GDS Holdings Ltd. Sponsored ADR Class A Stock

Hello, ladies and gentlemen. Thank you for standing by for GDS Holdings Limited's Second Quarter 2026 Earnings Conference Call. [Operator Instructions] Today's conference call is being recorded.

I'll now turn the call over to your host, Ms. Laura Chen, Head of Investor Relations for the company. Please go ahead, Laura.

Thank you.

Hello, everyone. Welcome to the Second Quarter 2026 Earnings Conference Call of GDS Holdings Limited. The company's results were issued via Newswire Services earlier today and are posted online. A summary presentation, which we will refer to during this conference call, can be viewed and downloaded from our IR website at investors.gds-services.com.

Leading today's call is Mr. William Huang, GDS Founder, Chairman and CEO, who will provide an overview of our business strategy and performance. Mr. Dan Newman, GDS CFO, will then review the financial and operating results.

Before we continue, please note that today's discussion will contain forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Forward looking statements involve inherent risks and uncertainties. As such, the company's results may be materially different from the views expressed today. Further information regarding these and other risks and uncertainties is included in the company's prospectus as filed with the U.S. SEC.

The company does not assume any obligation to update any forward-looking statements, except as required under applicable law. Please also note that GDS earnings press release and this conference call include discussions of unaudited GAAP financial information as well as unaudited non-GAAP financial measures. GDS press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures.

I'll now turn the call over to GDS Founder, Chairman and CEO, Mr. William Huang. Please go ahead, William.

Hello, everyone. This is William. Thank you for joining us on today's call.

AI is transforming our business. Our sales momentum is the strongest we have ever seen. In the second quarter, we achieved 260 megawatts of new bookings, bringing our total for the first half of 2026 to a record 470 megawatts. During the current quarter, we are well on the way to securing further major business wins with leading customers. We are confidently raising our full year sales target to 1 gigawatt. All of our sales agreements, including -- include a binding take-or-pay commitment. This is a metric which we disclose as bookings. The sales agreement specified delivery date, which is up to 4 quarters after bookings.

This allowed us to invest based on secured commitments. Following the delivery date, there is an agreed ramp-up period, usually another 4 quarters, which gives us visibility to the timing of new billings. Alongside the new bookings, our customers also request us to reserve deployable capacity at the same site for their future needs. Reservation has become an integral part of our sales agreements. So far this year, we have secured an additional 600 megawatts of reservations for our -- from our customers.

We expect to end this year with over 1 gigawatt of new reservations. This provides us with high visibility for new orders in the next couple of years as customers convert reservations to binding commitments. China's tech giants and emerging AI leaders are driving the adoption of advanced agentic models. This has led to a structural upgrade in the demand for computing power and AI infrastructure. GDS is uniquely positioned to address this opportunity as a result of our strategic customer relationships presence across all key markets in China, track record of execution and financing capabilities.

The strength of our platform is clearly evidenced in the composition of our first half bookings. We won significant new business from each of the 3 largest hyperscale customers. At the same time, we started to establish relationships with a group of emerging AI leaders, which have the potential to generate incremental demand in the future. Our new business wins are diversified across the markets. For the first half of the year, around half of our bookings came from the established markets and half from new markets, including the Ulanqab and Horinger in inner Mongolia and Shaoguan in Guangdong province.

We are progressing well with customers for our Changshu campus in Jiangsu Province, which is another new market. This new sales success validates our differentiated resource strategy. At the midpoint of this year, we have total binding commitments for over 2 gigawatts plus a further 600 megawatts of reservations. On the capacity side, we have around 3 gigawatts of developable capacity, which is not yet committed or to -- under reservation. It is mostly in new markets. In view of our current sales momentum, we are actively adding to the deployment pipeline in the markets where demand is growing.

While pursuing our ambitious target, we remain selective in terms of customers and the contract terms. We invested against binding long-term commitments for -- from the customers, and we are committed to maintaining financial discipline. I will now pass on to Dan for the financial and operating review.

Thank you, William.

I'll start from the backlog buildup on Slide 10. We started the current year with a backlog of 450 megawatts. By the middle of the year, our backlog had increased substantially to 757 megawatts. Based on the pricing in the contracts and our operating cost benchmarks, we estimate that we can generate RMB 2.2 million of adjusted EBITDA per megawatt on average from this backlog. Our booked but not billed adjusted EBITDA was therefore around RMB 1.6 billion. By year-end, assuming we achieve our sales target, we expect the backlog to increase further to over 1 gigawatt.

Turning to Slide 11. During the first half of 2026, our net move-in was 145 megawatts. During the second half, we forecast move-in of another 90 megawatts, making 235 megawatts for the full year. The move-in pattern over the course of 2026 reflects the timing of bookings last year. For 2027, we forecast move-in will increase substantially to more than double the number for 2026. The move-in will be heavily weighted to the second half of 2027.

Assuming we sustain our sales momentum, 2028 should see another step-up in move-in. Turning to CapEx on Slide 12. Our unit CapEx for the new capacity, which we are constructing averages around RMB 20 million per megawatt. As we just raised our sales target for the current year, we are also raising our guidance for CapEx paid from RMB 9 billion to RMB 10 billion, most of which is in the second half. Our plan is to continue financing new investments with around 60% debt and 40% equity at the project level.

Assuming we can generate a stabilized cash yield on new investments of 10% to 11%, this implies leverage of around 5.5 to 6x at the project level. Our primary source of debt is onshore RMB-denominated long-term bank borrowings. The onshore bank market remains highly supportive. During 2Q '26 alone, we were able to complete RMB 4.9 billion of new debt financing and refinancing. For the project equity, we have various sources. We have cash of nearly RMB 20 billion on our balance sheet, and we have delevered down to 4.7x net debt to last quarter annualized adjusted EBITDA.

We have operating cash flow, which continues to strengthen. And we have our onshore asset monetization program, which we are building up in a very deliberate way. Following our successful C-REIT IPO, the first post-IPO asset injection is currently under regulatory review. Turning to Slide 16. We are revising upwards our full year revenue and adjusted EBITDA guidance to reflect a more accurate financial outlook for this year, which includes the onetime items disclosed in 1Q '26.

Turning to Slide 17. In order to put our first half '26 financial performance and revised full year '26 guidance into context, we have made some pro forma adjustments. Starting from reported revenue and reported adjusted EBITDA, we deduct the onetime items in 1Q '26. For consistency, we also deduct recurring income in prior quarters, which was restructured into the onetime payment. and we deduct the revenue and adjusted EBITDA contributed by the monetized assets prior to their deconsolidation. These adjustments establish a clean basis for comparison. For the first half of 2026, our pro forma adjusted EBITDA increased by 12.7%. Taking the midpoint of our revised guidance for full year '26, the implied growth rate of pro forma adjusted EBITDA is 6.5%.

We'd now like to open the call to questions. Operator?

[Operator Instructions] And our first question comes from the line of Yang Liu from Morgan Stanley.

Congratulations on the upward revision of full year guidance. I would like to ask about the future potential move-in. I think that there's a lot of debate on your customers' CapEx and also the availability of GPU in the market and also the constraint of computing power. We also see that you expect your move-in to improve dramatically next year. What could be the downside risk for that? And if there's any concern or a delay in one customer getting the GPUs, will the take-or-pay contract protect GDS revenue?

Yes. Thank you. I think dynamics of the demand from the different dimension. I think, of course, the key driver is still the GPU. But the GPU, I think in terms of the domestic GPU, the supply is catching up. I think it took a while in the last couple of quarters, right, as we mentioned. But now it looks like on track to catch up. This is number one. But in the meanwhile, I think they also drive a lot of traditional cloud growth.

What we have seen is the new order quite a big number is driven by the CPU. So it will not impact in terms of the supply, it's no issue. So I think this is all positive. So that we take a more positive way to look at the current or future chip supply. So that's our view. If you look at the other -- a lot of the traditional cloud business, they are still raising their target and the growth is very significant as well. So I think let's be clear there.

How about the take-or-pay term protecting the GDS revenue?

Yes.

Yes, 2 comments. The first is that in each contract, there is a specific delivery date when the capacity has to be available to move in by the customer and that is a fixed date in each contract. It's up to 4 quarters from when the booking is disclosed. So that part, I think, is unchangeable. After that, there's a move-in period, and it varies from contract to contract. We've been very focused on trying to select contracts which have a shorter move-in period and a fixed commitment.

For the purposes of forecasting, we assumed that the move-in will be on average over 4 quarters on a straight-line basis. So that is what our forecast reflects. In reality, it could be faster or it could be slower. But I don't think it will materially deviate from that.

And our next question comes from Sara Wang from UBS.

Congrats on the really solid new order signs. As management just mentioned that there is increasing demand from emerging AI leaders. So just wondering, is there any difference in their demand profile or contract terms compared to established cloud or Internet hyperscale customers we already served for quite some time.

I think we are just starting to build up our team. So far, we are very selective business from some new AI leader. I think in terms of their demand profile, it looks like it's getting bigger and bigger, but we are still very selective. Our main customers and the new business mainly driven by the hyperscale, a couple of large hyperscale. But we think there are some new customer in future, it's the right thing to do to diversify our customer base. So we just start to build some relationship with them right now. So of course, the demand is obviously [indiscernible] in which we believe.

We will now take our next question from the line of Frank Louthan from Raymond James & Associates.

I wanted to get an update on what your new guidance is and what does that imply for the impact of potential action with the C-REIT contribution? Does that include any of that? And what would you expect that to be -- how would you expect that to impact revenue and EBITDA? And then secondly, if you could just address the slowdown in MRR, how should we think about that? And what -- and if we're looking forward, are you signing contracts that should be resulting in an improvement in MRR going forward? How should we think about that?

Frank, first of all, on guidance, to make clear that our guidance does not take account of any further asset monetization. There's a transaction in progress under regulatory review. We can't be any more specific about the timing of that. But to be clear, it's not factored in. For the [ MSR, ] we've provided guidance about the yield in terms of EBITDA per megawatt for the backlog and the new business that we're winning. And I think that will help for forecasting. If we go back to [ MSR, ] I always make the comparison on a same quarter basis.

So if we take 4Q '26 compared with 4Q '25, we forecast that it will be down 3% and then maybe by a similar amount next year. Part of that is the change in the location mix because there's a substantial amount of new business in new markets. And part of it is due to the legacy contracts where we have about another 18 months to go before we are through the transition of adjusting all of our contracts to the current market pricing. So our guidance this year and what we indicate in the future will fully reflect that.

I should point out that, I mean, the Tier 1 market, I mean, also the new market, the current price level is stable. It is all about the transition...

And our next question comes from the line of Daley Li from Bank of America Securities.

Congrats on the upward trend for the new orders. I have one question regarding the move-in. I remember in last earnings call, we are seeing a soft move-in rate in Q2, but it seems the number is better than our -- the market expectation. So what will be the -- what has been the key drivers for better move-in in Q2? And secondly, how do we see the demand, supply trend in the data center market in China, considering the power quota approval progress by the government?

I would not read anything into the quarterly fluctuations. Most of the move-in in the current year is the capacity that was booked in 2025 or even before. And if you look at the bookings in 2025, we had a very strong first quarter 2025 and then the second, third, fourth quarter were at a lower -- consistent level. And then from the first quarter of this year, our bookings increased by a very large amount. That's sustained in the second quarter.

We gave an indication for the full year that's sustained. So I think you can derive from that the outlook for move-in over 2020 -- remainder of 2026 and 2027, we see a significant increase in move-in in the second half of 2027, which is going to lead to a significant acceleration of EBITDA growth.

I think the current power, there's a couple of key points. Number one is now it's controlled by the central government and the municipal government as well. So basically, if you apply the polish, first step is to go to the municipal level because the local government commitment and their full support, right?

This is -- now government is quite selective right now. They try to give some [indiscernible] market leader more allocation. That's why we have built up our land bank in the last 18 months so quickly, right, and take some advantage of the GDS brand, right? So second then we go to the provincial level [ NDRC ] approval, then go to the final approval from the central government, the [ NDRC ] central government. That's the key process of how we get [indiscernible] location.

We will now take our next question from the line of Edison Lee from Jefferies.

So congrats on the good results. My question -- sorry, it's really centering around just reconfirming the definition of the bookings and the reservations. So I assume that bookings, contracts have been signed and reservations mean that is being -- is sort of an MOU with indicating interest by the customers, and you look forward to converting that into signed contracts over the next few quarters. Is my understanding correct?

Not exactly. What I'd like to make clear is that there's a sales agreement, which contains a booking, which is a contractual take-or-pay commitment. But within the same document, we undertake to reserve capacity to enable the customer to have certainty of being able to make commitments typically at the same site in future over a period of time. So the bookings and the reservations go together, and that's how the customers look at it from a resource planning perspective.

Yes. In the meanwhile, I think we should say based on our last 12 or 18 months experience, which the reservation -- our customers exercise their reservation in a 100% basis. That's our current experience. But in terms of the case by case, which negotiate, moving in general, reservation is quite certain -- provide a very, very high certainty for our future booking.

Okay. So can I follow up by asking your booking targets this year right now is 1 gigawatt. I think in the last quarter, I think your target was still 500 megawatts. So this doubling of the bookings target, I believe, is driven by your customers or your assessment of the customers' demand. And is it possible for you to split the customers' demand into training versus inference? Or you have no idea how to split that?

I think the campus like in the new markets, I think they will host a different workload. It's a training plus inference, both their workload increased the guidance. I think the increased guidance is number one is that the whole market demand we see is increased. If you look at our hyperscalers, they continue to increase their CapEx, and that's in line with that. That is number one. Number two, I think GDS still maintain a lot of advantage, which is our customers prefer. So everybody knows we step in the new growth and we started our new business plan. So I think in terms of the capital revenues, even better than the other competitors. So I think the customer will more rely on us.

And in terms of your power reserves, can you talk about the locations of your power reserves?

The part that we identify is developable capacity that is almost entirely new markets. We have capacity in established markets that it's under reservation. So there's only a small amount in established markets that is not committed or reserved.

So is it very different from what you disclosed in the last quarter in terms of locations?

[indiscernible].

We will now move to our next question -- and our next question comes from the line of Timothy Zhao from Goldman Sachs.

I think I just want to get more clarity on the move-in and how do you want to look at the revenue and EBITDA, I think beyond this year. Just wondering if you can give us a breakdown, like, for example, for this year, a lot of move-ins, what is the proportion between CPU based and GPU based? And into next year, it seems like you are looking for the move-in to be more than double to close to 700 megawatts next year. And what will be the breakdown between GPU and CPU next year? And with that 700 megawatts move in, of course, I think the majority will be more geared towards the second half of the next year. So if that is the case, then how do you think about the revenue and EBITDA growth, I think, beyond this year into '27 and '28.

It's -- I think it's not -- in general, we don't have the current detailed specific number in terms of the breakdown there. But in general, I think I can give you the general -- I mean, assumption, maybe it's around 50-50.

Yes, about growth in 2027, we provide annual guidance. Obviously, we won't be doing that until we give the full year results in around March next year. But what you can already see is that over the course of next year, there's going to be a very significant acceleration. The growth rate from 1Q, 2Q, 3Q, 4Q is going to be very different. I think what really matters is where we are at the end of the year and where we are in 2028. I believe it's already a strong indication that in 2028, GDS is going to be a pretty high-growth company.

And my follow-up on the breakdown 50-50. Just wondering if that refers to both this year and next year and onwards or how that mix can change into next year?

Yes. Maybe GPU will a little bit higher next year, that's what I guess based on the current domestic supply is catching up. I think -- yes.

Thank you. Due to the time limit of today's call, I would now like to turn the call back to the company for any closing remarks.

Thank you all once again for joining us today, and see you next time.

This concludes this conference call. You may now disconnect your lines. Thank you.

Hello, ladies and gentlemen, thank you for standing by for GDS Holdings Limited's First Quarter 2026 Earnings Conference Call. [Operator Instructions]. Today's conference call is being recorded.

I will now turn the call over to your host, Ms. Laura Chen, Head of Investor Relations for the company. Please go ahead, Laura.

Hello, everyone. Welcome to the First Quarter 2026 Earnings Conference Call of GDS Holdings Limited. The company's results were issued via Newswire Services earlier today and are posted online. A summary presentation, which we will refer to during this conference call, can be viewed and downloaded from our IR website at investors.gdservices.com.

Leading today's call is Mr. William Huang, GDS Founder, Chairman and CEO, who will provide an overview of our business strategy and performance. Mr. Dan Newman, GDS CFO, will then review the financial and operating results.

Before we continue, please note that today's discussion will contain forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Forward-looking statements involve inherent risks and uncertainties. As such, the company's results may be materially different from the views expressed today.

Further information regarding these and other risks and uncertainties is included in the company's prospectus as filed with the U.S. SEC. The company does not assume any obligation to update any forward-looking statements, except as required under applicable law.

Please also note that GDS earnings press release and this conference call include discussions of unaudited GAAP financial information as well as unaudited non-GAAP financial measures. GDS press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures.

I'll now turn the call over to GDS Founder, Chairman and CEO, Mr. William Huang. Please go ahead, William.

Hello, everyone. This is William. Thank you for joining us on today's call. Over the past few quarters, we have seen a resurgence in data center demand driven by AI. We believe this is the beginning of a multiyear growth story, supported by increasing availability of domestic chips.

Customers are planning their future deployments at unprecedented scale with a high degree of conviction. As market leader, GDS is well prepared to address these opportunities to the fullest extent. We have the trust of all the key customers, a multi-gigawatt development pipeline in strategic locations and a very strong balance sheet.

Up to the end of 1Q '26, our total bookings stood at 1.8 gigawatts. In our 3-year business plan, we target adding 500 megawatts to 800 megawatts of new bookings every year, with the potential to do more. To deliver this capacity, we are prepared to commit RMB 30 billion to RMB 50 billion of new investment over the next 3 years. The economics of the data center business in China is solid, and this new investment will create significant value for our shareholders.

On the last earnings call, we announced a sales target for 2026 of at least 500 megawatts. In the year-to-date, we have already done over 340 megawatts of new bookings, and we are still being selective. We are well on track to reach or exceed our full year target. We have won significant new orders from all of our largest customers for deployments across the whole of our platform, including the new markets.

For the hyperscale business, customers are planning gigawatt scale deployments in single cluster. When they sign new sales agreements with us, they commit to a certain amount of capacity, which we disclose as bookings and ask us to reserve the rest of the sites for their subsequent phases.

In the year-to-date, total new bookings plus reservations comes to over 1 gigawatt. The reservation give us near certainty of winning follow-on orders within the next 1 or 2 years. In order to fulfill our customer requirements, we expanded our platform to new locations, which can accommodate the largest AI deployments. These new locations integrate well with our platform in established market, enabling us to serve diversified customer requirements.

Anticipating the demand trend, we increased our secured landbank to nearly 4 gigawatts. Typically, we are purchasing land from the government exclusively for our data center development. As we obtain customer commitment, we will be granted a power quota for this site. We synchronized the timing of construction with new bookings and fixed move-in schedules.

Over the past 15 months, we initiated over 100,000 square meters or 400 megawatts of new construction, which is almost entirely pre-committed. Our backlog has increased to over 200,000 square meters or almost 600 megawatts, most of which we will become biddable within the next 6 to 8 quarters.

As this appears our growth will start to accelerate. AI in China is a transformational opportunity. We are super motivated to support this development and we will commit all the resource requirement to the expansion of our AI infrastructure platform.

I will now pass on to Dan for the financial and operating review.

Thank you, William. For our new business, the unit development cost averages around RMB 20,000 per kilowatt or USD 3 million per megawatt, depending on specification, cooling technology and location. Pricing for new business is stable. And at current levels, we're able to generate an adjusted gross profit yield of 10% to 11% for stabilized assets.

As shown on Slide 13, across the whole of our in-service portfolio, the adjusted gross profit yield is currently around 11%. We calculate this ratio based on adjusted gross profit, which includes the cash cost of operating assets, divided by gross PP&E, which includes replacement CapEx already incurred and for conservatism, we added back historic impairment charges.

The portfolio yield has been stable at around 11% for the past few years, based on a portfolio with utilization rate of around 75%. As our new bookings are delivered, we expect the portfolio yield to remain in the 10% to 11% range, which, in our view, is a reasonable return.

Assuming a 6-year investment cycle of development, ramp-up, stabilized operations and then asset monetization, we expect to generate a return on equity of around 20% from the incremental investment. This underpins our confidence in growing the business.

As shown on Slide 13, during the first quarter, net additional area utilized was around 16,000 square meters. During the current quarter, this metric will be slightly lower. And then in the second half of the year, it will rebound to around 20,000 square meters per quarter.

During the second half of next year, as we start to see the flow-through from this year's higher level of new bookings, the move-in rate will step up noticeably.

MSR on Slide 16 is a useful metric for financial forecasting purposes, but must be seen together with unit development cost. This is why we think it's more relevant to look at the gross profit yield or cash-on-cash yield as a measure of the economics of our business.

Turning to Slide 18. During the first quarter, we recorded 7.9% growth in revenue and 8% growth in adjusted EBITDA after excluding onetime items, which arose in the normal course of business. We find it useful to look at our growth rates on a pro forma basis, adding back the deconsolidated revenue and adjusted EBITDA of the assets, which we monetized in March and July of 2025. This shows pro forma revenue and adjusted EBITDA growing at 12% to 13% after excluding onetime items.

Turning to Slides 19 and 20. In 1Q '26, our organic CapEx was RMB 770 million. In addition, we received cash proceeds of RMB 2.7 billion or USD 385 million from the sale of a small part of our equity interest in day 1, which is recorded in investing cash flow.

We also received cash proceeds of RMB 2.1 billion or USD 300 million from the issue of convertible preferred shares, which is recorded in financing cash flow. As a result of the capital recycling and new issue, we are now sitting on over RMB 19 billion or USD 2.7 billion of cash and time deposits. This is an ideal situation to be in as we prepare for a new growth phase.

Turning to Slide 23. Our net debt to last quarter annualized adjusted EBITDA has decreased from 6.8x at the end of 2024 to 4.7x at the end of the first quarter of 2026. As we step up our investment, this ratio will increase to between 5 to 6x, which we consider an acceptable level.

Finishing on Slide 25, we maintain our full year guidance unchanged.

Now we'd like to open the call to questions. Operator?

[Operator Instructions]. And now we're going to take our first question, and it comes from the line of Yang Liu from Morgan Stanley.

I would like to hear your comment on the pricing for the data center business. I think Dan previously mentioned that the overall pricing environment is stable. But could you please break it down to different market or locations? Because from time to time, we hear that in certain markets, it's a little bit undersupply and also in certain markets, there are some relative aggressive bidding from telcos, et cetera. Could you please comment on the pricing in different markets, please?

Yes, Liu, I think this is -- I think in the last earnings call, we already said the new incremental demand, which is driven by the AI, right, large-scale data center demand. In general, I mean, the price is pretty stable, number one.

Number two, I think, of course, in the whole market, you cannot stop some bidder, right, they use some price tools to try to win. But it's not normal, right? It's not normal. And it's maybe -- in my view, in some regions, some deal is a onetime. It's not represented the whole market situation. Our thought remain what we experienced last quarter is quite stable.

Our thought is that it remain what we experienced last quarter, exactly, so quite a stable, yes.

Now we're going to take our next question. And the question comes line of Gokul Hariharan from JPMorgan.

My question is basically on the development cost, Dan, I think you mentioned roughly 20% -- sorry, RMB 20 million or $3 million per kilowatt, if I remember right. That number sounds a lot lower than what it used to be a few years back when you updated those numbers, I think. Could you talk a little bit about what the -- what are the variables that have changed? Is it mostly the location that has really changed? Or are there any other factors that have really changed to kind of reduce that development cost over the last maybe, I think, 2 to 3 years?

I would say that the unit development cost on a like-for-like basis, whether we're talking in established markets or new markets has decreased by about 15% over the past 3 years. That would be the case with the MEP, the mechanical electrical plant, which accounts for about 70% of the total development cost.

I'd also say that the land, concrete, steel and construction cost has been quite stable if we measure it on a per square meter basis, unit cost is relatively flat, but the power density has increased. So if we were to measure that part on a per kilowatt basis, it might appear to have come down as well. So that's why I think overall, on a per kilowatt basis, the decrease is about 15% over 3 years.

Yes. I try to add a couple of things. I mean, number one, the scale is unprecedented, right? So scale also makes it cost a bit lower, right? That's very nature. I mean this is number one. Even for a vendor perspective, scale -- that's larger scale gives a lot of the manufacturing product company a lot of benefit, right? So they're willing to reduce the cost -- reduce price. This is number one.

And number two, I think a lot of the AI data center, this is compared with the previous cloud, the architecture-wise also changed a lot. So this is another reason to drive down the cost, right? So that's 2 more reasons.

Now, we are going to take our next question. And the question comes from line of Sara Wang from UBS.

So I have one question regarding first quarter CapEx. So I think the first quarter CapEx is RMB 770 million. So it is a little bit modest given the strong orders we signed year-to-date and especially given the majority of the new orders should be new builds. So may I ask what's the reason behind the gap?

Sara, I would point you to our full year CapEx guidance, which remains unchanged. I mean the timing of incurring CapEx per quarter is not that significant, right? The first quarter is Chinese New Year, and it tends to be historically slightly below the level of the other 3 quarters. So I can't really -- have no other more fundamental explanation than that.

Now we're going to take our next question. And the question comes line of Frank Louthan from Raymond James & Associates.

Of the roughly RMB 3 billion that you discussed in capital you're spending, how much of that will you be funding yourself versus maybe with some JV investors or with capital recycling from some of your other assets?

Frank, it's Dan. Let me just go over these numbers again and make sure everyone is clear. So William was talking about having a sales plan of 500 megawatts to 800 megawatts over the next 3 years. That's our current view. And if you apply the logic of what I said is RMB 20,000 per kilowatt or USD 3 million per megawatt, that's how you end up with total CapEx over 3 years of between RMB 30 billion to RMB 50 billion.

So if we take the midpoint of that, say, RMB 40 billion, historically, we have financed our investments quite conservatively with around 60% project debt to total development cost. So we would be able to obtain and draw down on about 60% to RMB 40 billion, which is RMB 24 billion of new debt. So that would leave RMB 14 billion, which is less than USD 2 billion that we have to finance.

We have several different sources for that. We have our operating cash flow, which is -- last year was nearly RMB 3 billion. And we have our ongoing asset monetization program, which we're trying to build up step by step. And we also have $2.7 billion of cash on our balance sheet. So I think we're in a strong position to finance that level of investment and other options may arise, as you point out, development partnerships and so on.

Now we are going to take our next question. And the next question comes from the line of Ellie Jiang from Macquarie.

I just wanted to get a sense on the new bookings trajectory. The year-to-date 340 megawatts new bookings seems to be very encouraging. Considering the current token consumption and how AI agents are significantly boosting that compute demand, how would you kind of evaluate that upside surprises on the current scale?

Potential to upside.

Yes. We -- number one, I think we are -- 500 megawatts, we are very confident for this number with new booking. Definitely, that's the base case. We are looking at a more high number booking. But it's too early to say what kind of level we can reach. We will try to -- because we are still very -- we remain very disciplined to select the order in terms of the move-in price and customer types. So this is -- in general, I think it's -- we are very confident we can do more. But even though we still want to do high-quality order.

Got it. And if I may, just a quick follow-up. Would it be possible for you guys to consider kind of doing some of the Neocloud business models as well? Because it does seem like some of the peers are trying to accumulate more resources on the compute side. So that was being perceived as approach to boost the MSR or revenue in general. Is that something that we're considering as well?

Yes. I think the Neocloud actually is not something new in China already. Historically, they are a lot of big platform GPU service provider customer already, right? We already serve them indirectly, right? So this is number one.

But number two, I think we are -- from a long-term perspective, we also build -- start to build some relationship with them. So far, we haven't do any business with them, and we will see because in terms of -- maybe we can -- as I said, we will maintain our very discipline in terms of the financial return and the risk, everything, right? So if some Neocloud, high-quality Neocloud, we're willing to do something with them, start to build some relationship.

Now, we are going to take our next question. And the question comes from line of Timothy Zhao from Goldman Sachs.

Regarding the pace of the growth additional area utilized. Just wondering after the first quarter, can you share your latest outlook for the rest of this year in terms of the move-in pace and what are the key moving factors that may affect the rate ramp up?

Timothy, I couldn't hear you clearly, but yes, I'm told you you're asking about the move-in pace. So I did address that in the prepared remarks. As you know, it was 16,000 square meters in the first quarter. It will be a lower number in the second quarter, and then it will rebound I would say, to around 20,000 square meters in the third quarter of this year and the fourth quarter of this year.

And next year, we will see a significant step up, but it will be in the second half of 2027, in the third and fourth quarter of 2027. But if we look at 2026 and 2027 as a whole, I think the move in this year will be somewhat over 70,000 square meters. And then next year's number is going to be very substantially larger than that, maybe double something of that order of magnitude.

Sure. Understood. Can I ask a follow-up, if I may? Just wondering, I think behind this assumptions, I think we see factors. So like how much of that is contributed by the domestic chip versus the imported chips. I just wondering if you can share more color.

Yes. I think I'm not sure it's your question. I mean import chips will affect our movie, right? Is that your question?

Yes.

Okay. Frankly, this year's forecast is not based on any import chips. So all based on the domestic chips supply chain. So it will not impact our current estimation. So as import coming, maybe some upside, who knows.

Now we're going to take our next question. And the question comes from the line of Daley Li from Bank of America Securities.

My question is about our land and power resources. We have secured quite strong resources in 1Q. And are we planning to expand our resources in the following quarters? And if we have the plan in future and what kind of area we would focus on?

I think last quarter, we already answered the question. We will continue to develop the new market and established market as well because in China, what happened is the training and the inference demands all happening in the same time. So I think we are, we try -- because everybody knows GDS is a platform player, not just a project player, right? So we try to -- try to fulfill all the kind of AI demand, whatever is training or in the future or, let's say, inference. So we try to catch up and well positioned to catch up a different pace of the AI demand.

Thank you. Due to time limit of today's call, I would like now to turn the call back over to the company for any closing remarks.

Thank you once again for joining us today and see you next time. Bye.

Thank you.

This concludes today's conference call. You may now disconnect your lines. Thank you.

Hello, ladies and gentlemen. Thank you for standing by for GDS Holdings Limited Fourth Quarter and Full Year 2025 Earnings Conference Call. [Operator Instructions] Today's conference call is being recorded.

I will now turn the call over to your host, Ms. Laura Chen, Head of Investor Relations for the company. Please go ahead, Laura.

Thank you. Hello, everyone. Welcome to the Fourth Quarter and Full Year 2025 Earnings Conference Call of GDS Holdings Limited. The company's results were issued via Newswire services earlier today and are posted online. A summary presentation, which we'll refer to during this conference call, can be viewed and downloaded from our IR website at investors.gdservices.com.

Leading today's call is Mr. William Huang, GDS Founder, Chairman and CEO, who will provide an overview of our business strategy and performance. Mr. Dan Newman, GDS CFO, will then review financial and operating results.

Before we continue, please note that today's discussion will contain forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Forward-looking statements involve inherent risks and uncertainties. As such, the company's results may be materially different from the views expressed today. Further information regarding these and other risks and uncertainties is included in the company's prospectus as filed with the U.S. SEC.

The company does not assume any obligation to update any forward-looking statements, except as required under applicable law. Please also note that GDS earnings press release and this conference call include discussions of unaudited GAAP financial information as well as unaudited non-GAAP financial measures. GDS press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures.

I'll now turn the call over to GDS Founder, Chairman and CEO, William Huang. Please go ahead, William.

Thank you. Hello, everyone. This is William. Thank you for joining us on today's call. 2025 was a great year for GDS in terms of performance. We recorded 11% growth in both revenue and adjusted EBITDA. We beat the top end of our adjusted EBITDA guidance. And with the contribution from asset monetization, we were free cash flow positive. AI in China has really taken off. Increasing availability of domestic high-performance chips is a key enabler. All our major customers are investing in hyperscale computing infrastructure to support AI adoption. As a result, we are seeing robust recovery in data center demand across both new markets and established markets. It is an exciting time to be a data center company again.

To address this opportunity, we are building up our resources and funding. On the resource side, we are working on a 3-gigawatt pipeline comprising big clusters in new growth market. This will complement the 700 megawatts of powered land, which we are holding for future development in low latency established markets. On the funding side, we have increased our cash reserves to over USD 2.8 billion, including proceeds from the recent sell-down of our stake in DayOne and the CPS new issue. Furthermore, our success last year in opening up channels for asset monetization gives us a competitive advantage in accessing equity onshore.

During 4Q '25, our gross additional area utilization utilized was around 23,000 square meters. For the full year, gross move-in was over 86,000 square meters, our highest ever level. Our move-in target for 2026 is similar to last year. However, as our bookings step up over the current year, it will lead to higher move-in 1 year forward. During 4Q '25, our gross additional area committed was over 21,000 square meters. For the full year, our new bookings was over 96,000 square meters or over 300 megawatts, 3x the level of the past 3 years.

In 2026, we are aiming over 500 megawatts of gross new bookings, another big step-up from the last year. We expect 60% to 70% of new business to come from AI. So far this year, we have already secured 200 megawatts of new orders plus over 500 megawatts of MOUs, which are a strong indicator of future commitments. This 700 megawatts of total demand comes mainly from 3 of our largest customers. It's a great start towards our full year sales target.

Now the domestic chip supply is more certain. We are moving fast to secure multi-gigawatts of additional powered land in new markets, which can support big cluster deployments. We are focusing on the 3 locations: Horinger in Inner Mongolia, Zhongwei in Ningxia province and Shaoguan in Guangdong province. We have already won over 400 megawatts of new orders and MOUs for these locations. These new growth markets, all of which are official national hubs integrate well with our existing platform, enabling us to serve the different needs of our diversified customer base. We are very excited about the opportunities in front of us and look forward to growth in sync with China's AI development.

I will now pass on to Dan for the financial and operating review.

Thank you, William. Starting on Slide 17. In FY '25, revenue and adjusted EBITDA increased by 10.8% year-on-year. During the year, we completed 2 asset monetization transactions, an ABS in 1Q '25 and the C-REIT IPO in 3Q '25, following which we deconsolidated the underlying data center project companies. If we add back the deconsolidated revenue and EBITDA, the pro forma growth rates were 13.2% for revenue and 14.2% for adjusted EBITDA.

Turning to Slide 20. Our MSR per square meter has been declining due to a combination of lower market selling price and change in location mix to include more edge of town sites and going forward, new growth markets. Comparing 4Q '25 with 4Q '24, the decrease was 2.4%. At the same time, we have also seen a comparable decrease in unit development costs. As a result, the overall yield on our portfolio as measured by adjusted gross profit divided by gross PP&E, excluding construction in progress, has remained steady at around 11%. Looking forward, we expect further MSR reduction of 3% to 4% by the end of 2026 due to the same combination of factors. However, the yield on our new investments in both established and new markets continues to be in the 10% to 11% range.

Turning to Slide 23. In 2025, our organic CapEx was RMB 4.7 billion, in line with our guidance. Net of the cash proceeds from asset monetization of RMB 2.3 billion, our CapEx was around RMB 2.4 billion.

As shown on Slide 24, our operating cash flow for the full year was around RMB 3.4 billion. The significant improvement year-on-year was helped by a reduction in AR days from 109 in 4Q '24 to 82 days in 4Q '25 as a result of our tight control of collections. After taking into account asset monetization proceeds, we achieved positive cash flow prefinancing of RMB 1 billion. In 2026, we are guiding for organic CapEx of around RMB 9 billion, which corresponds to our 500-megawatt plus sales target. This year's CapEx will contribute to next year's growth. We have started work on a follow-on asset injection into our C-REIT. We have selected an asset which is larger than the seed asset for the IPO. We aim to complete the asset injection in the second half of 2026, if possible. However, we have not included any assumed proceeds in our CapEx guidance.

Turning to Slide 25. During the first quarter of 2026, we raised $385 million through the partial sell-down of our stake in DayOne. After the sell-down, our remaining stake is worth $2.2 billion or $11 per GDS ADS benchmarked to DayOne's Series C new issue price. We also issued $300 million of convertible preferred shares to Huatai Capital Investment. As a result, we are now sitting on nearly RMB 20 billion or $2.8 billion of cash. This is an ideal situation to be in as we prepare for a new growth phase.

Turning to Slide 26 and 27. Our net debt to last quarter annualized adjusted EBITDA decreased from 6.8x at the end of 2024 to 5.8x at the end of 2025. The decrease is mainly due to a combination of positive cash flow prefinancing, the deconsolidation of debt of the project companies sold to the ABS and C-REIT and the proceeds of the equity capital raise, which we did in 2Q '25. If we add back the purchase of time deposits, which is included in our reported investment cash flow and the proceeds of the capital recycling and new issue in 1Q '26, our net debt-to-EBITDA ratio decreases to 4.8x. At the beginning of 2023, we set a target of achieving positive cash flow prefinancing and net debt-to-EBITDA of below 5x within 3 years. Looking back, it was an aggressive target, but I'm pleased to say that we made it.

Turning to Slide 29. For FY '25, we achieved the midpoint of our revenue guidance and beat the top end of our adjusted EBITDA guidance. For 2026, we expect total revenues to be between RMB 12.4 billion to RMB 12.9 billion, implying a year-on-year increase of between approximately 8.5% to 12.8%. For adjusted EBITDA, we expect between RMB 5.75 billion to RMB 6 billion, implying a year-on-year increase of between approximately 6.4% to 11%.

As a result of the asset monetizations during 2025, the year-on-year growth rates are not directly comparable. If we add back the forecast revenue and adjusted EBITDA for the data center project companies sold to the ABS and C-REIT, the implied growth rate of our pro forma revenue and adjusted EBITDA guidance is approximately 1.6 percentage points higher. This is shown on Slide 30.

We'd now like to open the call to questions. Operator, please?

[Operator Instructions] Our first question comes from the line of Yang Liu from Morgan Stanley.

First, congratulations on the very strong booking year-to-date. I have 2 questions. The first is about the conversion from MOU to contract. What is the timetable behind this kind of conversion? Or what is the potential risk for this kind of conversion? Or what do we need to do or what do our customers need to do behind this kind of conversion? That's my first question.

And the second question is about the competition in the new key focus area like Inner Mongolia, Zhongwei and Shaoguan. We know that GDS has been far leading in Tier 1 market. But how about in those new focus areas? Are we seeing more competition or less competitor in those places? And what is the GDS advantage there?

Okay. The first question, I think this is a high certainty to convert to our order. This is number one. In terms of the timing, I think it's within 2 quarters. I think it's a high chance we can convert to the real contract. So we are very confident on that, the number one. Number two, I think in terms of the new market, right, new market, I think the current market before we step in, I think there's some data center already -- data center operator already there. But it's never too late because now the government set up a very high barrier right now.

Number one, they will measure -- I mean, they will measure the criteria for they choose a partner to able to acquire the land, they have a couple of the criteria. Number one, they will seriously look at the company's track record. Second, they will make sure you have the customer commitment behind you. And number third, they also will look at your financial capability as well. So this is the current government, how they look at a partner, a potential partner. So I think if that's the case, which it is, it's already set up a high barrier. And we think we will be able to still sit on the leading position.

Our next question comes from the line of Jonathan Atkin from RBCCM.

So you referenced a lot of your orders and current interest being AI-oriented. Can you talk about a little bit about the non-AI kind of traditional cloud, even enterprise types of workloads and the demand trends that you're seeing there? And then any further color around the types of AI workloads? And would you associate it primarily with large foundation models or inference or what?

I think the majority, I mean, the demands are driven by a very clear -- driven by the AI, right, GPU type data center demand. Of course, I think the traditional cloud still grows and they are more associated with the AI demand. So I think that's the kind of profile. It's not like before 100% driven by the traditional cloud. Now the cloud still grows, but more associated with the AI. That's a slight change in the driver, right? So what's the second question?

Type of AI application.

Inference or machine learning, the workload.

The workload of AI.

I think -- of course, the training still continue. That's for sure, because China is still behind the U.S., right, for a couple of years, and now it's catching up. So it looks like the training is still -- is a key driver to drive the data center demand. But in the meanwhile, I think the large language model owner, they start to, I mean, a lot of demand is also driven by the inferencing right now. That's why our Tier 1, let's say, traditional market still get the growth in the last year. And we are also -- we also estimate this year still have a very strong demand from both AI type and inference type and cloud.

And then I wonder if you could maybe just touch on the competitive environment, and it probably varies a little bit by region and market and so forth. But in terms of other projects that your peers are pursuing, how would you characterize competition that is meeting the demand? And is that at all different from the last time you gave us an update?

General -- competitive environment in new markets, competitive...

Yes, I think if you are aware, if GDS seriously step in a new market, we definitely will dominate the market. That's how we look. That's our behavior, right? So otherwise, we would not step in, right? We definitely get ready to step in and give. Over time, I think we will take the absolutely leading ship in this region. So I think the competition in AI in China, data center competition in China just a start for AI. So I think it's a good timing to step in, especially if you have enough financial capability that's more easy to win the battle, right?

Our next question comes from the line of Sara Wang from UBS.

Congratulations again on the really solid new bookings. So I have two questions. The first one is on supply. So I still remember that earlier last year, management took a very rational and cautious stance on taking new orders because back then, there were some uncertainties around chip supply. However, given the strong order and MOU momentum year-to-date, should we interpret that as a sign that chip supply has improved meaningfully? And then or in other words, from a supply side perspective, are there any factors constraining project delivery? So that's my first question.

And then my second question is that I noticed that GDS powered land reservation has increased from 900 megawatts last quarter to 3.7 gigawatts this quarter. So may I ask where are the main locations of the new resources? And how does the project returns in this new area differ from our existing projects?

Yes, I think we have the different view, right? So we are always looking more deeply to the industry. So that's why we are also. That means we are very disciplined to CapEx investment. So last year, we are slightly conservative because of the chip supplies still were not that certain. So this year, the certainty is more improved, right? I think in terms of the U.S. export policy changes and the domestic chips also catching up. If you recall, a couple of quarters ago, when we talk about this, we stay on the very -- we would always say we wait and see, right? So that's the right strategy to more disciplined to making CapEx investment.

Now we are much comfortable because the whole environment change adapted to a more positive, more certainty. So I think that's -- we think it's the right timing to step in, in a big way, right? Land bank, I think in terms of land bank, I just mentioned in my script, right? So I think there's Horinger in Inner Mongolia and Zhongwei in Ningxia province and Shaoguan, it is in the Guangdong province. I think that's all the national hub data center hub, right? So I think the location will be great for future and well recognized by all our existing customers.

And how should we think about the project returns?

Sara, as I said during the script, we're still able to generate a simple cash-on-cash yield of 10% to 11%, whether we're taking on new business in established markets or new markets. And with our business model of developing, ramping up and holding for the qualification period and then monetizing, that yield is sufficient for us to realize a return on equity above 20%.

Our next question comes from the line of Gokul Hariharan from JPMorgan.

My first question is on the 200-megawatt order that you've already secured. Could you talk a little bit about the nature of the urgency of the projects, obviously, given AI demand seems to be accelerating. When do you expect to deliver this to customers? Is it also more like an accelerated schedule like we saw with the 150-megawatt order that we saw last year? That's my first question.

And secondly, just trying to understand a little bit on the realized MSR trends. Previously, we were expecting MSR to start to flatten out a little bit in 2027. But now obviously, our location mix is probably changing a little bit in response to some of the new demand trends. So Dan, maybe could you help us understand how MSR is likely to shape up, let's say, 1 or 2 years out from now, the realized MSR based on the contracts that you're signing right now?

Sure, Gokul. The 200-megawatt new orders, you can assume for forecasting that it will take us 4 quarters on average to deliver. And then it will be a 4-quarter ramp-up. This is faster than historically when we were doing the more traditional cloud business, and it's consistent with our parameters in terms of selecting new business. The MSR decrease, it will continue beyond next year. I think in 2028, it's probably 3% to 4% again.

But the offset in terms of higher volume growth is going to lift our overall growth rate. In 2026, I think William said that we're expecting our move-in to be similar to last year, which is in the sort of 80,000 to 90,000 square meter range. But if all goes to plan in terms of meeting our sales target this year, we'll be looking at a move-in, which could be like double that next year. So it's a combination of those 2 factors is going to drive our growth higher.

Understood. Just one follow-up on the locations as we are adding some of these newer locations, which are a little bit more remote sites, but obviously, the new data center centers for the country. Is the customer concentration high in some of these new locations? Like previously, when we went to build-to-suit, I think it became very much like very customer specific. How should we think about the customer concentration in some of these new locations like Shaoguan or Inner Mongolia, et cetera?

Yes. I think the global views everywhere, every data center company now is getting more concentrated in terms of customer. I think that maybe global -- in China, maybe top 3. That's a trend, right? In the global point of view, I think it's 4 or 5 companies, right, everybody is at the trend. If you position you are a hyperscale data center operator, a dev ops center operator, right? It's not a traditional colo. If you look at the colo business, this looks like more diversified, right? But this is the reality.

I'd just add that the contract length for this kind of business is certainly at the long end or longer than what we typically have been before. So quite often find 10-year contracts, which I think derisks the investments in these projects.

Yes. I think in terms of customer number, GDS already had 1,000 customers, right? So of course, the new demand is mainly driven by the top 3 AI player in China, right?

Our next question comes from the line of Frank Louthan from Raymond James & Associates.

This is Rob on for Frank. So just looking at the demand and the bookings, what's the growth in demand that you're seeing from your nondomestic Chinese customers year-over-year? And what would you say is the outlook for that going forward?

Rob, the demand is from Chinese customers almost entirely. I think the market opportunity is around 3 gigawatts per annum, concentrated, as William said, in the very largest customers. So we talk about 500-megawatt sales target, put it into the context of that kind of scale of addressable market.

Our next question comes from the line of Timothy Zhao from Goldman Sachs.

Two questions here. One is really on the resource expansion that you mentioned about the 3 gigawatts pipeline into a new growth market. Just wondering, I think, between the 3 key hubs that you mentioned Inner Mongolia, Ningxia and Guangdong, what is the like difference or similarities among those regions in terms of customer preference or the IT workload, the pricing, et cetera?

And the follow-on question related to that is that what is the regional breakdown between those 3 key hubs out of the 3 gigawatts pipeline that you have or out of the 500 megawatts MOUs that you disclosed year-to-date? And second question is on the CapEx. I think given that you have very strong sales momentum year-to-date and to convert that 500 megawatts into contract probably around 2 quarters, do you see any possibilities to further revise up the CapEx guidance for this year?

I think there are 3 new market, I think the work is similar. Major workload is still training, plus partially is inference, right? And in the meanwhile, we just mentioned in the traditional market, which is low latency market, right? We also got a lot of order from our customer, the large language model customer because they start to -- already start to deploy the inference workload. So I think it's quite balance. In general, I think maybe it's 65% to 70% will go to the new market and still 30% to 40% is go to traditional market, right, which we used to call the Tier 1 market. That's sort of the difference of workload.

Tim, on your question regarding CapEx, as I mentioned earlier, you should assume that it takes us 4 quarters to build because in many cases, we're talking about new build in a site which is where we have no previous presence. So we commenced construction in 1Q '26, it's for delivery to customer in 1Q '27. So as we win new business going through this year, that will lead to more starts. But I think the CapEx guidance of RMB 9 billion is adequate. I would not expect to change that.

Our next question comes from the line of Ellie Jiang from Macquarie.

I just have one question that's more longer term. Just now management talked about the data center demand in China is just at the beginning of picking up. Would it be fair if we look in the next 3 to 5 years to kind of narrow the U.S. trajectory, I mean, especially on kind of how the large hyperscalers have been accelerating the CapEx deployment pace? And do you see similar commitments from our key customers? And lastly, how do you really see the longer-term kind of market size and positioning in that trajectory?

Yes. If you look at for our estimation, I think that yes, China demand will like a growth trajectory more like the U.S. because it's just behind a couple of years, right? That's happening right now. So if the -- last year, we just -- we talked about that demand is already there. It's all about just about the chip supply, right? Now it looks like it is getting much better, more positive right now, more certainty in terms of the chip supply. So it will go -- it will same pace to similar pace as the U.S., right? So I think the CapEx, if you look at the last 2 years, all the big AI company, tech company continue to raise their CapEx guidance, just like what happened in the U.S.

Got it. And if I may, sorry, if we continue on that route, would it be possible at one point because just now management talked about the MSR still declining slightly all the way until 2028. But would it be possible at some point, we still see hyper demand really building in, especially given how the open clause or all these agentic integration seems to be driving token consumption by 10x or even 20x. Then at some point, would it be possible for us to see even stronger pricing power down the road?

Yes, it could be. I think it could be. I think if you look at the U.S. price let's say, adoption profile in the last 5 years, it's -- if you look at it 5 years ago, the U.S. price used to be down to USD 60 per kW, right? Now it's 3x average, right? So 2x or 3x average. So that's profile, I think that will be a high chance it will be, right?

I give you a sense in the whole, I mean, western part of China, the total power capacity now -- as of now, just 30 megawatts -- gigawatts to available to supply the future growth. In general, it is still limited, right?

Our next question comes from the line of Daley Li from Bank of America Securities.

Congrats on the strong orders for year-to-date and for the MOU. My first question is about the 500 megawatts MOU. Could management introduce is this mainly for 2026 -- sorry, 2027? And how many years -- what could be the time horizon or time period? Is it like 1 or 2 year or like 3-year contract -- potential contracts?

My second question is about the CapEx and the financing, the update. And given the RMB 9 billion CapEx, how do we see the financing and the need and given we have strong cash on hand right now?

Yes. Sorry...

500-megawatt MOU, the delivery time...

Yes, the delivery time is, as I mentioned before, is 4 quarters. So the business that we're winning in the current first quarter of this year is going to contribute to move in next year. And it's logical that if we meet our sales target of 500 megawatt then next year's move-in could be around double current year's move-in. It would flow through like that. The contract lengths, I think much longer than what you mentioned. It will be 7 to 10 years, mostly at the 10-year end of that.

For financing, last year, we were self-funding in China. But that was achieved when our CapEx was RMB 5 billion, and we were able to complete 2 asset monetizations last year in ABS and the C-REIT. So now our CapEx has gone up to RMB 9 billion, and we have a plan for an asset monetization that we can't be sure, but we aim to complete that in the second half of the year. I don't know whether we will be self-funding in China. Let's say, CapEx is RMB 9 billion, operating cash flow is RMB 3 billion and maybe there's some proceeds from asset monetization. If there's anything left, it will be very easy for us to finance that in the traditional way with project debt.

There are no further questions at this time. So I'll hand the call back to Laura for closing remarks.

Thank you all once again for joining us today, and see you next time. Bye-bye.

This concludes today's conference call. Thank you for participating. You may now disconnect. Speakers, please stand by.

Hello, ladies and gentlemen. Thank you for standing by for GDS Holdings Limited Third Quarter 2025 Earnings Conference Call. [Operator Instructions] Today's conference call is being recorded. I will now turn the call over to your host, Ms. Laura Chen, Head of Investor Relations for the company. Please go ahead, Laura.

Thank you. Hello, everyone. Welcome to the third quarter 2025 Earnings Conference Call of GDS Holdings Limited. The company's results were issued via Newswire services earlier today and are posted online. A summary presentation, which we'll refer to during this conference call, can be viewed and downloaded from our IR website at investors.gdsservices.com.

Leading today's call is Mr. William Huang, GDS Founder, Chairman and CEO, who will provide an overview of our business strategy and performance. Mr. Dan Newman, GDS CFO, will then review the financial and operating results.

Before we continue, please note that today's discussion will contain forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Forward-looking statements involve inherent risks and uncertainties. As such, the company's results may be materially different from the views expressed today. Further information regarding these and other risks and uncertainties is included in the company's prospectus as filed with the U.S. SEC.

The company does not assume any obligation to update any forward-looking statements, except as required under applicable law.

Please also note that GDS' earnings press release and this conference call include discussions of unaudited GAAP financial information as well as unaudited non-GAAP financial measures. GDS press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures.

I will now turn over the call to GDS Founder, Chairman and CEO, Mr. William Huang. Please go ahead, William.

Thank you. Hello, everyone. This is William. Thank you for joining us on today's call. During the third quarter, our revenue increased by 10.2%, and our adjusted EBITDA increased by 11.4% year-on-year, maintaining the healthy growth trend since our business began to recover last year.

During 3Q '25, our gross additional area utilized was around 23,000 square meters. We are on track to achieve our highest every year of move-in. We continue to deliver the long-term backlog. In addition, we are now delivering the 40,000 square meter, or 152-megawatt order which we won in the first quarter of this year. By being selective with new business, we have successfully shortened the book-to-build period and brought down our backlog. Nonetheless, we still have visibility for over 70,000 square meters of move-in from the backlog next year.

Our total new bookings for the first 9 months is 75,000 square meters or 240 megawatts. We expect to achieve nearly 300 megawatts for the full year, which is a big step-up from the level of the past few years. Around 65% of our bookings in 2025 are AI-related. Nonetheless, AI demand in China is still at a very early stage. If we look at the big picture, the domestic tech industry has reached a critical juncture with major players making unprecedented financial commitment to AI infrastructure. This marks a definitive end to the previous downturn and signals the beginning of a robust recovery for the data center sector.

All of our major customers are committed to the massive scale of this new investment cycle, with CapEx plans of hundreds of billions, underscoring the intensity of the new AI arms race. Leading local chip companies are making continuous development progress in terms of performance, efficiency and capacity. The growth of the domestic chip segment will secure the long-term growth of the AI infrastructure industry. We have unwavering confidence in the AI demand to come basis on the development and the ramp-up of domestic technologies. We believe that new bookings in the coming years could be better, and this is what we are preparing for in our strategic plan.

There are 2 essential ingredients to win big in AI, powered land and access to capital. We have already secured around 900 megawatts of powered land in and around Tier 1 markets, which is suitable for AI demand, particularly for AI inferencing. In addition, based on our communications with our customers, we are in the process of securing more powered land in complementary locations, and we believe that 900 megawatts will not be enough.

On the financing side, we recently completed first IPO of a data center REIT in China. The transaction was a huge success. We intend injecting more assets in the REIT next year and establishing a continuous pipeline of asset monetization. The REIT gives us a significant competitive advantage in terms of accessing capital from the domestic equity market. It enables us to monetize assets efficiently, repeatedly and at the lowest possible cost.

The China market is at an inflection point. The outlook for the data center industry is very exciting. Our market position is as strong as ever. Over the past few years, we have taken a conservative approach. We improved our asset utilization and significantly strengthened our balance sheet. Going forward, we will maintain our financial discipline while, at the same time, taking a more aggressive approach to new business.

I will now pass on to Dan for the financial and operating review.

Thank you, William. Starting on Slide 15. As William mentioned, in 3Q '25, our reported adjusted EBITDA grew by 11.4% year-on-year. At the end of 1Q '25, we deconsolidated the data center project companies, which we sold to the ABS. And then during 3Q '25, we deconsolidated the data center project companies, which we sold to the C-REIT. In order to present a consistent trend, we have adjusted historic numbers to take out the EBITDA contribution of the deconsolidated companies for the first 9 months of 2025 and for the comparative period. On this pro forma basis, our adjusted EBITDA for the first 9 months grew by 15.4%.

Turning to Slide 16. Our C-REIT started trading on the Shanghai Stock Exchange on the 8th of August. As of yesterday's close, the C-REIT units were priced at RMB 4.375, 45.8% up from the IPO price. At this level, the C-REIT is trading on 24.6x EV to the projected 2026 EBITDA as disclosed in the C-REIT offering memorandum. The implied dividend yield is 3.6% based on the projected cash available for distribution, also as stated in the offering memorandum.

It is our strategic objective to grow and diversify our C-REIT so that it is a viable option for us to recycle capital on a repeated basis, thereby unlocking value for GDS shareholders and freeing up funds for new investment. Under current regulations, we are permitted to apply for approval for the first post-IPO asset injection 6 months after the IPO date, i.e. during 2Q '26. Thereafter, it will take some time to complete the regulatory review process.

For the first IPO -- post-IPO asset injection, we are preparing assets with a target enterprise value of around RMB 4 billion to RMB 6 billion. This compares with an enterprise value of RMB 2.4 billion for the assets which we injected into the C-REIT at IPO.

With the creation of the C-REIT platform, we have the opportunity to invest in new data centers, ramp up, operate and then, once the track record qualifies, to monetize over a 5- to 6-year investment cycle. Even if we take a very conservative view on potential future exit multiples into the C-REIT, the return on new investment is still very compelling. This could not have happened at a better time as we address the upcoming AI demand wave. We think it's a game changer.

Turning to Slide 17. For the first 9 months of 2025, our organic CapEx was RMB 3.8 billion. We still expect our organic CapEx for the full year to be around RMB 4.8 billion. However, net of the cash proceeds of the asset monetization, our CapEx will be around RMB 2.7 billion.

As shown on Slide 18, our operating cash flow for the full year will be around RMB 2.5 billion. Therefore, after taking into account the asset monetization proceeds, our China business is almost self-funding.

Turning to Slide 19 and 20. Our net debt to last quarter annualized adjusted EBITDA multiple decreased from 6.8x at the end of 2024 to 6.0x at the end of 3Q '25. The decrease is mainly due to the cash proceeds of the asset monetization and the deconsolidation of debt of the project companies sold to the ABS and C-REIT as well as the offshore equity capital raise, which we did in 2Q '25.

We are benefiting from the favorable interest rate environment in China, with our effective interest rate dropping to 3.3%.

Turning to Slide 22. After 9 months, we are on track to achieve the midpoint of our revenue guidance and at or above the top end of our EBITDA guidance for the full year of 2025. Our growth rate during the current year has clearly benefited from the strong new bookings in 1Q '25 and a short book-to-bill period. This gives a clear illustration of how our growth rate can accelerate with a pickup in demand.

The relatively subdued new bookings since 2Q '25 will affect our growth rate next year. However, in our internal projections, we foresee higher bookings next year, leading to gross acceleration thereafter.

We'd now like to open the floor to questions. Operator?

[Operator Instructions] Our first question comes from the line of Yang Liu of Morgan Stanley.

I have 2 questions here. The first one is regarding the China market inflection. As William just mentioned, the China market is approaching the inflection point. What do we need to see to see that really happen in the near future? And in terms of your strategy to go a little bit more aggressive in China, could you please elaborate more, for example, with location or what type of project, et cetera, are you planning?

The second question is regarding the overall investment profile because now we have a C-REIT platform, and it is a very effective way to recycle capital. And what is the new overall investment return with C-REIT scheme?

Okay. I think number one question is, yes, I think how to explain the aggressive approach. I think what we see in the market, demand is very strong in China. I think our customer announced their big investment in the next 5 years. I think now another signal is domestic chip is catching up. Just as what I mentioned, I think in terms of the efficiency, chips efficiency and production capacity, I think they all improved a lot. That means the real data center opportunity is coming. So we are well positioned. As I just mentioned, we still have the large -- I think the largest land bank -- powered land bank in and around Tier 1 market. This is very good for the future inferencing.

Another is, I think, the China tech player, they will continue to do massive training. So I think in order to capture this opportunity, we will acquire more land in some very cheap power location and more -- as much close to, let's say, the Tier 1 city, yes. So I think this is our strategy. And we are -- a lot of the land acquisition is in process. And maybe something will happen, we can announce in next earnings call. This is number one.

Number two, I think Dan may can explain about the REITs.

Sure. The unit economics of the data center investment in China is very solid. The selling price is stable. The unit development cost has come down to a level which is very efficient. And this allows us to generate typically 11% to 12% cash on cash yield on new investment. What has changed is the way that we can look at and evaluate investment. If we take the approach of investing, which maybe takes 1 year to construct and then 1 year for the customer to move in fully, we have to hold the asset and operate for 3 years to establish the track record, which is required before assets can be injected into C-REIT. But then in the year -- the following year, which would be year 5 or 6, we can consider an asset injection. But even if we use a exit multiple, a cap rate, which is being very conservative compared with even where we IPO-ed our C-REIT. If we look at the IRR over a 5- to 6-year period, then it is in the low to mid-teens. And the levered IRR, the return on equity, is well into the 20s. I think fundamentally, this is very attractive.

Yes. I'll add 1 more point. I think we believe now is the right timing to step in the market because, number one, I think the price is more stable; number two, I think the development cost is almost at the bottom of the -- in terms of history, right? So I think this is the right timing to maintain very good return. It's the right timing, yes.

Our next question comes from Sara Wang of UBS.

Congratulations on the solid results. It's glad to hear that GDS is being more aggressive in acquiring new business opportunities. So I have actually 1 question, but 2 parts. So I think Dan just mentioned, we are expecting higher booking next year. So regarding this booking, does that include our potentially new powered land acquired in relatively -- like regions with relatively lower power tariffs? And the second question is that, if we are going into complementary markets on top of our 900 megawatts resources then how shall we think about the -- like is there any difficulties in acquiring new power quota? Because this year, we have heard [indiscernible] like NBRC, they're actually relatively rationalizing or controlling the new power quota release in China in general? Yes, that's my question.

Okay. The first question, I think that was new booking next year, right? We're not fully relying on the new acquisition of the land. Definitely, we will -- if we can success to secure the land, power the land, we can do more, right? So this is our focus base. The second -- what's the second...

How difficult...

I think power quota always -- I mean, in general, always not easy, right? But based on our track record and the reputation, I see a lot of governments willing to work with us. So for us, it's not that challenge for us. We have a lot of the experience in the past -- in the last 10 years to build up the right relationship with the government and the power company.

The next question comes from the line of Frank Louthan from Raymond James & Associates.

Can you give us an update on DayOne on private round funding and potential updates for a possible IPO? And then what is the outlook on your customers getting GPUs and be able to ramp their installs going forward? When do we expect that to crack open?

Yes, I think I answer and maybe Dan can add more color. I think -- I have to say, I think after Series B, I think DayOne is fully independent. So we cannot represent DayOne anymore since that time, right? But we still can give some highlight information, right, about DayOne because we quite enjoy the equity value increase, right, for our shareholders. I think all business in Asia Pacific and in Europe, which we already announced the market what we already stepped in, remain very, very good, very, very positive, and the demand still remains very, very strong.

So I think the DayOne's business is on the right track and could be better. So that's all what I can tell you. Maybe if you are interested, maybe we can introduce to the DayOne's right people to explain in more detail.

Okay. And on potential for additional installs to ramp?

Frank asked about the new business in DayOne I think.

Yes. I just can -- what I can tell you is they remain very, very strong, positive view for the future, yes. I cannot tell any detail more. I cannot represent -- this is a GDS earnings call, right? Sorry about that.

The next question will come from Michael Elias from TD Cowen.

So in the U.S., when we think about the training workloads that we're seeing, we're seeing gigawatt scale projects getting deployed. And I'm curious, when you think about what training will look like in China, are you seeing the opportunity to deploy at that kind of the scale, i.e., in the gigawatt range? And then second question is, can you give us an update, as you think about these AI data centers that you expect to build, what the time to build those data centers are and how that varies from traditional cloud data centers? And if I can squeeze it in, any notable constraints or long lead time items that we should be aware of?

I think scale-wise, I think our client talk about gigawatt level, I mean, new demand, right? So I think this is just like 3 years ago in -- what happened in the U.S. And the number-wise, we are talking -- every big player talk about gigawatt size new demand. So I think that it's catching up. That's what we have been seeing -- we have seen. So in terms of time to market, right, I think, in China, we can build very fast. I think normally 9 months to 12 months is very normal start from the piling to deliver, right? The extreme, I mean, case, we can build -- let's say, even built within 8 month. So that's our record in China.

Any bottlenecks or...

No, I don't think the -- in terms of development, yes, supply chain in China is not an issue.

The next questions will come from the line of Daley Li of Bank of America Securities.

I have 2 questions here. First one is about we got new orders for the China market, like a near 30 megawatts. Could you share what's the...

300.

Can you hear me? Sorry.

Go ahead.

Go ahead. Sorry. Yes.

Yes. Yes. Could you give some color about the AI exposure? What's the percentage from AI? And is this about inferencing model training for the recent order? Number two, for the second cone is about the -- we heard the China government gave some window guidance in 2Q this year to tighten the data center supplier in China? And do you see any impact to us and to the market?

Yes, I think, new order from -- Yes, go ahead.

Okay. In our prepared remarks, we commented that we will probably reach nearly 300 megawatts in terms of new bookings for the whole of 2025. I think we hit 240 megawatts up to the end of the first quarter, and there's some good new business in the fourth quarter. We also stated that, by our estimation, around 65% of the new bookings this year are AI related. We are -- only have a presence in Tier 1 markets. So that is AI in Tier 1 markets. So that's going to be mainly AI inferencing or it can be a combination of AI inferencing and training, and it's being deployed within the established cloud regions and cloud availability terms.

The second question was...

Window guidance about the carbon quota. I think this has always happened in the Tier 1 market, right? So -- but we are lucky. We already prepared for that. And that's why I mentioned we still have almost 900 megawatts powered land. This power is all gathered carbon quota in or near Tier 1 market. It's very difficult to apply new around the Tier 1 market. But in a remote area, I think I didn't hear any about the window guidance because the power in those place, it's -- the big problem is how to sell, right? It's not -- so the power is -- capacity is very large in a remote area. So get the power, I think it's not very, very difficult. And the local governments are very encouraged the data center -- the operator built a data center in those places, location.

Our next question comes from Timothy Zhao of Goldman Sachs.

Congrats on the solid results. I have 2 questions. First is about the pricing trend. Just wondering if you can share some color on how you think about the MSR trend into fourth quarter and next year, especially given that probably the company is entering to a peak renewal period for the contract that were signed maybe 5 to 7 years ago, then how should we think about the MSR trend into next year?

Second is about the overall market and the competitive landscape. I think right now, you have been emphasizing time-to-market quite a lot. If you remember, I think maybe 5 years ago when there was a wave about the cloud data centers and 5G network, there was also a wave of increased data center supply in China. Just wondering if you think, from where we are right now, how do you think about the overall industry supply and demand dynamics?

The first part of your question about the downward price reset when our installed base contract come up for renewal. And this has been going on for a few years and will continue for a few years more. And the impact of that gets reflected in our MSR. And I was -- give some comment on future expectations.

Now I'd say that, over 2026, we expect the MSR to decrease by 3% to 4%. That's on average, comparing 1Q versus 1Q, 2Q versus 2Q and so on. And that is not only a function of the downward price reset, we also have elevated higher levels of move-in. And that also has a dilutive effect on MSR. So that 3% to 4% reflects the combination of those factors.

Yes. I think I add a little bit of my points. I think all the new build data center, the price is quite stable since 2 years ago. Nothing changed. I think this is very good. But in the meanwhile, I think the cost is more stable, right? So if you look at all the new-build asset return, it's very decent. So I think this is a way to look at the MSR, right? Because the new campus, new building is, in general, I think compared with like edge data center, the enterprise data center, even cloud data center, the price definitely go -- went down a lot. But if you look at the asset return since 2 years ago, it's very, very similar, very -- and this price is very, very stable. Return is also very stable. It's 100% fit the REITs to inject to the REIT.

Tim asked about the competitive landscape.

Competitive landscape, I think the new competition, I think, if you try to get your customer trust and reliable, you should show your financial capability. Now our customers more care about the financial capability, not just the capability you can build. Everybody can build easily, right? So I think if you try to commit a customer 500-megawatt or 1-gigawatt campus in the future, I think the financial -- our customers definitely will consider about do you have the capability to access the capital market, what's the cash position you have right now? So this is very -- this is the new competitive advantage.

In terms of this, I think we are more -- much more way ahead than any competitor else, right? So I think this is not just a land/power competition. It's also the capability to access capital market. So in terms of this, if I look around, I think not that much company, both has the land capability -- power the land capability and well position and let's say, financing capability.

Thank you for the questions. Due to the time limits of today's call, I would like to now turn the call back over to the company for any closing remarks.

Thank you once again for joining us today and see you next time. Bye.

This concludes today's conference call. You may now disconnect your lines. Thank you.