it. And that's largely because the more software that's written and optimized for ARM makes everyone stronger.

And we asked for their support of the strategy at the Arm Everywhere day.

And every single partner we asked said yes. We had probably over 50 different partners, the names I just gave you, all of them. In some cases, they provided a quote. In some cases, they provided references to other partners.

And then in some cases, they did a video for us, which we used at the event. So I think by the fact that everyone was asked and everyone said yes, and how can I help? I think it's about as good as an endorsement I could give for that. So we're -- we're deeply grateful for it. We don't take it for granted. We're very, very appreciate of it. And we think, again, it lifts all boats.

And probably the last thing I would say on this is that the primary reason we did this was that our customers asked for it. And at the end of the day, we are responding to customer demand in a market, and we see this today because we're sold out and we've got people looking for more products, there is demand for these products.

And at the end of the day, customer demand speaks volumes.

Our next question today comes from the line of Vijay Rakesh from Mizuho.

Great quarter and guide here. Rene, just a quick question. Actually look at look at the demand on agent CPUs. And obviously, the CSP seem to be ramping up on that as well. How is your data center royalty revenues growing, I guess, as you look at fiscal '27 or last year? I'm sure that's picking up, but any color on that?

The royalties associated with customers building chips based on neovers, those royalties have doubled year-on-year. and I'm looking at Jason , I expect they're going to double year-on-year again in this year. So the answer -- the answer to that statement is yes. So that business is incredibly strong. And when we talked last quarter in February and I said that this was going to be our largest business.

I was speaking about it in the comp so the royalties were. So now when we add the ARM AGI CPU business, we have 2 extremely strong sources of revenue that I think the best way to think about it is, don't cannibalize each other. They're going to run in tandem with each other. We're going to have a strong business around the RMA GI CPU.

We talked about $15 billion by fiscal year '31. And then we also have an IP business that we expect to double to $10 billion. And that IP business will largely be driven by data center prices.

Got it. And then on the licensing revenue side, should we still think of it as a high single-digit percent growth fiscal '27, '28? Or is that trending higher to with 9 and

Well, for this year, I think we said you should expect license revenue to be more in the 20-ish percent range. Over time, I would expect I think the long-term target is probably in the high single digit, low double-digit range.

But it's hard to say to say. We've seen this kind of AI investment, super cycle or whatever folks are calling it has now gone on for 3 years. So who knows how much longer it goes, but it's at least going to happen for the next year. and beyond that, hard to say.

But I would expect at least at least 10% year-on-year growth for the long term is probably putting the floor that we can see for right now.

We will now take the next question. And the question comes from the line of Harlan from JP Milan.

And congrats on the expanded AGI demand profile. On the $1 billion of revenues for AGI across '27 and '28, most of it obviously is going to be in '28. I think at the event, you told us gross margin profile on these first gen products is about 30% plus. I know it's embedded in your total cost structure.

But Jason, can you just give us a rough sense on the OpEx attributed to supporting your chip business this year and next year? I know it's early revenues, but I'm just trying to figure out like the scale of the chip business cost structure. And as you drive your chip revenues higher, like what -- when can it drive accretion to the earnings power? I know the calendar '30 endpoint, I think, is roughly in OpEx for the chip business, but where is it distort next year?

Yes. I would -- it's a good question. I would say, yes, the revenue split for '27, '28, something like million for Q4 '28 and then $910 million or whatever for '28 million. That's kind of what we laid out 5 or 6 weeks ago. And as said, we have demand above that. But for right now, let's just assume that's the number until we work through some of the wafer memory shortage issues.

In terms of the OpEx, yes, so the OpEx that's in our plan and what we've guided to, that does include the support related to that business. If you kind of just look at that business on a stand-alone basis, we see it as a business that because a lot of the cost of development is really -- probably the most expensive part of developing the chip is really the compute die, which really is effectively kind of the CSS.

So we're able to kind of leverage that, that automatically makes this business much more profitable as a stand-alochip business given that we get kind of that work as part of our IP business already. So the incremental costs and OpEx are apt add to the chip business is not that significant. It's a team that's in the -- probably in the dozens of people, not hundreds.

And so you can assume that it's operating profit positive next year, -- and when you go out to say, 2031, I think we said the IP business probably gets to something like a 65% operating margin business or EBITDA margin business and that the chip business is probably in the 35-ish percent range, and that's where we expect to get to. How quickly we get there is probably going to be a function of how fast the revenue grows and at least at a $15 billion rate, I think those are the right numbers based on everything we see right now.

Thank you. We will now take our final question for today. And the final question comes from the line of Lee Simpson from Morgan Stanley.

Great. at the end here. I just wanted to go back to the ratio of CPUs to GPUs particularly as it relates to orchestration, -- and I think, Rene, you mentioned it's better to look at this really from a core count perspective, particularly when handling these agentic sessions. I'm just trying to get a sense from a bottom-up perspective -- are we looking at this as each agentic flow requires 1 core -- or could we look at this differently and say that there is an average number of ARM instructions that are needed to orchestrate each token generated by the accelerator.

Yes. Thank you, Lee. You've gone into deep math there. I think the latter is a little too complicated to think about it. Maybe a more straightforward way to think about it is -- each of these agents are running a batch or running a job themselves.

There is certainly a level of complexity in terms of the way the branch prediction coding is handled and essentially the way you would code the example. But if you just think about the nature of an asynchronous workload for an agent, it runs a job, it does some scheduling it stops, it waits, it pauses. It's actually pretty good for a single core design to handle that as opposed to having multicores having to run that all together in unison. It's going to be more power efficient if you run it through a single core.

And the more cores you have, in theory, the more batches you can run. So our viewpoint is very much 1 of more cores is better. And that's why I think you're going to see increasingly larger core counts in the CPU chips. So you'll see more CPUs cores being shipped.

You may not see 3x to chips, but the chips will be more expensive, which is why when people look at $100 billion TAM in 5 years or $125 billion, whatever the number is, largely, it's going to be driven by the fact that the CPU chips are going to have lots of CPU cores which will drive ASP up.

But I think it's a core per core batch job not multiple instructions across multiple cores.

Thank you. I will now hand the call back for closing remarks.

Thank you, and thank you for all the questions. This was an amazing quarter billion approximately in revenue, not long ago as what ARM did in a single year and do nearly $5 billion in revenue a few years after the public offering of the company, when we projected around this time, we'll be doing about $4.5 billion is a great testament to all the work done by arms employees, customers and partners. What we are seeing, to be clear, is unprecedented compute demand and that we are at the center of that demand growth.

The best way to think about ARM's growth trajectory, particularly now that we've announced the ARM AGI CPU is that we have 2 growth vectors that will drive this. The ARM AI chip, as we talked about, with $1 billion of demand in March 24, we're now seeing over $2 billion of demand over the next 2 years, 2x what we talked about, and we're on track to achieve $15 billion of revenue by fiscal '31 and our business around IP, the new overuse IP with CSS has doubled year-on-year. We're projecting it to double again year-on-year.

And we are seeing adoption and acceleration with AWS, Google and Microsoft. Both of these vectors represent a structural growth for ARM that is very, very strong and very sustainable. With that, thank you again for all your questions, and we appreciate all your interest in ARM.

Thank you. This concludes today's conference call. Thank you for participating. You may now disconnect.

[Presentation]

Please welcome Arm's Chief Executive Officer, Rene Haas.

Such a nice warm welcome. Thank you. So welcome to our live stream audience, watching the Arm Everywhere event. I don't think we've ever done a live stream event like this and to the folks here in the audience, thank you so much for coming to the historic Fort Mason.

And you may not know that Fort Mason here in California, was actually an official defense site for the Civil War. And this is where a very famous battle between Alabama, Georgia and California took place.

Now you're thinking to yourself, I don't remember that battle. That's why this area looks so pristine. There actually was not a battle. But it actually was a fort. So I thought that was actually kind of neat. I didn't actually know that. So thank you again for attending a big day for us. We have a lot to share with you. So I'm going to jump right into it.

When we thought about how to name this event and how to talk about our company, we thought Arm Everywhere was really appropriate because one of the things that we're very proud of that we don't always think about in our daily lives at Arm, but it's really quite impactful. It's just the scale of the company and the scale of the magnitude we have.

So when we start looking at numbers, 117 billion, what is the number? That's the total humans ever to live on earth. So that's -- if you count up by all of our calculations, how many people have lived on the planet since inception, about 117 billion. 350 billion plus are the number of Arm chips to have ever shipped. That is 3x the total number of humans who have ever existed on the planet.

So it's not just 1 for every human. It's 3 for every human to have ever lived. 7x the total number of non Arm-based CPUs shipped combined. Just think about that number. And 160 Arm chips for every global household, Mine is probably larger than 160, but 160 is about the average. So that just gives you a sense of the scale of what we've done, and it's really important because it feeds into everything that makes us what we are today and, of course, could not be done without our ecosystem partners.

Now the company's DNA was really born to run off batteries. Company started in the early 1990s. It was a spinout of a British computer company named Acorn. And that company had a mandate to build a chip, and that chip had a couple of requirements. One was it had to run in a plastic package, which back then was really important. And number two, it had to be really low power. The first part was important because of heat.

The second part was important because battery life met everything because this was going into the world's first PDA. So the company, we nailed that. We nailed that objective so solidly that -- and this is a true story, that when the first Arm development board that had the first Arm1 processor was powered up, and these were plugged in now into a back of a wall.

So you had a development board, lots of logic chips plugged into an AC outlet. When the AC outlet plug was removed, the chip kept running. And the chip kept running based upon the leakage current that was coming off all the other chips on the board. So the folks came in the next night and they saw the silico was still driving a signal.

And that is really what for us launched the revolution of smartphones. We were designed into the very first GSM phone for those who remember that Nokia brick on the far edge. But then the BlackBerry, which many of us who had loved, still love. Wish it came back, all the way to the modern smartphones of Android and iPhones. That is where we started in terms of the battery life.

It launched a generation of smartphones. Now one of the breaks we got about 10 years ago was when SoftBank bought Arm. Yes, it was about 10 years ago, it was 2016. And when SoftBank bought Arm, Masa gave us an opportunity now that we're a private company to invest into areas that we were not able to invest in before. And that gave us the opportunity to expand the platform to a number of other verticals. We took everything that we knew about smartphones and then expanded that out into the cloud. We launched Neoverse.

We got our first design wins in the data center. And then we were also able to invest into autonomous, automotive, physical AI. We could not have done that without that 2016 moment. And this is my thank you to Masa for allowing us to do that. We could not have made that all happen. It's paid significant benefits for the company.

However, as good as our products are, as competitive as the platform is for physical AI, for autonomous, for the cloud, it is really what I like to call the ecosystem of ecosystems that really differentiates us. And this is where the partnership really comes to life because that mobile platform that we built cannot happen without the software. And the software layer in the case of the mobile area is iOS, it's Windows, it's Android, it's MacOS.

And then the litany of applications that not only run on the Arm compute platform, but they're highly optimized, highly tuned and allows the partners in the ecosystem to build great products. That formula applies to every vertical that we participate in. It applies to what takes place in the cloud, whether it's Linux or OpenAI or Anthropic and then the platform that runs with it.

And this is why we like to call this the ecosystem of ecosystems because it's not just one vertical. And you can see when we look at the physical AI platform with automotive, same formula. 22 million-plus software developers that are very unique to a vertical, but they leverage a lot across the ecosystem that allows people to get started in other areas. So this is the magic, and this is what is uniquely Arm.

It is what's very, very unique about our compute platform. There's no one on the planet who can serve the edge to the cloud in the way our ecosystem does. Now over the past few years, we've been evolving our strategies largely because we see the demands in the marketplace are around the chips are more complex. The cycle times to build these chips are getting longer, 5-nanometer to 3-nanometer to 2-nanometer means longer fab times, longer peg times.

There's a need to do more and to do it faster. We've traditionally provided IP, IP in a stand-alone form, the CPU, the GPU, system IP. And that has served us well for the first 30-plus years of the company. But as I said, we were starting to see huge demand for the need to go faster, make products better and get time to market sooner. And we introduced something called compute subsystems.

We did this about 3, 4 years ago. We invested very heavily in terms of the engineering requirements to do this. And what this does is it takes all the blocks of IP and puts them together in a finished way, verified, performant, tested that the end customer can then take to market. And in some cases, it shaves a year, in some cases, 18 months off the time of starting design to get into production. It was a very significant investment for us.

We put a lot of effort and engineering into it. But we've already seen massive benefits in terms of the customer base. We introduced this 3 or 4 years ago. Our business model is a license plus royalty. Royalty is the laggard. So royalties start to show up 2, 3 years after we license a product. Already, CSS represents almost 20% of our royalties and growing.

Now that's our evolution. But of course, we're now in an era where everything is different than we knew it before. And when I think about artificial intelligence, and I get a lot of questions when I talk to analysts or media about did AI just come up on us by surprise. And I think back to a time I was in Bletchley Park about 1.5 years ago, and the Bletchley Park is where the original crypto work was done by Alan Turing to help the West against the Germans and World War II.

There is an area there where you can go in the museum and you see papers from Alan Turing about can machines think. I think those papers were written in the 1940s. So the idea of AI is obviously not new. And if you're a sci-fi, aficionado or fan, I certainly was growing up. Arthur C. Clark was one of my favorite authors. 2001 A Space Odyssey, now we have people who weren't even born in 2001, who are here.

I always looked at this and said, of course, this is going to happen. I just didn't think in my lifetime, I would see it at the pace that we've seen it. And for anyone who says this is a bubble and it's going to pass, it may be a financial bubble in the case of investment may slow down and it may be an investment bubble in the sense of the valuations may not be what they are today tomorrow.

But if anyone thinks that this is something that is going to go away, it's a little bit of an ostrich syndrome. This is here with us. And it's really changed how people think about computing. However, somewhere along the way, people kind of thought CPUs were dead. And there was a thought that the only way you handle AI is through accelerated computing, that the CPU's role in the AI world is no longer relevant.

Now if we think about the role of the CPU and what happens in the cloud, now this is the cloud before AI. So I'm going to say it's before that last slide that I showed. Huge growth in compute cloud. We saw growth from AWS, Microsoft, GCP. And the conventional use of the cloud was you type in an answer, you do a search, -- any seats left for the Warrior's game?

I think there are a lot of seats left for tomorrow's game, by the way, I have seen or tonight's game. You got the prompt back. This is the cloud. Very simple, you do search, but CPU is very heavy. So when we look at the growth of SaaS 10-plus years ago, 10, 15 years ago and all the growth around cloud, the CPUs were doing literally all the work. Now when you add the AI cloud, if you will, and now you are a human and you're putting in a prompt into your device, whether it's your phone or your PC, Well, of course, there are still CPUs involved.

The cloud is servicing that request, and that request gets sent for a token, which the accelerator generates and a CPU in that data center orchestrates and sends a token back, the token being a word or an answer that provides the request for the query. So this is all the work that's being done by the AI data center. So CPUs are involved both in the cloud and obviously, they're involved in the AI data center.

And we estimate that in this data center, there's probably 30 million CPU cores per gigawatt. So there's a lot. And data center here is a combination of what sits right in the AI cluster, whether it's your head node to your accelerator or what sits next to a dedicated rack. But the math is basically about 30 million CPU cores per gigawatt, okay? And that is the world that we've seen coming up to about the last year or so or maybe even less.

And what has changed in the last number of months has been this explosion of agents. Agents are essentially tools that act on a request and come back with a full flow of answers. So it's not just a query for an answer, but it's actually work. It's run a payroll task, do a scheduler, go off and write a number of analyses relative to a tool flow and provide me an answer.

And we heard so much about OpenClaw here in the last few weeks as an example, and it's not the only example. Now why is this important? Why am I talking about this? Because as we move to agentic query, the number of tokens per human go up by 15x, if not greater. And if you think about the why of that, it's pretty straightforward.

Agents can generate requests, a, far faster than humans; and b, they don't sleep. They're at a 24/7. So the agents are now pushing these requests into the cloud into the data center and what's happening? The data center is choking. These accelerators, which are very expensive, that generate the tokens now need to send those tokens back through the cloud. Now if we think about what an agent is, an agent is a workflow.

As I said, it's a payroll task, it's a scheduler task. It's asynchronous. It is a lot of work relative to scheduling. That's what CPUs do. That is what CPUs do. That is not a work that can be done by an accelerator. The way to think about this is the accelerator generates the tokens, but it's almost like pushing a dump truck up and someone's got to move all that dirt. The CPUs are the pieces of equipment that move that dirt and Agentic AI only increases that.

So what you see is a huge bottleneck now in terms of flow. So what does that mean? You need more and more CPUs, lots of them. CPUs near the head node, CPUs next to the accelerator rack, more CPU racks inside the data center, you just need more. And by our calculations, and we think this may be a little bit light, goes up about 4x, 120 million CPU cores for that same gigawatt, okay? So in that same profile, we now need 120 million CPU cores.

Now we're trying to put 4x the amount of CPU cores in that same power envelope. Power is precious, obviously. The capital required for it is precious. So trying to put all those extra CPUs into a data center that is already stuffed to the brim with accelerators and CPUs doing the core work, that is a problem.

[Presentation]

Now every tough problem needs a good solution. And we're announcing our first silicon chip that we are selling to customers for revenue.

The Arm AGI CPU. Now this is a big, big deal. And I would love to tell you every feed and speed about the product right now, but Mohammad will kill me if I do that. So we'll go into a lot of detail about the product and how we conceived it and the why. But let me be clear, we are now in a new business for Arm, and we are supplying CPUs as chips. The biggest reason we're doing this is that our partners have asked for it.

But we're also really doing this to solve the problem I just described. As Agentic AI becomes mainstream, all of the work required to make that happen is CPU bound and you need a CPU that has the DNA of being born to run off a battery. So as I said, -- reason 0 is our partners have asked us for it. And one of the partners we work the closest with on this is Meta.

And I'm super pleased to have Santosh Janardhan with me today, who's going to do a better job than I can to tell you why Meta made that choice. Santosh?

Hey folks, welcome. Every year, I try to run the San Francisco half Marathon and they distribute the bits the day before you run right here. I can tell you, it looks very, very different compared to what I guess you're seeing now. So hi, my name is Santosh Janardhan. I lead infrastructure at Meta.

So what does that mean? Well, it means that we traditionally go and custom build and design our data centers, we run it. We custom build our hardware, our GPUs, our CPUs, and we'll get into that quite a bit, the network that connects them. And obviously, the software that sort of binds it all together. It's a fancy way to say that if your Instagram is not working, if your WhatsApp is not working, your message is not arriving, I am the person to blame.

Now if you think through our family of apps, that marks about 3 billion, 3.5 billion users that use our products daily. Every single day, about half of humanity logs into one of our sort of apps and hammers away at it. And as you can imagine, that creates a decent amount of scale. We run a decent amount of the Internet. And we're probably the only hyperscaler that's not a cloud, right?

So if you think about gigawatts of capacity, tens of millions of servers and increasingly, more and more, you're seeing bigger and bigger CPU and GPU, AI clusters. Rene sort of went through that quite a bit. I think it's interesting to go and look at how this has grown over the last years. AI clusters are a fairly new thing, really started sort of post-COVID 2022, '23, just after sort of a ChatGPT came along.

And our initial clusters were pretty small. In fact, when I look back for this, in '23, our initial clusters are about 128 GPUs. That's it. But as you can see, even in '23, we started scaling quite a bit. And as you sort of fast forward, it really started growing. The demand for this has far surpassed sort of what any one of us could imagine it was. We are in the tens of thousands of GPUs stitched together in a single cluster now.

And if I project it forward, and this is the thing I really want to set context, there is absolutely no sign of this slowing down. In fact, it's almost exponential. I only see it accelerating. right? So the demand is exponential. And as Rene was saying, power is constrained. I want to talk a little bit about some of our clusters. That is Prometheus. Prometheus is one of our bigger clusters. It will surpass well over 1 gigawatt by the end of this year.

There's a lot of GPUs, I can tell you. And we stitched together a bunch of data centers, a bunch of tent. That thing you see, the blue colored thing is actually a tent. It's a fancy tent, but still a tent, right? It's weatherproof. It can survive about a category 2 hurricane. But we're putting together all of this, stitching it together with a network.

And so to our developers, to our researchers, what they end up getting is about 1 gigawatt worth of an AI cluster in a single combined entity, which is pretty powerful, as you can imagine. But like I was saying, the demand is exponential to put it mildly. That is Hyperion. It is going to go up to 5 gigawatts in a few years. Most people can't fathom what a gigawatt is. A gigawatt is about 10 Palo Altos, the town of Palo Alto, 10x what it consumes is 1 gigawatt.

This will be 5. That's 50 Palo Altos, right? That's what we are building out. So it's going to go and go really, really big. So why do we do this? At Meta, we have this vision of delivering personal super intelligence for every single one of our users. This means creating models that can go and figure out the most relevant experience, the most engaging experience for every one of you on our platforms.

It means creating a personal assistant for every one of you, right? Now if you have to go and deliver sort of personal super intelligence to billions of people, what kind of systems would that take? We're talking about billions of people each using sort of exact amount of compute over and over. And like I said, over 3 billion users a day, right? This -- if this advances, there you go.

So what does it take? Well, it takes power, it takes land. It takes a decent amount of hardware, software, obviously. And most of all, it takes silicon, a lot of silicon, right? And this is why I think Arm is such a natural partner for us. What we want is a partner who can match our ambition who can match our cadence of velocity of innovation. And what we realized when we're sitting down with Arm is that they codevelop it.

They were as hungry as we were and most importantly for us, we were as power conscious and as efficient as we wanted sort of them to be. This is why while Arm is now the primary co-collaborator and the primary sort of partner, the CPU that we are ending up developing is pretty foundational. It can be -- it's not just a meta CPU. It's not just an Arm CPU. This is something that I think will end up being a foundational CPU for the whole ecosystem.

I think we are at the threshold of something pretty sweet here because you're going to hear more and more about sort of the constraints that data centers are facing. You're going to hear more and more about while the demand for compute is growing, the power is not growing at the exact same curve. So this marriage is -- I think about it personally as a win-win situation, right?

So it's extremely sort of heartening to see Arm moving on from not just being an IP license provider, but actually getting into the game of sort of building something that is production scale and production ready. Exciting times. 2 years, 3 years in the making, but I think about this as the sweetest of things take some time, but we're getting there. Now like I said, we are obsessed with efficiency.

And if you think about one of the biggest appeal that Arm has had over the years, it's power profile. Arm can go -- Rene had this fascinating experience that he was talking about taking 30 million cores instead of 30 million, now making it 120 million and fitting in the same power envelope. But that's one thing. You don't want to compromise on performance, right? This is the thing that I really want to make sure we drive here.

The biggest reason why we sat down with Arm and had this conversation was we want to put in a lot more cores per watt, but we do not want to compromise on the performance piece. That marriage is why I really think it's a win-win situation here. In fact, about 2, 2.5 years ago, we sat down with Arm, we actually first surveyed the market to see was there a CPU that could meet the specs that we wanted. If we met the performance, we couldn't get the power.

If we got the power, we couldn't get the performance. And this is why Arm ended up being such a partner. The ability to scale that Arm gives us when you push in a lot more cores. And if you think about personal super intelligence, if you think about the orchestration that Rene showed, you don't want to starve your CPUs nor do you want to starve your GPUs. That marriage that you end up doing is, I think, that most people are going to realize pretty soon.

Now the design point that we chose for this was something to minimize risk for this iteration. We wanted to make sure we get our first CPU right, get it working out of the box. But this is a multigenerational partnership. I just want to emphasize this. When we look at subsequent iterations of things that are already in the hopper of what we're going to build out, I truly believe that this chip is going to expand sort of the performance on multiple axis.

In fact, this ecosystem is actually going to be awesome. When you challenge the incumbents, you see innovation across the board. That, I think, is what all of us will end up achieving. Now I want to talk about why. I want to take this back, I guess, to why we do this work. But like I say, 3 billion, 3.5 billion people use our products every single day. This means there's your friends messaging each other on WhatsApp. It could be a small or medium business messaging the users on a platform.

It could be somebody going and doing an AI interaction with Meta AI. None of this is possible without infrastructure. Infrastructure has now become -- has gone from being on the backside of sort of technology innovation to being the enabler of technology innovation, right? AI is built on the backbone of infrastructure. So every interaction, every post, every feed, every call is done on the basis of what we build out on the back end.

And at least for us, we're custom building data centers, we're custom building hardware and custom building silicon. That's why Arm, I think, is such a big partner for us because for us, we want to squeeze every bit of performance out of what we build out. We think about optimizing things like performance per watt, performance per gigawatt, and Arm allows us to do that. It allows us to go and increase the efficacy of everything we build out.

Why? So that we can go and serve more users so that we can hopefully improve every one of your lives in some way, shape or form. And that's why I think Arm has been an awesome partner. So thank you, Rene and team. It's been absolutely a pleasure to work with you, and hopefully, we'll do this for years together. Thank you.

Amazing. Santosh, thank you. That was terrific. I have someone else I'd like to ask to join us to also talk about how they plan to use our Arm AGI CPU. And that's Kevin Weil from OpenAI. Kevin?

Thank you, sir.

Kevin, thanks for joining us.

Thank you for having me.

Welcome to Fort Mason. Have you been here before?

I have. There are a few conferences in the past.

Well, welcome. So first off, just tell us and tell me, why does -- why does this launch today matter to OpenAI?

Well, I thought you did a good job painting this. AI performance these days is system performance. And GPUs kind of get top billing wherever they go, but really the CPU is playing an incredibly important role as an orchestrator.

But also, I think as AI becomes more agentic, when you look at a rollout that an agent is doing, it's using it's using tools inside containers, that's CPUs, it's running Python scripts as it does -- as it performs skills. Those are CPUs. So the CPU plays an incredibly important role. And it's really the whole system together that makes it all possible.

Now your role at OpenAI is a pretty cool one, right? You're doing math and science and the stuff that's super compute heavy. And when you think about compute constraints, and I know when I talk to you or Sam or Mark or anyone at your company, it's I need more compute. Yes. Tell us about that.

Yes.

Tell us about that.

That is one of the most common things I hear inside OpenAI. I need more compute. It's kind of the coin of the realm. I mean the root of it is we have more demand from customers. We have more ideas internally that we want to experiment with. We have more things that we want to do than frankly, the industry can keep up with.

And when you get to the bottom of all this, it's certainly it's about silicon, but it's also about power. And so if you have a CPU that can draw less power, it could be just as performant, but use less power, it means you have more leftover for everything else that you want to do. That means more inference and more compute. That means more intelligence. And if there's one thing that I've learned in my couple of years now at OpenAI, it's that more intelligence leads us to be able to build better products for all of you.

The thing that I keep coming back to that I try and remind myself of all times is as amazing as the models are today, -- and every year, I'm blown away by the amount of progress we make. As amazing as the models are, the model that you use today is the worst AI model that you will ever use for the rest of your life. It's the worst AI model you're going to use for the rest of your life.

And a year from now, you're going to be -- you couldn't imagine coming back to the AI models of today because they're getting better at such a rapid pace, which just means there's basically infinite demand for intelligence. So we are not stopping from here.

And in your world, in your new role, where you're looking at verticals that are somewhat untapped today, math and science and things of that nature. When you think about the Arm AGI CPU or more broadly, what does more compute do for you in that space?

Well, I mean, the more compute you have, the more inference you're able to do, the longer the rollouts you're able to do. AI, as we go -- as we're sort of progressing from this world of AI as chat to AI solving harder and harder problems. And just like you or me, when you solve harder and harder problems, you're going to need to think a little bit longer.

So the more important problems we solve as we start to think about things like enterprise AGI, science, you're going to need more compute, which means if you can draw the power that you have, which will always be finite, you can draw that more efficiently. You can do more and we can solve more problems.

And for you personally, what are you most excited about broadly in terms of everything we see going on with AI?

Well, I mean, I kind of -- I think I have the coolest job in the world. I get to work on accelerating science with AI. And you've seen sort of a revolution in the past, even just 3 months with GPT 5.2, 5.4, codecs. I mean it used to be that people said, oh, well, these are just stochastic parsers. They're sampling from a distribution of data that they were trained on, but they can't do novel things.

Now we're seeing every day AI solve open problems in science, in mathematics and physics and biology. We're seeing AI help us understand the nature of the universe. We're seeing AI work for weeks on end using a robotic lab to run 36,000 different experiments to optimize the synthesis of a new protein faster and better than any human could. So it's an exciting world. I think science is going to move faster than ever, and it's all built on the kind of infrastructure that you're providing.

We are grateful for your support. Kevin, thanks.

Thank you so much.

Thank you. I love the idea that the model that we're using today is about as bad as it's going to get. That's crazy. I want to repeat, in case I wasn't crystal clear on the first go around. We are now delivering IP, CSS and chips. IP, CSS and chips. Contact your local sales representative. Will is here. He can be reached afterwards.

Now seriously, I talked earlier about the ecosystem of ecosystems and none of this could be done without the ecosystem that we have, particularly around Neoverse. We have many partners that we work with on the supply side, whether it's around memory or connectivity. But we've also got great customers who use our IP today.

And they are so supportive of what we're doing. Santosh talked about the demand. The market is so large. The demand is so significant that no one company can serve it. So what I'd like to do is rather than me going on and on and talking about it is have you hear from some of our partners and friends who I think you'll probably recognize a few.

[Presentation]

Charlie and Matt and Sanjay and even my old boss did better than I could in terms of talking about this. But this has not happened without a fantastic partnership and support from the ecosystem.

Now I know you are dying to hear about this product as am I. And I'm now going to turn over to Mohamed Awad, who is going to tell you all about the Arm's AGI CPU, and why it is absolutely amazing. Mohamed?

Thank you, Rene. Thank you, Santosh. Thank you, Kevin. Thanks to all of you. Thanks to the entire Arm team that made today possible. We have been looking forward to this, and it is so exciting to be here. It's so exciting to talk to you guys. Thank you. Thank you. Thank you.

Rene talked about how the world is transitioning from sort of legacy data centers to AGI data centers to agentic data centers heading down this path and how the CPU is at the heart of it. We've designed our AGI CPU around 3 simple principles. We believe that's the heart of what we're doing. It's the heart of what we focused on. It's the heart of how we think about it.

First, performance, performance, performance. With this many threads going on, with this much work to do, with this much orchestration to happen, you can't slow down 24 hours a day, as Rene said, these agents are going to be running. And if they're not performing fast enough, then the rest of that infrastructure that's relying on it grinds to a halt. So we focused on performance. Second, we focused on scale.

The scale of what we're talking about here is just incredible. You heard Santosh talk about gigawatts, gigawatts, scale at the CPU level, scale at the Board level, scale at the rack level, scale at the warehouse level, all the way up. We focused on that. And finally, we focused on efficiency, maybe most importantly, because at the end of the day, with this much at stake, with this much compute we're trying to deploy, we're not going to get there unless we provide that performance, we provide that scale and we do it in an efficient package.

Those are the principles that have guided us. Wait for it. Those are the principles that have guided us, and we refuse to compromise. We've designed on all 3. Play the video now.

[Presentation]

I got to tell you, we are so, so proud. Our team has done a fantastic job on this, and it's really been designed for the ground up from this. Let me tell you a little bit more about what you just saw because I know there was a lot packed into that video.

Arm AI CPU starts off with our standard Neoverse V3 compute subsystem. That's the same compute subsystem we make available to the entire ecosystem, and we have other partners building on it. We're incredibly proud of that. We pack in 136 of those cores, which are very high-performance cores designed to be high performance. Our V-Series is our most performant line, and you've seen it set records across lots of different hyperscale implementations and those of other system providers.

We add to that a dedicated 2-megabyte L2 cache, and we support up to 3.7 gigahertz in frequency. But it's not just the CPU core. We thought about the entire system. As part of the design, we went with 96 lanes of PCIe Gen 6, which supports CXL 3, which means you can attach it to any accelerator you like. It also means that you can support things like memory expansion.

On the memory side, DDR5, with up to 6 gigabytes per second of memory per core, which can be sustained to each core. That is unique. That level of performance to every single core on both the I/O and the memory is unique to us in this type of a package, in this type of a performance point at this efficiency level. And it's not just about the bandwidth. It's not just about the I/O.

It's about the overall design. You see we designed the whole thing to be low latency so that you could get to less than 100 nanoseconds of latency from the memory. We did so by sticking with a dual chiplet design, each chiplet having all of the memory in the I/O directly on it rather than having to worry about complicated pneuma domains and multiple hops across the silicon. The result, it wasn't a typo in the slide, 300-watt TDP, 300 watts.

That is amazing. It's built on a 3-nanometer TSMC process and allows for that maximum compute density. This is what purpose-built design looks like. This is what we're so proud of. The AGI CPU is breaking records all over the place for performance, for scale and for efficiency. You saw some of that in the video. This is a standard OCP air-cooled rack. Nothing unique about it, nothing especially exotic about it, just OCP rack, standard, right?

That's our Head of OCP right there clapping, just so everyone's aware. 36 kilowatts, we pack in over 8,000 of these performance CPU cores. We do so by going to a 2-node 1-use server, 30 of them. You can't do that in other systems because the power consumption is just too high. This is setting records for air cooled. But you know what, if you want liquid cooled, we can do that, too.

Over 45,000 CPU cores and a 200-kilowatt, again, a standard rack from OCP. Over a petabyte of memory in this thing. And oh, by the way, fun fact on this one. It's a 200-kilowatt rack. We actually will consume about half that much power. We ran out of space. That's why we couldn't put more cores in there. Yes, it's pretty wild. The scale of this stuff is crazy. It's just really inspiring. These are standard racks, but there's nothing else like them.

To get to this level of efficiency, we really had to design the Arm AGI CPU from the ground up. And that's what I'm so proud about, and I'll tell you about in a minute. But before I get that, I want to just talk about the fact that these are standard racks because it's not only about the fact that we are using -- there we go. It's not only that we're taking from OCP and leveraging some of their platforms, we're also giving back.

We're in the process of making a bunch of contributions to OCP, things like Arm server ready, authenticated access control and diagnostic tools. And those contributions won't just be for the Arm AGI CPU, they will apply to the entire ecosystem. So it will make it available so that -- and they will be beneficial for all Arm-based platforms because it really is an ecosystem that we're building here. Arm has always been about nurturing and partnering with the ecosystem.

That's always been core to our identity. And those relationships are paying great dividends now. You saw the video that Rene played, and we're so grateful about all those partnerships. It's those partnerships actually, which have allowed us to build the Arm AGI CPU. Some of them are very long-standing. Partners like TSMC and Samsung and Micron and SK hynix, these are partners that we've been working with for well over -- for decades, literally for decades.

And we've also got some new partnerships, which is why we're so proud to say that the Arm AGI CPU -- went a little bit far there. Can you go back, please? -- is available now. Said it there. It doesn't say it there. Yes. So Arm available. Arm AGI CPU is available now, and we're so proud of that. It's actually in customers' hands. Customers are actually evaluating it as we speak.

We are ready to go. And we're so grateful for our partners, both on the ODM side, on the memory side, on the CPU side, on the manufacturing side who have helped us get to this point. We'll be in production by the end of the year, and we are excited to share that with you. We've got -- today, we've got firmware ready to go. We've got specifications ready to go. I talked to you about platforms. I talked to you about supply.

The one thing I haven't talked to you about yet is software. Let's talk about software. Now the next slide. Okay. So the reality is that Arm has been investing in data center software ecosystem for well over 15 years. I don't know if everyone understands how long we've been investing in the software ecosystem. For the beginning of that time, in the early days, it was just Arm investing in the software ecosystem.

And then something happened in 2019. We launched Arm Neoverse. And what Arm Neoverse did, that compute platform when we launched it, it allowed our customers to begin to launch products with a much lower barrier to entry. It allowed them to build their own silicon and start to coalesce around a common platform. And that started that software flywheel turning.

You see when tech leaders started adopting Neoverse, they started to optimize software around it. And the more of those tech leaders that adopted Neoverse, the faster that flywheel started to spin. Today, we've got AWS and Google and Meta and Microsoft and Oracle and NVIDIA, all investing alongside us in the software ecosystem. And that really was what allowed us to kind of really make some great traction in software.

Together, we've made Arm a first-class citizen on most modern software packages. And for our AI software ecosystem specifically, not only are we a first-class citizen, not only the software run well on Arm, software actually runs best on Arm. And the reason for that is very simple. For AI, the Arm software ecosystem, the Arm architecture is the primary CPU architecture in support of AI today.

In fact, the work we've done together with technology leaders means that tens of thousands of companies today run their software on Arm in the cloud on over 1.25 billion Arm Neoverse cores, which we've already shipped into data centers around the world. And that growth is only accelerating. That's actually the curve. You see Arm in the data center just works.

This is a key point. And I don't know if I'm making it well enough. So I'm going to bring somebody on stage who's got a little experience with software. Paul Saab has worked on Meta's infrastructure for over 18 years. He's one of the longest tenured employees at the company. There's a laundry list of things that he's been responsible for, including the adoption of flash storage all the way through the implementation of IPV6. Today, he's specifically focused on making AI more efficient in their infrastructure. And that's how we got to know each other.

Please welcome Paul Saab.

Thank you.

Thank you. Thanks for being here. Thanks for being here.

Thank you for having me.

You've told me the story before, but I really want to hear you guys have had a long history with Arm. -- goes back longer than just a couple of years ago. Can you maybe give everybody a little bit of a history lesson as to kind of how things started?

Yes. I think it was like 2014, 2015, we were looking at Arm. We were really excited about the efficiency wins that we were seeing. We were really back then just targeting our hack/PHP platform called HBM.

And it was working great. Like we made it work. It was performant and then the market kind of went away for us. We didn't really have a platform anymore. And so we just sort of tabled it. And we ripped all that code out. Everything in the code base was removed.

Okay. So that was 2014 and 2015. Obviously, something must have changed or you wouldn't be standing here today, right? So kind of where do we go from there?

Well, the story is kind of funny is like we were -- it was like post -- we're just coming out of the COVID bubble, and we had a bunch of people over at the house sitting around socializing and whatever. And I turned to one of my colleagues and I said, hey, I want to build -- I want to port the Arm again. I kind of had this gut feeling that the ecosystem in the world had changed.

And if we didn't start then we would be kind of playing catch-up when it actually happened. I didn't even ask my boss here for permission to buy these machines or even to start the project.

It's a good thing he approves now.

I don't really ask him permission for much to do. But yes, so we started -- we decided we found some machines out there. I went to some other colleagues. I said, hey, I want to port to Arm and he actually responded was like I was wondering when you were going to ask me. So we got the machines in, started porting, making great progress, but it was super slow.

We only had 8 machines. But we had this vast x86 ecosystem. And I went to the guys and I was like, hey, can we cross compile? And that's what we ended up doing. We ended up like working around the clock. It took us about 90 days, 5 engineers, and we had a full complete port, full system ready, but then we ran into another problem. We had no silicon to buy.

And this -- and Santosh referenced this that like we looked at every partner. And I think this is about the time you and I started talking.

So you'd say the market was a little bit underserved maybe for what you guys were.

I think underserved is an understatement.

But let's go back to the 90 days, 5 people. I mean, really -- it's okay. So I'm going to take your word for it. It was 90 days, 5 people, but that's just getting the code working. Like now you've got to now operationalize and get it performing. Like how is that going?

It's still a small team. I mean it's a lot of very devoted people bringing the systems up. From the time we finished that initial port in 2022, it took us about 2.5 years to actually get some sort of production worthy systems in that were TCO effective, performance per watt. And it was still a very small team.

And even today, it's really a small team that's focused on hyper optimizing. It started off with -- once those performance systems landed, it was really just one engineer until a few more came in. But that engineer never had written a single line of Neon, never written a single line of SVE and single-handedly took some of our most precious workloads and made them work on Arm.

And how is it performing now generally, like on typical workloads? Like how should we think about the performance in the general?

We're seeing performance that is basically equal to anything you can buy in the market today at massive performance per watt improvements.

That's great. That's great. Okay. My light is going to start blinking in a minute here. So I'm not going to keep you on stage too long. But I just -- first of all, I want to say thank you. But before I let you go, I guess one question for you.

If somebody is out there thinking about, hey, because there are tens of thousand companies that are using Arm already, but there's still a few that aren't. What sort of advice or guidance would you get? And what would be your kind of recommendation to them?

I think small focus teams doing the port. But like if I were starting the port today, I would be using an LLM. I mean what we're -- what I'm seeing some of the engineers that are now optimizing even existing Arm accelerated code, they're using LLMs to even boost those by 10% or 20%.

So the barrier to entry today like porting to Arm is, I would say, close to 0. because like the LLM is just going to do it for you. I don't even write any handwritten code anymore myself. It's just all LLM, all test cases all across the board. So like there's no excuse to port to Arm today.

Excellent. Thanks, Well, that was inspiring. I mean, Paul and I have obviously known each other for a little while. And the tenacity and what I hear around is once Paul gets something in his mind, it just kind of happens. So I appreciate all the support, Paul. Thank you. We're so part of the partnership that we've had with you and with Meta more broadly. So thank you very much.

What I love about that story is that they had a need, the market was underserved. And together, we worked together to go address it. The reality is the opportunity for the AGI CPU is broad. The software is ready, and we have a great product. And that's why we're seeing such great customer traction. We're seeing it in multiple areas.

If you think about companies like Cerebras and Positron and Rebellions, they're joining Meta and OpenAI by using Arm AGI CPU for things like managing head nodes that they're building or managing accelerators they're building, so a head node type use case or also for agentic orchestration and fan-out.

These are specific use cases that they're looking at. And then in the cloud, we see companies like SAP and SK Telecom and Cloudflare who are actively using or planning on deploying Arm as part of their infrastructure. These are just a few of the customers that are planning on using Arm AGI CPU. But rather than me tell you let's listen to them.

[Presentation]

I just want to say thanks again to all of our customers and some partners that are supporting us here today. The support we've gotten has really just been incredible. We built Arm AGI CPU for you, and we're so pleased with the response. You see Arm AGI CPU has been designed from the ground up to make sure that performance scales and power stays predictable.

That's the superpower, performance, scale and efficiency. And it's resonating with our partners. You see that's a very different approach than it's taken by x86. They've burdened their execution -- they are burdened with execution overhead and legacy feature support. They chose to focus on things like modularity, support for lots of different markets and esoteric use cases. We are ruthlessly focused on improving efficiency and reducing latency.

Ultimately, this is about architectural philosophy. We're not strapped to the past. We are not strapped to the past. Listen, we don't support Lotus Notes, okay? We just don't do it. We're focused on exactly and only what the AGI at data center needs, performance, scale and efficiency. Let me take you through that in a little more detail. It starts with performance. And performance for us is all about doing more work for every clock cycle.

This has always been an area. Great IPC has always been an area where Arm has shined. How much work do you get done every single cycle? Our AGI CPU absolutely shines here. Now what we see is that legacy CPUs, they sometimes try to compete on this vector by doing things like increasing the frequency, going to boost modes. But here's the reality. When you increase the frequency, what else do you increase? Power.

That's a problem. These boost modes are not sustainable across long periods of time. They're not sustainable across a chip. With Arm AGI CPU, what we give you is full performance sustainably all the time. And ultimately, that means scale. We linearly scale across cores, and our memory and I/O subsystem is specifically designed to be matched to those cores so that we can continue to feed them, 6 gigabytes per second of memory bandwidth to every single core. In order to scale, what we see some of these legacy architectures do is multi-threading, right?

What happens when you do multi-threading? You throw 2 jobs at the same core. That's how they get to a high thread count or try to get to a lot of devices. But the reality on that is your I/O and your bandwidth, that doesn't double. So you've just moved the bottleneck elsewhere. And oh, by the way, the CPU needs to be burdened with managing that back and forth. And so your performance degrades, you end up starving your processes.

What we see over and over again is that data center operators have to overprovision their data centers by 30% or more to deal with this lack of nonlinear scaling. This is an actual thing that happens. We take pride in not having to do that. There's actually a great demo of this out on the show floor. I encourage you all to check it out after the keynote. And then finally, we have this maniacal focus on efficiency.

Obviously, that's always been Mark's -- that's always been Arm's hallmark. It's always been something that we've been great at. We're leveraging all those techniques and methods and experience that we've built up over the decades around building incredibly efficient processors, incredibly efficient technology. And we're packaging that all up in a custom design specifically for this use case. AGI CPU is purpose-built without that legacy overhead because it all comes back to performance, scale and efficiency.

It's my efficiency bullet. At the end of the day, no wasted cycles, no stranded compute, no wasted power or silicon, and we're super proud of that. Let's look at what it means in practice. I'm going to show you the results, and they kind of speak for themselves. First, let's talk about sustained performance. What you see here is the performance that you can expect to achieve consistently. So this is consistent performance.

No performance throttling because you're over power budget, no memory or I/O contention. This is the sort of performance you're going to see. You can see with AGI CPU, it's world-class. You've got world-class performance, you can take to the bank. Next, let's talk about scale. How many threads or agents can you run in each rack? How much compute do you actually support with a fixed power budget with a fixed physical footprint. Remember those racks I showed you earlier, there you go.

That's where we land. And of course, there's efficiency, performance per watt. What's going on with my screens. They're flipping all over the place. Can you go back, please? Go back one more. So what you're seeing here, these are -- all of these charts are with SMT disabled. So these are single-threaded cores for us, single-threaded cores for them. So no multi-threading whatsoever, okay?

I told you what I thought about multi-threading, which is why we elected to show it to you this way. But oftentimes, what we hear is that multi-threading is going to improve that middle chart. It's going to allow for more scalability. Multi-threading is going to improve the performance per watt. Let's take a look at what happens if we turn multi-threading on, okay?

See, first of all, your performance goes down. That's the chart on the left. And the reason why the performance goes down is because you can't just add more work and expect performance to be the same. So that's pretty self-explanatory. And in this particular case, again, we've held it at kind of based on the memory and the I/O bandwidth available, kind of where you land. That second one, the sustained threads per rack.

The reality is that because of the limitations on the device and all of the bottlenecks, you end up in a scenario where you can't actually use all of those threads. Many are left idle. And then finally, performance per watt. Yes, there is a small improvement there, but not enough to change the calculus. At the end of the day, the results are clear.

This is a killer product and Arm is a class of its own. Performance, scale and efficiency. I'll say it one more time. This is what the Arm AGI CPU is built for. And the impact on the AI data center is going to be profound.

Let me turn it back to Rene. Thank you.

Thank you, Mohamed, and thank you, Paul, and your LLM agent that's going to do all the conversions for us. So we've shared a lot with you today, and I'm grateful for your patience and time. If there were just a few things to take away from this morning, I think it starts here. performance per watt, which translates to performance per rack. When you look at an x86 equivalent structure, same power delivery, 36 kilowatts, 2x the performance in the same power.

That's what you need to remember. For those of you who are paying for that power, there's another number you need to remember. If you think about 1 gigawatt of capacity and you think about the CapEx associated with that extra power you're spending at the sake of performance, it's up to $10 billion of CapEx. Obviously, these are serious numbers.

So again, the takeaway from the Arm AGI CPU is 2x performance per watt, probably more than 2x -- now you heard a number of comments in the videos, including Santosh, that when you embark on the kind of engagement and partnership we're talking about, while a day like this and an event like this is wonderful and amazing and we're talking about a great product, it's really not about the day, but it's about the future and commitment to a road map.

So we are committing to future generations of this product. Arm AGI CPU 2 is coming out soon as is Arm AGI CPU 3. As you heard in the videos again, these are multigenerational engagements. We are investing a lot. Our customers are investing a lot. The ecosystem is investing a lot. We are absolutely committed to a road map and a future around this product line. In addition, we will continue the CSSs around these products.

And as Mohamed mentioned, one of the big benefits of the CSSs are the speed it allows our customers to get to market. It also enables a lot of benefit for us as well. So the CSS road map will continue. So I want to close a little bit around what we think the financial opportunity is for Arm. So before this day, our business has been IP and IP compute subsystems.

And we have been doing extremely well in that business, far better than what we had talked to investors about 2.5 years ago when we did our roadshow for the IPO, we're actually ahead of that. When we look at the AI data center business, that represents today about a $3 billion TAM. And now I'm just talking about roughly the royalties. So I mentioned on one of the earnings calls that the cloud AI business will probably be our largest business in a few years.

And this is really driven by all of the growth that Mohamed talked about, the deployment of 1.25 billion Neoverse cores and forward. When we think about our business going forward, the Arm AGI CPU and as Mohamed mentioned, we have committed customers, Meta, OpenAI, Cloudflare, SAP, F5, customer you saw in the video. When we think forward about what is the market opportunity for this business, it is a dramatic sea change for the opportunity.

When we look at what's going on with Agentic AI, the growth of CPUs, the benefit that power-efficient CPUs bring to the data center, we think this represents about $100 billion TAM for us in the future. So today, it is all about the Arm AGI CPU. But there will be some tomorrows. And don't ask me about tomorrow today, but there will be some tomorrows.

And we think this opportunity to take the work we've done across all of the markets, as you've heard in the video from edge to cloud, from milliwatts to gigawatts, we think we have an opportunity to address greater than a $1 trillion TAM by the end of the decade. So we've got some work to do, but I couldn't be more proud of what our company has achieved, grateful to the ecosystem that helps us achieve it and the customers that are now committed to buy our product.

I want to close by saying that we stand on the shoulders of our ecosystem. None of this is possible without the ecosystem that we have nurtured for 35-plus years, many of you who are here today and watching on video. Thank you for attending today. Arm is everywhere, and we appreciate your support.

Good day, and thank you for standing by. Welcome to the Arm Third Quarter Fiscal Year 2026 Webcast and Conference Call. [Operator Instructions] Please be advised that today's conference is being recorded.

I would now like to hand the conference over to your first speaker today, Jeff Kvaal, Head of Investor Relations. Please go ahead.

Thank you very much, and welcome to our third quarter fiscal '26 earnings call. On the call are Rene Haas, Arm's Chief Executive Officer; and Jason Child, Arm's Chief Financial Officer.

During the call, Arm will discuss forecasts, targets and other forward-looking information about the company and its financial results. All of these statements represent our best current judgment about future results. Our business is subject to many risks and uncertainties that could cause actual results to differ materially. In addition to any risks that we highlight during this call, important risk factors that may affect our future results and performance are described in our registration statement on Form 20-F filed with the SEC. Arm assumes no obligation to update any forward-looking statements.

We will refer to non-GAAP financial measures during this discussion. Reconciliations of certain of these non-GAAP financial measures to their most directly comparable GAAP measures can be found in our shareholder letter, as can a discussion of certain projected non-GAAP financial measures that we are not able to recognize -- reconcile without unreasonable effort and supplemental financial information. Our earnings materials are available at investors.arm.com.

And with that, I'll turn the call over to Rene.

Thank you, Jeff, and welcome, everyone. Arm delivered a record third quarter. Revenue grew 26% year-on-year to $1.24 billion, our fourth consecutive billion-dollar quarter. Royalties increased 27% to a record $737 million, driven by record units with strength across AI and general-purpose data center. Our data center royalty revenue has grown more than 100% year-on-year, and we expect in a few years, our data center business to be our largest business, larger than mobile.

License revenue was $505 million, up 25% year-on-year as more leading companies signed high-value licenses for next-generation technologies. That performance lifted our non-GAAP EPS to $0.43, even as we continue to increase R&D investment.

Our performance this quarter reinforces the strength of the Arm platform and our continued commitment to investing in innovation across a broad spectrum of compute technologies. The fundamentals of the Arm business have never been stronger.

AI is changing how compute is built and where it runs across cloud infrastructure, edge devices and physical systems. The industry requires platforms to deliver high performance, energy efficiency and flexibility across a broad range of power envelopes and use cases. Only Arm's compute platform can address these demands, supporting AI workloads ranging from milliwatts to gigawatts.

To align with how our customers deploy AI, we've organized ourselves around 3 business units: Edge AI, Physical AI and Cloud AI. Edge AI comprises the smartphone and IoT businesses, Physical AI includes automotive and robotics, and Cloud AI encompasses data center and networking.

A key driver of our royalty momentum is Compute Subsystems, or CSS. We launched CSS nearly 2.5 years ago, and demand continues to exceed expectations. This quarter, we signed 2 additional CSS licenses for Edge AI tablets and smartphones, bringing us to 21 CSS licenses across 12 companies. Five customers are now shipping CSS-based chips, including 2 shipping a second-generation platform. And the top 4 Android smartphone vendors are shipping CSS-powered devices.

CSS helps customers get to market faster by lowering integration risk and complexity. As demand scales, it increases the value that Arm delivers per chip, creating a significant tailwind to royalties.

In Cloud AI, the shift towards inference is reshaping data center design. And increasingly, that inference is agent-based. These workloads are persistent, always-on and power constrained. This is a fundamental change in how AI systems operate. This is because agent-based AI requires coordination across many agents running continuously, and that the CPU can only do coordination. As this model scales, customers need CPU chips with higher core counts and better power efficiency to operate continuously within tight power and cost constraints.

This trend directly benefits Arm. Arm-based CPU chips deliver industry-leading performance per watt, enabling customers to scale core counts and run always-on AI workloads. We're now seeing this trend play out in the market where Neoverse CPUs have surpassed 1 billion cores deployed, and Arm's share amongst the top hyperscalers is expected to reach 50%.

Leading hyperscalers are launching new products with increased core counts to address this opportunity. AWS launched its fifth generation Graviton processor with 192 cores, doubling the core count from Graviton4, and delivering 25% higher performance and up to 33% lower latency versus Graviton4.

NVIDIA's next-generation Vera CPU features 88 Arm-based cores, up from 72 cores in the Grace CPU generation.

Microsoft introduced Cobalt 200, built on the higher-performance Arm Neoverse CSS V3 with 132 cores, up from 128 cores in Cobalt 100, which was based on the prior Neoverse N2 platform.

And Google previewed its second Arm-based server processor with Axion-powered N4A instances delivering up to 2x better price performance and 80% better performance per watt in the comparable x86 offerings. Google has now migrated over 30,000 applications to the Arm instruction set.

We are also seeing more integrated platform designs to improve system efficiency, often translating into more AI output or more tokens per watt within the same power envelope.

AWS integrates Graviton with Arm-based Nitro DPUs and Trainium accelerators, and NVIDIA pairs GPUs with Arm-based Grace CPUs and Arm-based BlueField DPUs with its transition to Vera, delivering a 6x increase in DPU compute capability over the prior generation.

Together, these trends make clear that as AI inference becomes more agent-based, the importance of CPUs is only increasing. And as a result, Arm's role at the center of the modern data center architecture continues to grow rapidly.

Outside the data center, AI is now moving to everyday devices. The edge and physical AI markets are opening up new growth opportunities. These systems operate in real time under strict power, safety and reliability constraints, where efficient and predictable general-purpose compute is essential.

Arm's strengths, power efficiency, predictable latency and always-on operation are best suited to on-device agents that continue to monitor inputs, prioritize tasks, and invoke models when needed to preserve battery life. Our common software foundation across devices, vehicles and robotics let customers scale deployments without rebuilding software stacks.

We now see that momentum in customer innovation. Rivian announced its third-generation Autonomy Computer based on the Arm-based Rivian Autonomy Processor. The first production vehicle based on a custom Arm chip and the first to deploy Armv9 in production car. Tesla's upcoming Optimus humanoid robot is also powered by a custom Arm-based AI processor, and platform from leading silicon providers like NVIDIA's Jetson Thor and Qualcomm's Dragonwing platforms are scaling Arm-based solutions across robotics and autonomous systems.

To close, AI is moving to every environment and every powered envelope. Arm provides the foundation for that shift, a platform that spans milliwatts to gigawatts, and a developer ecosystem over 22 million developers, more than 80% of the global total. We are now seeing the results of strategies we put in place years ago, focusing on the data center, power efficiency and compute subsystems. As a result, as more and more applications move to agentic AI, Arm will be the compute platform connecting cloud, edge and physical AI use cases.

And with that, I'll now hand it over to Jason.

Thank you, Rene. We have delivered another strong quarter. Total revenue grew 26% year-on-year to a record $1.24 billion, marking our fourth consecutive quarter above $1 billion.

Royalty revenue exceeded our expectations, growing 27% year-on-year to a record $737 million. The biggest growth contributors were smartphones with higher royalty rates per chip, and in the data center, where our revenues continue to grow triple digits year-on-year as we see ongoing share gains from custom hyperscaler chips.

Royalty revenue from Edge AI devices such as smartphones continues to grow much faster than the market. All the major Android OEMs are now ramping smartphones with chips based on both Armv9 and CSS.

In Cloud AI, data center royalty revenue continues to double year-on-year with the ramp of Arm-based chips by all major hyperscaler companies. We are getting a further benefit as the build-out of these new AI data centers are driving increased deployment of networking chips, particularly DPUs and SmartNICs, where Arm has a very high market share.

In Physical AI, the automotive market grew double digits year-on-year and contributed to our strong royalty performance. Overall revenue -- royalty revenue growth continues to reflect Arm's increasing royalty per chip and rising market share.

Turning now to licensing. License and other revenue was $505 million, up 25% year-on-year. Growth was driven by strong demand for next-generation architectures and deeper strategic engagements with key customers. We signed 2 new Arm ATA or Arm Total Access agreements during the quarter and 2 new CSS licenses, both with leading smartphone handset OEMs. These agreements reflect the continued investment by our customers in our next-generation Arm technology.

Of the $505 million of license revenue, our agreement with SoftBank or technology licensing and design services contributed $200 million.

SoftBank has become an increasingly important customer as they build out their AI compute strategy, including their recent acquisitions such as Ampere and Graphcore. We believe that the revenues we are receiving from SoftBank are durable as they relate to current generations that will continue as SoftBank executes on its road map.

As always, licensing revenue varies quarter-to-quarter due to the timing and size of high-value deals. So we will continue to focus on annualized contract value, or ACV, as a key indicator of the underlying licensing trend. ACV grew 28% year-on-year, maintaining a strong momentum following the 28% year-on-year growth we reported in Q2 and Q1. This continues to be above our long-term expectation of mid- to high single-digit growth for license revenue.

Turning to operating expenses and profits. Non-GAAP operating expenses were $716 million, up 37% year-on-year due to strong R&D investment. These investments in R&D reflect ongoing engineering headcount expansion to support customer demand for more Arm technology, including innovation in next-generation architectures, compute subsystems, and into our exploration into chiplets and complete SoCs.

Non-GAAP operating income was $505 million, up 14% year-on-year. This resulted in non-GAAP operating margin of about 41%.

Non-GAAP EPS was $0.43, close to the high end of our guidance range, driven by both higher revenue and slightly lower OpEx than expected.

Turning now to guidance. Our guidance reflects our current view of our end markets and our licensing pipeline. For Q4, we expect revenue of $1.47 billion, plus or minus $50 million. At the midpoint, this represents revenue growth of about 18% year-on-year. We expect royalties to be up low teens year-on-year and licensing to be up high teens year-on-year. We expect our non-GAAP operating expense to be approximately $745 million and our non-GAAP EPS to be $0.58, plus or minus $0.04.

The strength of customer demand we are seeing today, combined with a growing base of long-duration contracts at structurally higher royalty rates, provides increasing confidence in our future revenue profile. This confidence allows us today to invest in next-generation architectures, compute subsystems, and silicon that are needed to enable higher performance, greater efficiency and more AI use cases. We believe this virtuous cycle of customer demand and ambitious investment positions Arm for sustained growth over the long term.

Just before we get into the Q&A portion of the call, as you will have seen, Arm is hosting an event on March 24, and I'm sure there will be interest about what we are planning to announce. There will be a million ways of asking what we may or may not be announcing. Please be patient, as we won't be providing any details ahead of the event.

With that, I'll turn the call back to the operator for the Q&A portion of the call.

[Operator Instructions] And your first question today comes from the line of Joe Quatrochi from Wells Fargo.

Rene, you touched upon in the prepared remarks, but I was kind of curious if you could just maybe give us a little more detail of just how you view Arm's role and the role of the CPU in AI and cloud data centers? And just how does that change as we start to see more proliferation of AI agents?

Yes. Thank you for the question. There is a number of shifts taking place in the data center, as I mentioned in the opening remarks. First off, as the shift moves away from exclusively training to predominantly inference, that is a workload that launches a number of different solution paths. One of them that we're seeing is around agentic AI. And the agents that are actually talking to other agents or having to control workflows such as service tickets or other work streams, those are very, very well suited for CPUs because CPUs are very, very power efficient, always-on, very, very fast latency. And what we are seeing is already an increased deployment of CPUs to address that problem.

Now it's just not CPUs that are good for that problem. It's the number of CPUs you have. And obviously, given the power constraints inside a data center, the efficiency of those CPUs. So for all those reasons, that's a very positive tailwind for Arm. And in particular, we're seeing those proof points now, as I mentioned, where the latest generation of CPU chips from the hyperscaler providers and also NVIDIA have increased the number of cores, and we think that only continues.

And just as a follow-up, maybe one for Jason. I know you're not giving fiscal '27 commentary today, but just how do we think about the puts and takes of this royalty revenue growth and the risks that are associated with the potential like demand destruction that we're seeing in consumer electronics potentially from memory?

Yes. Yes, that's a -- it's a great question and something we spend a lot of time looking at. So in particular, I think MediaTek last night talked about something like around a 15% reduction in unit volume for next year. And that's pretty consistent with what we've heard from other smartphone and handset providers around what they think the memory supply chain constraints could provide.

And so we've done our own kind of analysis of it. And what's interesting is we're hearing from our various partners that they're really trying to make sure that they protect the high end of the market, so the premium and flagship portion of the market, which is great for us because that's where all of our CSS and v9 royalties are, so the highest by a significant margin.

And then on the very bottom end of the segment, that's where most of the supply chain constraints will probably be felt. For us, that's v8 and even older generations that are dramatically smaller royalties. So I think if you were to say, what if there's a 20% reduction in volumes next year, for us, that would translate to probably somewhere around a 2% or 4% at worst impact on smartphone royalties. If you then project that across the whole business, it'd be a 1%, maybe 2% negative impact on total royalties.

The good news is, because, as Rene mentioned, the Cloud AI or infrastructure business has been continuing to grow ahead of our expectations, it's actually growing at a level that's more than compensating for those kinds of risks on the memory and mobile side. So I think we have a very good setup for next year, and not too concerned about at least the royalty revenue impacts that we might see from these unit volume and supply chain constraints.

Your next question today comes from the line of Simon Leopold from Raymond James.

First one is, I'm hoping you're able to shed some light on this, but wondering what your thoughts on are whether or not SoftBank will potentially need to sell some of the Arm stock that it holds to finance some of the investments you talked about making and how we should think about the implications for your shares? Then I've got a quick follow-up.

Sure. Thank you for the question. That's one that we read a lot about, and there's a lot of speculation on chat boards and whatnot about that. I can tell you from talking to Masa about this, and I would quote him directly, he is not interested in selling 1 share of Arm stock. And that doesn't mean 2 shares or 3 shares. That means any shares. He's very long on the company. He's very, very bullish, as am I, about our long-term prospects. And he has no interest in selling. There's been a lot of writing about it, but I can tell you from a direct conversation and direct conversations, plural, that I've had with him, that's just not the case.

And then just as a follow-up, you've provided a forecast for some deceleration in the royalty revenue growth. I'm just wondering if you could elaborate on the trend? Is it more difficult comps? Or is there something else shifting that we should be considering?

Yes, this is Jason. I'll take that. I would say the royalty trends for next year are pretty consistent in absolute dollars, maybe a little bit lighter just because of what you're now seeing on the memory shortage side, like I said, maybe 1% or 2% impact largely due to that. The growth percentage is down a bit because of the overperformance that we saw last quarter and expected to see again this quarter. So we are coming off of a stronger comp.

Now the obvious question then is because you've had stronger growth both in Q3, we thought we'd grow about 20%. We grew 27%, so a $30 million beat or more. And now seeing some of that flow through into Q4.

Will that flow into next year as well? Right now, too -- I'd say, too hard to say. There's a lot of talk about memory and even wafer shortages. And so that stuff doesn't affect us as much as many of the full fabless semiconductor companies.

But -- so I'd say right now, we'll give you updates as we learn more. But overall, the absolute magnitude of royalties for next year are expected to be pretty close to what we were thinking what we said earlier this year. But we'll see if this recent strength continues and allows us to take things up as we proceed in next year.

[Operator Instructions] And your next question today comes from the line of Vivek Arya from Bank of America.

I actually just had 2 clarifications. One is, I was hoping you could quantify the exact amount of data center revenue? I know you said that it doubled, but how much is it? So we can get a sense for, right, what the magnitude is versus the overall company sales.

And then the clarification -- the other clarification I had was, I think you mentioned SoftBank contributed $200 million. I somehow recall the original expectation was about $178 million, $180 million. And if you could clarify that? And what are you embedding for March and onwards from that contribution?

Yes. The -- well, the $178 million last quarter, it was -- so no new deals were signed. It's just the deals from last quarter, it was $178 million for the quarter. The full quarter has the impact now is about $200 million. So nothing new, it's just a full quarter impact. I would expect that $200 million going forward -- is the right run rate going forward.

And the data center revenue?

Yes, data center revenue, we provide the details on that once a year. I think at the beginning of this year, we said it had just hit double digit. And because it's growing so much faster than the rest, I assume it's going to be somewhere in kind of the teens to probably getting closer to 20%.

As Rene said, over the next 2 to 3 years, you should expect to see it get similar or maybe even larger than smartphone business, which is in the kind of 40% to 45% of total business.

Your next question comes from the line of Mehdi Hosseini from Susquehanna Financial.

Yes. Just as a follow-up to the smartphone topic. Two, how should I think about the migration to the v9 higher royalty is going to help offset the lower smartphone units?

Yes. So I'll let Jason provide the detail. But again, as a reminder, with the way that we handle v9 for smartphones, particularly v9 CSS is, every smartphone cycle, we deliver a brand-new CSS. Each time we deliver the brand-new CSS, the royalty rates are generally increased year-on-year.

So when we think about v9 in smartphones, the appropriate way to think about it is it's all CSS. It's all moving to CSS now. And as a result of that, we get price every year with the royalty increase year-on-year.

Yes. And in terms of the guidance kind of that I just gave in terms of if there's a minus 20-degree unit impact, there's at most a kind of 4% to 6% revenue impact, just specifically within smartphones. That would be incorporating the higher royalty rate per unit that's already been contractually agreed to and that we assume we'll be shipping later in the year.

Your next question comes from the line of Vijay Rakesh from Mizuho.

Rene and Jason, just a quick question on the -- on your partnerships. As your partner, SoftBank executes on its AI road map, should we be expecting like an Arm custom ASIC down the road given the substantial partnership that you have with them with the $200 million a quarter NRV that you're getting? How should we look at that, the timing and how that will impact the fiscal '27, let's say?

Yes, nothing. Hi, Vijay. Nothing we can say specific about any products that you're asking about. So unfortunately, not much more we can say there.

Your next question today comes from the line of Krish Sankar from TD Cowen.

Rene, I just wanted to find a little bit about how to think about Arm's IP penetration rates or percentage rate in AI data center semis today? And where do you think that evolves over the next 3 to 5 years?

It's a wonderful question. I think what we're going to do in the next 3 years is a evolving of how these data center chips are built out. What do I mean by that?

Today, you've got a classic architecture, where you've got a CPU, which connects into an accelerator. The CPU does some work, the GPU does some work. I think what we're going to start to see over time is a morphing of the workloads that the CPU takes that the GPU used to do. And as I mentioned you go to agentic inference, that's going to mean more CPUs, which could be more different custom chips that are CPU-based.

In addition, the inference workloads which are dominated by 2 pieces of area of work, specifically prefill and decode, you could see some specific solutions around that, that continue to extend things like what a Grok has done, for example, you could still see more kind of innovation across that area.

I also think -- you asked about the data center, but I think we're going to start to see a lot of that migrate to the smaller form factors, where different combinations of IP and solutions are going to be needed to address areas where power is much more constrained, particularly around physical AI, and then the lower edge devices.

So I think there's a lot of innovation to come in solving the AI problems because one thing that's clear is that these AI workloads are going to be running on every single piece of hardware that has compute. And because the vast majority of the compute platforms out there today are already Arm-based gives us a gigantic opportunity to move where that goes.

Your next question comes from the line of Harlan Sur from JPMorgan.

On Compute Subsystems, obviously, you continue to drive solid momentum with 2 more licenses added in the quarter. The value-add of CSS that we hear from your customers is resonating extremely well, right? It improves their productivity, it improves their overall system performance. They're willing to pay a higher licensing fee and higher royalty fee for that value add, as you mentioned.

I'm curious to know what percentage of the royalty mix is CSS today? And what proportion of the royalty revenue could it become over the next 2 to 3 years?

Yes. Thank you, Harlan. I'll let Jason take that.

Yes. So Harlan, yes, a lot of progress on CSS with the 5 CSSs that have actually already been turned into silicon and actually something we're receiving royalties on. It's had a material impact.

I think of CSS last year, I think it was just kind of approaching double digit. And this year, it's well into double digit. Think of it as being into the teens. And then I would say over the next couple of years, I expect it to probably -- it could be upwards of 50%.

But we'll have to see. I think the primary drivers for acceleration of CSS has really been mostly around our customers needing to shorten the cycle time, and CSS that cuts cycle time about half.

And so stay tuned, but I would expect to continue to see that acceleration occur and to continue to see. I think right now, every CSS customer that's had a chance to sign up for the next version or kind of renew for the next generation has all done that. So that's certainly a really key indicator of the value that, as you said, customers are seeing from it.

Your next question comes from the line of Charles Shi from Needham & Company.

I think going back, maybe it was 1 year, you guys kind of soft guided FY '26 and FY '27 growth should be around 20%. You're definitely delivering that FY '26. We definitely will see how we think about FY '27, about the quarter. But any early view you guys can provide FY '28? I know I'm asking -- and plus 2 years here, but you guys did do that going back about a year, and I was hoping if you can provide any early view into the outer year?

Yes. I would say for '26, as you said, we had said at least 20%, and I think now we're guiding to 22% at the midpoint. So as you said, exceeding that target.

For '27, not guiding on full year, but in terms of kind of at a high level, the 20% growth rate, I think, certainly is very reasonable and not anything that we back away from.

In terms of '28, we haven't drawn anything out there yet. I'd say maybe stay tuned. There are opportunities as we contemplate other possible offerings and what that could do to our numbers is still something we're working through. So we'll give you an update on '28 sometime down the road.

Your next question today comes from the line of Srini Pajjuri from RBC.

A couple of clarifications, guys. On the memory impact, I guess, you talked about -- you quantified that impact. But Jason, the outlook for the next quarter on the royalties being up low teens, do you think memory is already having an impact on the smartphone volumes? Is that why it's only up low teens?

And then to add to that, you talked about CSS accelerating. I'm just curious, given the pressure on the bill of materials, do you anticipate or are you seeing any impact in terms of the adoption of CSS and v9, I guess, as you look into the next few quarters given the bill of materials challenges?

Yes. Thanks for the question. I'll take the second part first, and then Jason will take the first part on memory. Question was regarding CSS pricing impacting bill of materials. No, we're not seeing any of that at all. What we are seeing is that the value gain by accelerating time to market outweighs anything that customers are considering. Given the complexity of building these chips, the increased cycle times through the fabs going from 5-nanometer to 3-nanometer to 2-nanometer means that the design windows are really short, and missing the first few months of shipment or having any kind of delay is critical to profits.

So based on that, we've really not had many discussions with anyone regarding the BOM impact. The value that we create relative to profits gained by the customer is what really drives the decision point.

And then regarding the memory impact on the next quarter, I'll let Jason address that.

Yes. The memory impact, very minimal, I would say. And that's not really the driver of the guidance on the growth. The absolute growth in royalties has much more to do with typically, seasonality. Our Q4 or calendar Q1 is always one of the slower quarters. And the one thing that happened a year ago is we did have a MediaTek chip come out in Q4 of a year ago -- our Q4, calendar Q1 of a year ago, which was unusual timing. So we are lapping that. So it's much more about kind of what we're comping, and to some extent, seasonality.

But overall, full year royalties, I would expect to be in that north of 20% range, which is kind of what we were expecting early in the year and still expect Q4 or calendar Q1 to be stronger than what we previously expected. So it's really -- the year-on-year growth piece is really more of a seasonality/comping kind of an unusual onetime release from a year ago.

And your next question comes from the line of Andrew Gardiner from Citi.

Jason, perhaps one for you. On the OpEx side, we've clearly seen significant investment in the business, particularly in R&D, given everything that you guys are doing.

You've given us a bit of a steer on fiscal '27 revenue growth. Clearly, R&D has been growing at a faster rate than revenue in the current period. Is that something we can expect to continue into fiscal '27 given everything that you guys have got in front of you? Or will we actually start to see R&D growth slow relative to the revenue?

Sure. So a little early to talk full year. I can tell you right now, our expectation is that the Q4 to Q1 step-up will be similar to last year. I think last year, it was low double-digit sequential growth, and you should see the same kind of sequential growth as a year ago.

Right now, I would say the growth after Q1 is probably going to moderate more so than it did this year. We did see pretty significant step-ups throughout the year. I don't expect there to be quite a significant step-ups for next year. But as we progress more into next year, we'll give you a little more color, but that's the high level, I'd say, modeling approach that we take right now.

And the question comes from the line of John DiFucci from Guggenheim Securities.

Rene, you've seen a lot in technology over the years. So I'm going to ask a question that's kind of a little bit self-serving here. I'm curious how you'd characterize what's happening in the stock market recently as it pertains to the software sector? And if you might, since you're at least partially a software company, how does AI affect your business other than driving demand? In other words, how should we think of how you'll leverage AI in the design of chips and systems?

Yes. Well, regarding the stock market's reaction to the software company, I had agreements with that. I probably would be in a different position than the one that I have. And I'm not sure I can -- I'm in a great position to discuss what the near-term impacts are to the stock market. But what I can say after watching and being in technology my entire career, we do see these kind of things time to time where investors or the market gets jittery around what the broad impacts are, when we're in the midst of fairly significant technology disruptions.

I can say, for our business, given the fact that we are an intellectual property provider that goes into physical things, chips, AI is not going to replace a physical chip anytime soon. They're kind of linked at the hip, if you will, relative to -- you need the hardware to run the software.

I think there's just enormous opportunity, however, still for growth in the overall sector because when I think about where AI actually is operating truly inside the enterprise, it's very lean. When I think about our own company and things like our payroll systems or purchase order systems or our SAP systems, there's some AI going on there, but not nearly enough to be massively transformative yet. And I think part of that is just the complexity of integrating these large systems and changing software workloads.

So I think we're in super early days, to be quite frank, and having been in technology, again, my entire career and have seen lots of technology disruptions, this one feels a little bit like the final frontier in terms of the amount of productivity and change that AI can benefit, and we're still all trying to get our arms around it.

If you just even look at the numbers of spend, I heard earlier today, Google or Alphabet announcing $180 billion CapEx spend. That used to be what semiconductor companies used to spend a year on fabs times a few. So we're in uncharted waters, and maybe that's why you're seeing some jittery numbers relative to how the market reacts. But from where we sit, there's just huge demand for compute, and that's what Arm does. And so I think in the long game, I'm super excited about the opportunity for us.

We'll now take the final question for today. And the final question comes from the line of Timm Schulze-Melander from Rothschild & Co.

It's a 2-parter for Rene, please. You've talked a lot about inference in the AI future. You just referenced the Grok architecture. And I really wanted to ask you what are your thoughts or how should we think about SRAM, SRAM at the edge, some of these different memory structures and what they could mean for your business?

And then the second part is just the cadence of power efficiency for Arm. Is there something that we should think about in terms of the average annual or per v8 to v9 energy for compute efficiency that you see going forward?

Yes. So I'll take the latter part first because it kind of bridges into the first. We look at how to address power efficiency 24/7. And the reason for that is increasingly, as you get into these smaller form factors, the one thing that you don't get much liberty on is battery life and space. So as a result, we have to always think about operating in a constrained environment where you're adding more and more demand of compute. When you add AI onto something that already has to drive a display or open an app or recognize a voice, it's a constant thing that we think about and worry about.

I think we're very well positioned to address it because we are the incumbent in many of these platforms. So it is something we spend a lot of time and energy on.

To your first part of the question on SRAM and different memory technologies, absolutely, that's something we're highly involved in. To oversimplify a computer, a CPU needs memory, and memory needs a CPU. So when you're designing a piece of hardware, the two go very much hand in hand. And there is a lot of work and research being done about not just SRAM, but alternative memory technologies and solutions that can address these increasing demands on AI.

So again, this question prior to yours in terms of the overall broad opportunity, what people in our space tend to worry about is that there isn't hard problems to go think and work on and develop new technologies for. We don't have that problem. Every single end application is going to be impacted by AI. We believe every end application will run AI through Arm. So we're spending a lot of time and energy, and you can see by our investments to come up with innovative ways to address that.

Thank you. I will now hand the call back to Rene for closing remarks.

Yes. Thank you, and thanks for all the thoughtful questions. And we could tell by the range of the questions, we were talking about memory prices inside the quarter and then what alternative memory technologies could look like years from now. I think that's a very good way to sort of describe the current quarter, but how we're very, very bullish about Arm long term.

We delivered the best quarter in our history. We delivered the best quarter in our history on royalties, which is really an indicator for the strategies we have going forward. And we have a huge amount of customers shifting to Arm in a big way with more CPU accounts. That being said, the quarters that we're most excited about are the ones ahead of us. We think we have huge opportunity, as I mentioned, in the new areas of Physical AI, Cloud AI and Edge AI. And we intend to do everything we can to make Arm the compute platform of choice for all AI workloads. Thank you.

Thank you. This concludes today's conference call. Thank you for participating. You may now disconnect.

Good day, and thank you for standing by. Welcome to the Arm Second Quarter Fiscal Year 2026 Webcast and Conference Call. [Operator Instructions] Please be advised that today's conference is being recorded.

I would now like to turn the conference over to your first speaker today, Jeff Kvaal, VP, Investor Relations. Please go ahead.

Thank you, Sharon, and welcome, everyone, to our earnings conference call for the second quarter of fiscal '26. On the call are Rene Haas, Arm's Chief Executive Officer; and Jason Child, Arm's Chief Financial Officer.

During the call, Arm will discuss forecasts, targets and other forward-looking information regarding the company and its financial results. While these statements represent our best current judgment about future results and performance, our actual results are subject to many risks and uncertainties that could cause results to differ materially. In addition to any risks that we highlight during this call, important risk factors that may affect our future results and performance are described in our registration statement on Form 20-F filed with the SEC. Arm assumes no obligation to update any forward-looking statements.

We will refer to non-GAAP financial measures during the discussion. Reconciliations of certain of these non-GAAP financial measures to their most directly comparable GAAP financial measures can be found in our shareholder letter as can a discussion of projected non-GAAP financial measures that we are not able to reconcile without unreasonable efforts and supplemental financial information. Our earnings-related materials are on our website at investors.arm.com.

And with that, I'll turn the call to Rene. Rene?

Thank you, Jeff, and welcome, everyone. We continued fiscal year 2026 with strong momentum, fueled by accelerating demand for AI compute from milliwatts in the smallest of edge devices to megawatts in the world's largest hyperscale data centers. Artificial intelligence is reshaping every layer of technology and Arm is the only compute platform delivering AI everywhere.

Q2 is our best second quarter ever, with revenue of $1.14 billion, up 34% year-on-year, marking our third consecutive billion-dollar quarter.

Royalty revenue reached a record $620 million, up 21% year-on-year, driven by growth in all major markets, including data center, smartphones, automotive and IoT. Unprecedented compute demand has led to our data center Neoverse royalties to more than double year-on-year.

Licensing revenue rose 56% to $515 million as companies continue choosing Arm to build their next-generation AI products.

Our strong results lifted non-GAAP EPS above the high end of guidance.

During the quarter, we announced a strategic partnership with Meta to scale AI efficiency across every layer of compute from AI-enabled wearables to AI data centers on a consistent compute platform. This partnership combines Arm's leadership in energy-efficient compute with Meta's innovation in AI infrastructure and open technologies to deliver richer, more efficient AI experiences to billions of people worldwide.

In the data center, access to power has now become the bottleneck, and this is accelerated adoption of Arm's Neoverse compute platform, which has now surpassed 1 billion CPUs deployed. Our compute forms the foundation of custom silicon from leading partners, including NVIDIA Grace, AWS Graviton, Google Axion and Microsoft Cobalt.

For example, Google's Arm-based Axion chip delivers up to 65% better price performance while using 60% less energy. And as a result, Google is migrating the majority of their internal workloads to run on Arm.

Customers are increasingly deploying Arm Neoverse CPUs alongside their AI accelerators to orchestrate massive clusters, highlighting the versatility and scalability of our platform.

The addition of 5 new Stargate sites this quarter further expands visibility into future AI capacity and reinforces is Arm's central role in the hyperscale build-out.

As AI chip design becomes more complex, our compute subsystems, or CSS, are helping customers accelerate their development cycles and reduce execution risk.

Demand for CSS continues to exceed expectations. During the quarter, we signed 3 new CSS licenses, 1 each in smartphone, tablets and data centers, bringing our total to 19 CSS licenses across 11 companies. We also expanded our collaboration with Samsung, which is leveraging CSS for its Exynos family of chipsets, driving up to 40% AI performance over previous non-CSS generation.

As a result, the top 4 Android phone vendors are now shipping CSS-powered devices. CSS has quickly become the starting point for customers, building next-generation silicon, offering faster time to market and delivering higher royalty rates for Arm.

In the quarter, we also launched Lumex CSS, our most advanced mobile compute platform to date. Lumex enables rich on-device AI experiences such as real-time translation, image enhancement and personal assistance. Flagship devices from partners like OPPO and vivo are expected to ramp later this year, bringing console-quality performance and new AI capabilities directly to mobile devices.

At the edge, AI is transforming how people interact with their devices in their hands, homes and vehicles. Google launched the Pixel 10 smartphone featuring the new Arm-based Tensor G5 chip, which runs Gemini models up to 2.6x faster and twice as efficiently as prior generations.

NVIDIA began shipping its Arm-based DGX Spark system for AI developers, a compact desktop supercomputer for local model training, fine tuning and inference.

In automotive, a flagship electric vehicle built on Arm's platform introduced advanced park assist, voice control and safety features featuring Arm's Automotive Enhanced technology.

Tesla's next-generation Arm-based AI5 chip delivers up to 40x faster AI performance, enabling the next wave of intelligent vehicles and autonomous machines.

Our leadership in AI is amplified by our unmatched software developer ecosystem, now more than 22 million strong, representing over 80% of the world's developer base. This ecosystem is a powerful growth engine for Arm. Every new ARM-based device brings more developers, which drives more software innovation, which in turn fuels greater demand for our compute platform across every market we serve.

As mentioned in our last call, we are continuing to explore the possibility of moving beyond our current platform into additional compute to subsystems, chiplets or complex SoCs. As a result, we continue to accelerate the investment in our R&D as we are seeing increased demand from our customers for our work from Arm.

AI is shaping how the world computes and Arm is a foundation making it possible. From milliwatts to megawatts, we deliver the performance, efficiency and scalability to meet this moment and the years ahead.

And with that, I'll hand it over to Jason.

Thank you, Rene. We have delivered another strong quarter. Total revenue grew 34% year-on-year to $1.14 billion, a record for Q2. It exceeded the midpoint of our guidance range by $75 million and marked our third consecutive quarter above $1 billion.

Royalty revenue exceeded our expectations, growing 21% year-on-year to a record $620 million versus our guidance of mid-teens. The biggest growth contributors were smartphones with higher royalty rates per chip and in data center where we continue to see share gains from custom hyperscaler chips.

Royalty revenue from smartphones grew in order of magnitude faster than the market as multiple OEMs ramped smartphones based on Armv9 and CSS chips.

Data center royalties doubled year-on-year given the continued deployment of Arm-based chips by hyperscaler companies.

Automotive and IoT both continued to grow year-on-year and contributed to our strong royalty performance. Overall, royalty growth rates continue to reflect Arm's increasing royalty rates and rising market share.

Turning now to license. License and other revenue was $515 million, up 56% year-on-year. Growth was driven by strong demand for next-generation architectures and deeper strategic engagements with key customers. We further expanded our license and services agreement with SoftBank. We also signed 4 ATA and 3 CSS deals. These agreements reflect the continued investment by our customers in next-generation Arm technology.

As always, licensing revenue varies quarter-to-quarter due to the timing and size of high-value deals. So we continue to focus on annualized contract value, or ACV, as a key indicator of the underlying licensing trend.

ACV grew 28% year-on-year, maintaining strong momentum following the 28% year-on-year growth we reported in Q1. This is well above our usual run rate of low teens growth rate -- low teens growth and is also above our long-term expectations of mid- to high single-digit growth for license revenue.

Turning to operating expenses and profits. Non-GAAP operating expenses were $648 million, up 31% year-on-year on strong R&D investment and slightly below guidance. These investments in R&D reflect ongoing engineering head count expansion to support customer demand for more Arm technology, including continued innovation in next-generation architectures, compute subsystems, and possibly chiplets or complete SoCs.

For example, over the past 4 years, we've invested heavily in developing the technology that makes up the Lumex Compute Subsystems for smartphones, which we announced in September. This project took around 1,000 man-years with a team size peaking over 450 engineers and required around hundreds of billions of dollars in investment -- hundreds of millions of dollars in investment. Lumex CSS has attracted strong market interest and we're already seeing royalty revenue from an early licensee.

Non-GAAP operating income was $467 million, up 43% year-on-year. This resulted in a non-GAAP operating margin of 41.1% and an improvement from 38.6% a year ago.

Non-GAAP EPS was $0.39, $0.06 above the midpoint of our guidance range, driven by both higher revenue and slightly lower OpEx.

Turning now to guidance. Our guidance reflects our current view of our end markets and our licensing pipeline. For Q3, we expect revenue of $1.225 billion, plus or minus $50 million. At the midpoint, this represents revenue growth of about 25% year-on-year.

We expect royalties to be up just over 20% year-on-year and licensing to be up 25% to 30% year-on-year.

We expect our non-GAAP operating expense to be approximately $720 million and our non-GAAP EPS to be $0.41, plus or minus $0.04.

Our higher revenue allows us to both accelerate R&D investment and pass-through upside to EPS. We are seeing strong demand from our customers for Arm technology, which gives us confidence in our long-term growth trajectory, and our strategy to enable AI everywhere, in the cloud, at the edge and in physical devices. And we will continue investing aggressively in R&D to capture these opportunities and ensure that AI runs on Arm.

With that, I'll turn the call back to the operator for the Q&A portion of the call.

[Operator Instructions] And your first question today comes from the line of Sebastien Naji from William Blair.

Congrats on the nice results. Rene, I wanted to ask about the AI opportunity. There's been a seemingly nonstop stream of new data center deals announced over the last quarter, calling for tens of gigawatts of additional computing capacity to be stood up.

How do you feel about Arm's strategic positioning with respect to these AI deals? And what do you view as the opportunity across the build-out?

Thank you for the question, Sebastien. As a Board member of SoftBank and also given our heavy involvement there with Stargate and regular dialog with OpenAI, I believe I have a unique perspective in terms of visibility in terms of this market.

One thing that's become quite evident is that power has become the bottleneck for everyone and power not only means access to energy, but everything underneath it in terms of infrastructure build-out, turbines, transformers, everything associated with generating power.

So in that environment, everyone wants to move to the most efficient compute platform as possible. Arm is about 50% more efficient than competitive solutions. We've seen that across the board in benchmarks, but also more importantly, in real-life performance. And that's why we see NVIDIA, Amazon, Google, Microsoft, Tesla, all using Arm-based technology.

We see an unprecedented demand for compute and all the incremental compute that we've seen announced literally has all been based on Arm. So that's driving huge growth opportunity for us, and it's one of the indicators as to why we've seen such growth in our Neoverse business more than doubling year-over-year.

Your next question comes from the line of Joe Quatrochi from Wells Fargo.

I noticed in the filing you announced your intention to acquire DreamBig Semiconductor. Curious just kind of what's behind that? And how does that kind of fold into your plans to potentially expand beyond your current kind of offering platform?

Yes. Thank you for the question. So DreamBig is a great company. They've got a lot of interesting intellectual property particularly around the Ethernet area and already make controllers, which are very, very key for scale-up and scale-out networking.

So when we look at the demand for what's going on inside the data center and particularly in the area of high-speed communications, that type of technology will be very helpful for us to broaden our offering to end customers. So we're very excited about the company and DreamBig has got some fantastic engineers.

Your next question comes from the line of Jim Schneider from Goldman Sachs.

I noticed in your disclosures that you saw a material step-up in related party revenue. So I was wondering if you could maybe talk a little bit about -- there's also been many announcements related to Stargate and SoftBank since the last earnings call. Can you maybe give us any kind of color you can on the nature of that relationship and how things are changing in terms of design activities?

So one of the ways to think about Stargate and particularly given the relationship between Arm and SoftBank is a huge opportunity for Arm to partner with SoftBank and SoftBank's partners to provide technology into all those solutions.

So without getting into too many of the specifics, but at a high level, if you think about what's associated with building out these data centers, you have the compute, obviously, you have the networking, you have everything associated with power distribution, you have a potential technology that gets into the power mechanism of the data center and then everything associated with even potential assembly of the data center.

So as a result of all the work that SoftBank and the SoftBank family of companies are doing, it provides huge opportunity for Arm to provide solutions into that space. So that, at a high level, is the way to think about how the SoftBank family works together on these designs.

Your next question comes from the line of Ross Seymore from Deutsche Bank.

I wanted to go back to the OpEx side of things. I know it was a little bit below your guide in the second quarter, but the fourth -- third quarter looks like it's going to step up again.

Kind of a bigger picture one. You mentioned about exploring different sorts of go-to-market methodologies, chiplets, et cetera. When do you expect to give us more color on when that's going to go from exploration to return on investment or the actual strategy, how should we monitor that and expect to get more information from you?

Yes. Thank you for asking. The best detail I can give you is there's nothing I can talk to you about today in terms of time line, about products or technologies. When the time comes for us to announce it, you'll be the first to know in terms of what we're doing.

Right now, the best commentary I can give is that everything associated with those solutions does require a significant level of R&D. Now as you've seen on the guidance going forward, our revenue go forward is higher than our OpEx increase, which is something we've been very careful to manage. So we feel comfortable about that.

But at the same time, what we're looking at in terms of the opportunity for compute and more importantly, compute using Arm has never been greater. So as a result, we want to make sure we're in the best position possible to capture it. We're looking at all possibilities in terms of how to do that. And when we're ready to talk about what that is, we will certainly advise.

The only thing I would add is, I think last quarter, we said, as soon as the way we think about when we announce something, if it were to be something related to full SoCs, it would be once there's tape-out, once there's samples back and once there's actually noncancelable customer orders, when we achieve all 3 of those milestones, that's when we would probably talk about something because this would be a new business and something we haven't done before. So whenever those milestones are achieved, that's when you should expect to hear from us.

Your next question comes from the line of Vivek Arya from Bank of America.

I just wanted to clarify how much was the SoftBank contribution in Q2 versus what you thought? And then what is baked in for Q3 and hopefully, if you have the number for Q4?

And the real question is how long can this quarterly rate persist? And if you do move into physical chips or chiplets or any other products as part of target, does it start to cannibalize this licensing stream?

Yes. So thanks for the question. In terms of the impact, it was about a $50 million increase from last quarter. So last quarter, we think we were about $126 million. It actually went up $52 million, so now about $178 million. and that's a good run rate to assume going forward.

The only way it would change is if we have any additional deals. And again, these are license plus design services. So think of it as being licenses to our IP to work with SoftBank on exploring solutions. But then think of the design services being effectively a kind of a funded R&D model. And so that's a lower margin revenue, of course.

So these -- in terms of how long these revenue streams will occur, we're not at liberty to say yet, but I would say, as Rene said, at some point, probably in the next year or so, you'll hear us talk about what products those might be. But obviously, that's not just up to us. It's when SoftBank is ready to talk about what these products could look like and what the revenue profile, et cetera, is.

And so when that would occur, it's likely to assume that there would be some different revenue source, whether it's royalties or gross revenue from selling a chip if in fact, it's a full SoC. Those are all things that are still to be worked out. And yes, I would think of that as being, to some extent, cannibalistic of whatever the current license and design services.

But then, of course, if there is a product, you could also assume there could be successive generations of products after that, in which case you could stack royalty between license and design services. But then, of course, there could also be royalties or whatever the revenue relates to whatever the product that ships in market is.

So I would think of it as very much durable revenue, in that I think if SoftBank wasn't a related party, we would just be booking license and design services, and it wouldn't be a related party, but then the numbers would be pretty similar. And so the fact that the related party I think is probably what makes it look somewhat unique. But the reality is we also, as Rene already mentioned, this is not really just between us and SoftBank. They also have contracts with many others, OpenAI, other Stargate partners as well. So I would think of this as all being part of a larger effort.

Your next question comes from the line of Timm Schulze-Melander from Rothschild & Co Redburn.

I had 2, please. Just following on, on the Stargate theme and the sites. Can you maybe just talk about the shape of what that revenue opportunity looks like on a sort of 1-, 3- and 5-year view just kind of when it's going to start having an influence on the revenue -- the annual revenue or quarterly revenue of the business?

And then my second question was, just to make sure, I wasn't sure I caught it right. You talked about the Lumex CSS. I think that's a product that you launched in September, but I think you also said that you already have royalty revenues associated with that. If you could just maybe expand on that a little bit, that would be really helpful.

Sure, sure. I'll take the first part of that question, and I'll let Jason take the second half.

Without giving you kind of a go-forward forecast of 1, 3, 5 years, maybe a way to think about it is, back in January of this year, OpenAI with Oracle and SoftBank announced Stargate, which was a $500 billion project to build out data centers over the next number of years.

When we go back to where we are now 11 months later, I would say the demand picture for compute is greater than it was at that time. So this is a bit of why you're seeing all kinds of different accelerated announcements around spend, et cetera, et cetera.

So if nothing else, I think the opportunity for compute has only grown since we made that Stargate announcement. And to be clear, that announcement is around a joint partnership with OpenAI and SoftBank being equity partners in this investment for compute.

So we are quite bullish in terms of this overall demand for compute. Right now, what is in the way of realizing that potential is all of the infrastructure required around the power. But from everything that we can tell from people we talk to inside the ecosystem, the demand for compute to train these new models, reinforcement learning to make them great and then inference to serve them, the demand opportunity is stronger than what we announced 11 months ago. So this is why we're accelerating all the investments that we talked about to take advantage of that opportunity.

On the Lumex CSS royalty question, I'll let Jason answer that one.

Yes. So I would say the licensee that's already actually -- that we're already receiving royalties from, that is, I'd say, earlier than expected. And the way -- because we just launched this in September, the way it's happened so quickly is this actually -- we're not able to say which partner it is, but it is a partner where this is not their first CSS, this is their second CSS.

So as a result, there was already kind of close partnership on the first generation. And so then when we launched the next generation, because the teams have already been working pretty close to each other, it allowed that second generation to be adopted very quickly and for royalties to come really just within a couple of months after the technology was delivered.

So kind of unusual, a little ahead of what we had expected, but it very much speaks to exactly why CSS has been more successful even than we thought when we launched it 2 years ago. It's really about speeding up time to market, and this is an excellent example of that occurring.

Your next question comes from the line of Harlan Sur from JPMorgan.

Rene, you talked about Neoverse royalties growing 2x year-over-year with all these cloud-based CPUs ramping. And then on top of that, with these high-performance AI clusters, right, they're using more DPUs or SmartNICs that are also using Arm cores.

On the networking side, data center switching and routing chips have multiple Arm cores embedded in them for things like telemetry, load balancing, overall system management.

The bottom line is that there's significant Arm compute going into all aspects of the data center, right? We're also even seeing Arm taking over x86 in the service provider networking markets as well.

So last fiscal year, cloud and networking accounted for about 10% of royalty revenues. We're midway through this fiscal year. Maybe you guys could just true us up, I assume, this mix has increased. Is it approaching 15%, 20% of total royalty revenues for the team? Any color here would be great.

Yes. I'll let Jason address the numbers, but thank you for being a great salesman and describing our penetration across domains. You're 100% right. There's Arm technology in virtually every set of the networking stack. The BlueField technology at Mellanox, DPU-based, that's Arm. Significant technology goes into the switches around Tomahawk and Arista are all using Arm technology.

So we are definitely seeing an acceleration of all that. And at the same time, I think the power efficiency piece is probably the biggest accelerant I think we're going to see just in terms of being able to offload as much as everything you can on to the more power-efficient domain of the compute platform.

So I'll let Jason comment on royalties scheme in terms of where that is going directionally.

Harlan, so on the royalties, yes, I mean, it ended the year at around 10-ish percent. And so we're certainly with the growth rate in infrastructure being double, I'd say, all the other categories in overall average royalty, you should expect it to continue to increase.

We'll provide a full update at the end of the year. But your trajectory of somewhere in the 15% to 20% range is not a bad assumption and probably a reasonable expectation for where we expect to trend throughout the year.

So I would say it's probably going faster than we expected a year ago.

Your next question comes from the line of Krish Sankar from TD Cowen.

I have a question for Rene. Clearly, you kind of highlighted how you have strengthened smartphones and also increasing market share in data centers. I'm kind of curious, when you look over the next few years, how do you see chip demand and token generation playing out and its implication for Arm, especially as you move into more of an inference world where edge devices may play a bigger role?

I think from some accounts of people who I talk to will say that today on some of these data centers, these build-outs of multi-hundred megawatts that still -- and again, depending on how you define training versus inference and reinforcement learning, majority of compute is being used for training still. That clearly will flip. Well, at some point, it has to, we think. And then that demand starts to move to inference.

What we're seeing is all kinds of demand for different architectures and compute type of solutions to run inference not in the cloud. Obviously, you're going to not rely 100% on something on the edge. But today, it's the reverse. It's about 100% on the cloud. And we think that is going to change.

We are seeing already lots of demand for the CPUs and Lumex that have these scalable matrix extensions, and these are the extensions that allow you to run AI workloads at higher performance. That's only going to continue.

And I think for Arm, that is an enormous trend for us on 2 levels. Number one, huge trend for us because the further you move away from the cloud on to battery-level devices, that's a domain that Arm can play in, in the sense of the software workload running exclusively there. But at the same time, customers would love a scalable software solution between the cloud and the edge.

And that's a lot of what's behind the announcement that we made with Meta in October. This is around working in such a way with Meta where whether they're running something in the cloud or running in the edge, for developers, they're able to port models in such a way that it's as efficient as possible no matter where you're running.

So this is all, I think, a good thing for us because more tokens means more compute, more computes means more compute needed at the edge, and more compute at the edge is really good for us because that's a -- I think we're in a very, very unique position to address that.

We will now take our final question for today. And the final question comes from the line of Lee Simpson, Morgan Stanley.

Well done everyone on a great quarter. I see China is maybe 22% of sales this Q. And I was just wondering what is driving that? Is it more licensing or royalties for strength in the quarter?

And maybe just as you look at the licensing pipeline for the rest of the year, have you seen more reason to be confident in the growth this year for licensing, especially as you look to Q4, which, as I believe we said before, there's potential for good renewal deals this year.

Thanks for the question, Lee. In terms of the China performance, yes, it definitely has done well. And I would just overall say the demand in China looks to be as strong as we've ever seen. We did have one of our largest license deals actually come out of China.

And so I would say license was slightly more of a -- I'd say, more of the overperformance came from license. Royalties are also growing strong in China as well, but license was a little bit of a bigger driver this quarter. And our pipeline indicates that we have a pretty strong license pipeline for the remainder of the year.

In terms of overall license revenue, hard to say as we get into Q4. There are some large deals as we always have. In terms of timing. Right now, we're just guiding on Q3. But next quarter, we'll definitely have much more clarity around what deals are going to be able to land in Q4 and whether there's any pull forward, pushouts or whatnot.

But as a reminder, we don't -- the deal cycles on large license deals are usually 6 to 9 months, and we don't really lose deals. It's really just about what exactly are the market needs for customers and when do they need it. And given the certain -- the current CapEx kind of forecast and all the AI cycles that continue to be as strong as they've been for the last couple of years, I have a lot of confidence, but we'll give you a little more detail next quarter on what's going to land in Q4.

Thank you. That was our final question for today. I will now hand the call back to Rene for closing remarks.

Thank you, and thank you, everyone for the questions. As we stated, we could not be more happy with the results last quarter. Royalties at a record, 34% growth year-on-year, just terrific results.

But more importantly, when we think about the opportunity for Arm going forward, the future has never been brighter because if we look at what's going on with artificial intelligence, artificial intelligence is driving unprecedented demand for compute.

And given the unprecedented demand for compute, we are seeing all kinds of constraints on power and infrastructure to deliver that compute, which means that the compute that's being delivered for AI needs to be as efficient as possible. That's also a great place for Arm.

And then as more and more of this AI compute moves from the cloud to edge devices and requires the most efficient compute on the planet, that's a great place for Arm, too.

So we are extremely excited about the future going forward. We continue to invest to ensure that we can take advantage of that opportunity. And on behalf of everyone inside Arm who made this quarter happen and to our partners and customers, thank you so much, and thank you for all the questions.

Thank you. This concludes today's conference call. Thank you for participating. You may now disconnect.