Nebius Aktie

Okay. Good morning everybody we got a packed house Welcome to day 2 of the conference. Really pleased to have Nebius here. Speaking with us, we got Arkady Volozh, who's the Founder and CEO. I know there's been a lot of announcements and exciting news in your space. But Arkady, maybe for folks that are newer to the story, give us an overview, I think, of your background, which is also really interesting, but how you would position Nebius in kind of the AI landscape.

Nebius is a young start-up, a little bit more than 2 years old not so young by experience. It was formed with several hundred of very well educated and experienced engineers with a very small one small data center, 10 megawatts and some pocket cash of $2.5 billion. That was our initial state.

That's how we separated from our previous company. And we started this business in summer 2004 (sic) [ 2024 ], understanding that with our experience of building big infrastructure projects, being a hyperscaler in the past, there is a new opening on the market. AI will need a lot of infrastructure to be built, and we know how to do it. And there's not too many other people, teams who know how to do it. And there is a business to do.

And that's how we started [ $40 billion ] raised later, 5 gigawatt contracted later. We are now here a company with real business, real revenues, engineers, products, sales, marketing teams. What we actually are building is a small -- yet another hyperscaler. So we think what we are doing, there's not too many other teams can do it. There is -- if you look around on the market, who builds all this infrastructure from the full stack from the ground, from the plot to data center to racks to software, cloud layer, application layers and so on. There's not too many companies in the world who do this full stack. We do everything just like -- again, we're not as big, but just like AWS, Azure, GCP today.

Got it. And given there's a number of different players out there, hyperscalers, kind of more GPU infrastructure, neoclouds that primarily focused on providing capacity on the GPU side. Maybe let's dive a little bit deeper into the differentiation that you have that maybe puts you more in that hyperscaler camp versus kind of the bare metal GPU providers.

Yes. The infrastructure layer of AI is actually -- it's many layers. There's different layers. There's people who build -- develop plots. They take plots, build electricity, build shell and then lease these data centers to somebody else with profits. Then there's people who lease these data centers, bring active equipment on the site, GPUs, racks, all these big systems, network connect altogether. And then they sell it to the next layer on it's called bare metal. It's a basic infrastructure, almost not virtualized, basic virtualization.

And then people, bigger companies buy it, usually the same hyperscalers and big labs buy this infrastructure and then run their product on this infrastructure and resell it to the next layer, which is application developers, corporate users, people who build applications to build AI applications, which then on the next level, generate ultimate value, AI brings value to this world in different areas, generates new coding, security, pharma, whatever retail, AI generates a lot of value.

And actually -- and the food chain is the opposite. AI generates a lot of value in real life. It goes down to people who provide tools -- they buy infrastructure capacity and capacity buys the land and electricity, it goes a waterfall down. But it starts there generating value. If AI stops generating value, all this is dead. So far, it generates more value and there's -- nobody knows when it stops.

You're on the road and phone, I'm sure, talking with customers every day. Give us a sense of kind of the level of the demand environment right now, maybe compare and contrast that versus what you've seen in prior quarters? Like how would you describe it 'to this packed room?

So supply/demand is a well-known problem now, and it's a good problem to have because demand exceeds supply significantly today. That's because AI found a lot of niches to grow and to bring new value to generate value. And these niches are growing exponentially, whereas infrastructure supply, it is grounded in the real world, and it cannot multiply this fast. You cannot make physical things 10x more a year of physical things. It's not software, which you can multiply very quickly. So supply is scarce. Demand is growing and the situation, at least for now is getting worse or better. More and more severe, there is more and more demand. And unfortunately, with all this growth in infrastructure, it's not enough to cope with growing demand. So the difference is increasing.

And it all reflects in prices in the whole this waterfall of pricing, prices going up, go up. And we can see it -- you can see it all in expert management systems. Prices are going up in our business. We just reported last quarter that we started selling like everybody else, bare metal prices were at the level of $10 million, maybe $15 million per megawatt. We realized that demand is much higher. We tried manually to adjust this pricing. And until we found the next level of $20 million, $25 million per megawatt for consumer level, retail level and then on even shorter contracts, the demand is even more -- even higher and there people who need capacity now, not ready to wait until it's available in 2027, '28, and they're ready to pay much higher, like $40 million, $50 million per megawatt for this shorter-term contracts.

And this is what we discovered, not only we as an industry, we as a company as well, discovered this year. We discovered that kind of manually, and we understood that on this disproportionate market when demand is high and supply is not enough, the best way to -- for price discovery to balance it through the price, you need some automated procedure to always balance growing demand versus limited supply. And we were among the first, I think, who went to develop the system of auctions. We first experimented.

Again, it's why we were able to do it because in the past, we were the previous 20 years, we worked in consumer markets of Internet advertising or transportation services and all those industries, consumer industries, they're all balanced by some kind of auctioning systems, both advertising and search pricing in Uber taxi business and many, many areas, they balance it's huge industries and the balance in those industries are all found in some kind of auctioning systems.

They are not just simple auctions. It's very complicated systems, started simple and then eventually became very much complicated with a lot of people serving this area of price discovery. It's a separate business actually. And we think that probably something like this will start happening here in our industry, which is B2B now. But still price discovery should be determined by some kind of this auctioning system, probably like with commodities or financial markets. Every day, somebody trades stock. It's also kind of an auction system.

Yes. I thought the auction comments last quarter were really, really interesting and just sort of kind of revolutionary in the space. But how do you think -- I mean, you described this dynamic, a good problem to have, demand way more than supply. And you're going to be bringing on a lot of capacity. You still have a lot of capacity to sell. You talked about 5 gigawatts by 2030. How do you think about what makes most sense for Nebius in terms of the types of contracts you want to sign? Like for the next gigawatt of capacity, how much do you want to be auctioned, maybe short-term capacity through token factory versus multiyear committed deals? What makes most sense?

It's a question of optimization and derisking. Of course, it's great to put everything on short-term auction, try to sell everything at [ $50, $60 ] per megawatt. But then if something happens, then it happens, you have nothing. We divide our customers in 3 big buckets. The main -- our main business is in the middle bucket where we have, let's say, midrange customers, people who buy thousands and tens of thousands of GPUs -- for 1, 2, 3 years. This is our sweet spot, and this is our main market. We have dozens and hundreds of companies who are there. And this is our target market. But to serve it, you need to have capacity to sell. So we had to start with a different sector. We started -- we became famous for our contracts with Meta and Microsoft, which is great contracts, great customers, profitable and so on. But we basically had to do it. It's not our core business.

We cannot sell -- they don't need our software. They need just bare metal like from everybody else. And there's many other companies who provide bare metal services or new cloud. So we are one of many who serve this sector of the market. But -- we had to do it because we use those big contracts to help finance our main part and to build our independent part, our cloud part. Now we have it built -- started building it. It's just close to 1 gigawatt of capacity this year. And we want to build something like 1 gigawatt a year and then probably eventually more, gigawatt 1.5, 2 gigawatt a year.

To give you a scale, the industry today builds, I don't know, 15, maybe it was 5 gigawatts a couple of years ago, it's now 10, probably this year, it will be 15, 20 gigawatts a year built in '27, '28. So we are probably, I don't know, 5%, 10% market share in the world. We are a very small company. But this 5%, 10%, again, small in market share, but the industry is huge. And GPU cloud, unlike previous classical cloud is probably 10x larger. It's a $1 trillion revenue business. So to be 10% of that is not bad for a start-up.

Yes. Got it. Okay. And as we think about the execution of bringing gigawatts online, potentially multiple gigawatts online per year down the road, there's been a lot of investor concerns around local oppositions, state mandates. What gives you confidence in the ability to navigate some of those complexities as you bring that type of capacity online?

To build on the ground, to interact with a physical world is very challenging. And this is good, actually, this is good for us. It means there's not too many people can do it. So it's -- we are kind of protected. There is not too many people who can build this kind of complex infrastructure in this scale. We can do it. We definitely face the same problems, the same challenges as the whole industry. This year in U.S. specifically, this whole wave of people concerned with the consequences of AI and data center build. It's a genuine concern. People are generally frightened. They're afraid of AI, they're friend of the huge construction sites in their neighborhood -- understandable. We experienced it just like everybody else in the industry. How we deal with it? First of all, we diversify.

We build not only in the U.S., we built a lot in Europe. We -- I think we are the largest building in Europe. We built in the Middle East and we build in India, in Japan now through partnerships. It's not just one market. And when we build our own sites, actually, we do it from the beginning. We just recently realized that other people didn't do it here in the U.S. All the big constructions of data centers in the U.S., we are under NDAs, nobody guess who is building. We always have come to the site openly like a developer company, I would say, build a booth where we show the project, we invite community people to discuss what we're building.

We have our educational programs for the communities. We employ local unions to build it. We try to be open and friendly. But again, this is real life, real people, real concern. And specifically this year, there is a lot of delays in the industry. Eventually, I think again, it's not -- I don't believe that the situation will go so dramatically that the U.S. will stop building data centers. If it happens, they will be built elsewhere, maybe, but I don't believe it will happen. Somehow people will understand what it is. Actually, it's a huge -- I think data center is a very good industrial development for any community. First of all, it's much cleaner than anything else. Like in France, for example, we built -- we took a brownfield site, an old factory, which used to produce tires, a Bridgestone factory, which produce tires.

And now they have a clean, quiet data center, clean -- I mean CO2 clean and no noise and it's much bigger than having a factory in your background. So again, there's a lot of concerns, but I think maybe the industry should do better work explaining it. There will be much more explanations. -- but eventually, people, I think, will understand that it's good to have data centers, which generate a lot of revenues for the communities.

Yes. I wanted to shift to some of the headlines this week. There was a Palantir partnership that you announced. And what stood out to me was in the press release, both you and CEO, Alex Karp, were quoted talking about the partnership, which certainly implies a relationship from the top of the house down. So maybe talk to us about the evolution of that relationship? And specifically, what does this open up for you and potentially for them?

Where the synergy goes, we build infrastructure, which we eventually need to build up to the customer level. The customers ultimately will be enterprises. That's where AI will be creating all this value. To get to the enterprises, you need channels. For us, it's just -- first of all, it's a great channel. They have -- the majority of their business is commercial enterprises, big commercial enterprises. For us, it's the way to get there.

For them, they are proponents, their customers big corporations, they are concerned that when they use AI, they need to feed their data back to other companies. And they lose -- they formulate losing their sovereignty. They kind of give up all of their secrets data to somebody else, which then they incorporate in the universal models, which then later used by competitors to these enterprises. So they would rather see a model which they can control and the infrastructure they know and they can look through to the ground.

And Palantir actually have choosen us because we provide this full stack infrastructure, which they have -- they can have full control of -- they know what's going on up to the data center, racks, software, everything. On top of it, they put their software. Yes, we -- our level ends on the providing open models to them. Our Token factory provides all kinds of token models, open models, Chinese models, Nemotron and others. They take the models, they have instrumentation. Palantir has instruments for the enterprise, which allows them to take their data, take an open model, feed the data to the model. The model generates outcome.

They use it as their new data, corporate data, which they generate. They feed it back to the model. And after several cycles, there are several cases, even public cases showing that if you have 2 options, you can take a very high level best on the market proprietary model, Anthropic or OpenAI, something and actually give up your data to them and use them, pay a lot or you can take an open model, train it with your data multiple times.

And after several cycles, you get the open model, which is generally weaker than the universal model. But after several cycles of training with your own data in your specific domain, it becomes higher quality than the universal model. And that's what happened with Shopify. There was a famous tweet a week ago when they reached -- after several weeks, they reached quality -- they started with open model Qwen, I think. And after several cycles, they reached the level of quality in their specific domain higher than GPT-5.6.

So this cycle, open model and your own data in the cycle work helps to generate intelligence, which is smarter -- narrow domain smarter than the Super Duper universal models. That's the thesis of Palantir. We fully support it, NVIDIA supports it. Probably this is the way to go, one of the ways to go. Of course, universal models will not disappear. They will have their place. But this is the thing which Palantir wants to do. For this to do it, they need to partner with open model and full stack, which they can control and guarantee to their clients that it's a verified stack, the data will stay there and the quality is there. That's a partnership.

Okay. And just diving into the open weight models, open source models, which have been very topical in our world over the last few months. Certainly, I think the mix of open source models now versus where we were 6 months ago, it's probably a lot higher than people expected. But why does this matter for Nebius? What are the strategic reasons why if we see more open source, open weight models, how might this benefit for you? I mean, obviously, you kind of win if there's more compute and demand for AI broadly. But just curious how you would monetize that or what new opportunities that brings to you?

For us to have this full stack up to the cloud and up to the inference level and actually higher agentics, it gives us more job to do. Otherwise, we would -- just like any other Neo cloud, we would be just providing bare metal services at $10 million per megawatt to people who really then make a product with OpenAI and Anthropic or Microsoft, whatever, Google. Instead, we can build our own full stack and our own product and sell it much higher level with materially different margins. You can see the prices. It's like 2 or 4x higher, which makes very good margins, very good return on investment. We just recently announced that we pay back our CapEx in less than 2 years and 22 months on these prices. This is -- this product allows us to be this effective.

Otherwise, you go and just serve other people's interest, selling at very same margins, paying down the stack to other providers to buying racks, paying margins the leasing data centers, paying margins there and you work on a very -- probably you're a big company, big revenues, very small margins. We have a different business. We built a full stack. Our margins are very healthy. because the product allows us to get up to the users who pay these higher margins, not the intermediaries who buy cheap and then resell to the same market.

And presumably, those higher-margin services, things like Token factory and can you maybe just talk about the product portfolio in terms of the specific software inferencing services that you offer? And what should investors kind of track to gauge those progress? Because obviously, the underlying business is just growing so rapidly, right?

Yes. We develop a lot of products, but I'm ashamed to say that -- and we try to make them the smartest and the best. The reality is that you can unfortunately sell anything today, the quality of the product, if you have capacity, you can resell it at basic price or higher price, but you can sell everything which is built. But this situation is not forever. It will -- supply/demand will balance. Will it happen in a year, 18 months, 22 months, 24 months, it will balance some time. And then by then, people will start comparing products.

What -- and we want to be sure that we offer the full range of products, the full stack, full stack allows us to provide very -- it's low cost to us. We don't pay all these intermediaries in between. We control the quality of the stack. We optimize our product. It works faster, better when we control all of it. So hopefully, when the market stabilizes and the tides go, we will stay there as a company with real product with real customers, with real margins.

Yes. And obviously, this industry is changing by the day with new model releases and different supply and demand variables. But what's your best guess on how long this elevated pricing lasts? I mean you talked about conceptually 18 to 24 months. Is that kind of your working assumption? Or anything you would offer as a view on that?

It depends on how much value AI will be creating in real -- in real industries because again, today, new cloud sell it at [ $10, $15 ] to people who then sell it to companies developing applications at $20, $30, sometimes $50 short term, which then we sell it on the token level. Anthropic tokens recalculated per megawatt is something like [ $100 ] per megawatt. They are very high-level product. They have a lot of margin. They can afford buying capacity at not just [ $10, $20, $30 ], it's fine with them, still huge margins. But it's only because people who use the models who buy Anthropic, OpenAI tokens or open model tokens, they generate somewhere, I don't know, legal services, coding, whatever, math recently science in general.

They generate value per megawatt, which is much higher than this [ $100, maybe $200, maybe $500 ] per megawatt. While it continues, again, it depends on how AI works. I don't know who can predict it. Everybody expects that AI is serious. There will be more value creation, GDP will grow fast for the whole world through this. And it means that the prices may not just go down. Maybe they will go up, who knows, which means that even more infrastructure will be built. Actually, still not enough infrastructure. Physical world is limited. That's why Elon goes to space.

No Nebius space announcements.

No. Well, when they launched something in space, like they launched, they reallocated some of their capacity for the market. How they did it, they offer bare metal to whom to Azure, okay, we could serve this capacity. If Meta decides to -- they're not still there. There were discussions whether they need to provide their capacity to the market. If they decide to do it, we were the first to run after them to take this capacity. Actually, we're building for Meta now.

Our second contract comes in Q1, Q2 next year. If Meta says that they change their mind and they want to offer this capacity to the market, we were the first to beg them to give it to us because we will take this bare metal and resell it on the model level with much higher value. And I don't know, maybe we will be able to offer them some backstops if they offer us this capacity. If this happens, actually, the industry as a whole needs more capacity. So whoever can build it, will be well off.

Yes. And just in the closing minute here, your goal is to be an AI hyperscaler. What are the top 2 or 3 things you're focused on that investors should be on the lookout to track over the next year so we can see that vision play out?

First of all, the first thing for everybody is to build more. And to build it's hard. It's a lot of things every day. It's not easy. So we are looking for the ways how to physically build and how to finance it. We announced 5 gigawatts. It's easy to announce -- we signed it 5 gigawatts. We have this capacity to build. But to launch 5 gigawatt, multiply it by [ $50 billion ] per megawatt, it's $250 billion of financing. We -- so far, we financed maybe 20% of that. We need another $50 billion and then another $50 billion. So physical capacity, the sites, the physical world dealing with everyday hassle, communities included, everything financing all this stuff is the second big area of concern and effort.

And then we need to guess what product will be used, what these new customers will be needing -- will be using in not today, but next summer and a year from now. And finally, very -- actually, it's a new area to build the system of balancing demand and supply. It's like when we were a search engine, we had several thousand engineers on search and probably 1/4, 1/3 of this on this auction system for contextual advertising. It's a huge system around the main product. We need to build the system here.

Yes. Great. Well, I think that's an awesome place to leave it there. Arkady, thank you so much for joining us for a great discussion. Thanks for the awesome attendance in this room.

Appreciate it.

Fantastic. Well, hi, everyone. I'm Alex Duval. I head up the Europe Tech, Hardware and Semis research team based in London at Goldman Sachs. I'm delighted to be joined by Arkady Volozh, Chief Executive Officer at Nebius; and Marc Boroditsky, Chief Revenue Officer. Thank you very much, both of you, for being with us.

Thank you.

Great. So before we get started, I'd just like to state that this conversation is not intended for the media and is off the record.

Fantastic. So let's get straight into it and perhaps start Arkady with a bit of a discussion on AI compute demand. One of the key investor debates has been around how durable the current AI infrastructure cycle will ultimately prove to be. So can you talk about what you're seeing in the market today? And what does that tell us about the duration of underlying AI compute demand and to what degree has your view beyond '26 strengthened?

Well, I doubt I will say something new here. Everybody sees what's going on with the prices. It's obvious that demand today is much higher than the real world can build for this demand. And the prices always reflect this unbalance. What's going on is that AI is creating a lot of added value in some way in real life in the industries, be it coding or recently security or office work and many, many other areas which are to come. There is a lot of value created there, which goes down to people who provide tools to serve this demand, who provide models, which goes down from there to people like us who provide compute in the cloud to serve these models and then go down to people who build the infrastructure, which we also do. So demand is huge. It grows, I would say, times a year, if not tens of times a year. And the physical world is limited. It can't grow 10 times a year. That's why we have this unbalance. We publicly have said that -- a couple of months ago that if we could -- if we wanted, we could sell all the 2027 demand today, we have much more ask than we can physically serve.

We do not presell much. We are focusing on free capacity, which we will be selling later. We're building capacity, which will be sold later. But since we last publicly spoke, actually, the size of this demand went even further. Now we receive orders for first, second quarter of 2028. People are demanding tens of thousands of Vera Rubin GPUs now. So we see demand today as unlimited. Nothing is unlimited in this world. But for now is the demand in all the visible perspective is much higher than anybody can serve.

For how long it will go, we were talking about 18 months. Now I think we can easily talk about 24 months, maybe more. But today is the situation that AI is working. AI creates real value in real life. It needs to be served. We just -- as an industry, we can't build as fast. That's why this pricing growth.

Fantastic. And it's amazing how fast the industry is evolving, and it's also fantastic to see how Nebius is evolving. And I think you announced today a partnership with Palantir where Nebius was named as a sovereign partner. Could you explain what that sort of sovereign partnership entails? And which customers you'll be serving through that. And particularly, it'd be great to get a sense of why Palantir chose Nebius.

Well, what Palantir calls sovereign is sovereign for the companies. The companies they use AI, by using they produce a lot of data. This data go back to the models, improve the models, produce even more data and this loop continues. And the companies have actually today a choice. They can go to commercial models and start feeding their data back to them and improve their models or they can actually use open source, open-weight model on their own -- under their own control, feed data to these models. Keep improving these models for themselves, not sharing that data with the whole world. And actually, the practice shows that if you do this with generally open-weight models are lower quality than commercial models. But commercial models are very universal, very expensive. Open-weight models could be trained with this data in repetitive loops. So that in this specific narrow domain of this enterprise could be trained better to a higher level of intelligence than general models. And there's a lot of cases in our practice, in which it works.

Just recently, one of our clients, Shopify, there was famous tweet of CEO of Shopify, which show that they used open-weight model, trained it with their own data repetitively, and they achieved the quality which is higher than they had with GPT 5.6. Open model, your own data, repeated many times in the loop, better quality in your domain. That's the -- what enterprises should be doing. And by this, they don't share their data. They keep their data. They don't train the whole world, potentially their competitors. And this is what Palantir actually and NVIDIA as well and us see as the vision as the main mechanism, how enterprises should be using their AI. Palantir -- to implement this, Palantir has instruments to create this loop, open models, data back and forth, but they need infrastructure on which it could be run. And they used to use hyperscalers for that, sending data there.

Their clients don't like it because, again, their data is shared with everyone. And they needed a controlled stack which could run open-weight models, which is controlled down -- through the models to cloud to racks and the data centers. Palantir can control the whole stack. We provide this and the synergy is that we provide this whole stack up to our token factory, where we run a variety of different open models. They take it, add their tools on top of it, their customer base for us, it's an excellent channel, enterprise -- channel to enterprises. And together, hopefully, it will be nice tool to serve the market.

Fascinating. And perhaps my question to follow up on that with Marc. Palantir's commercial installed base is obviously quite different from the sort of AI native customer base that you've garnered hitherto. So perhaps you could talk a bit about how this fits into the enterprise customer base that you've been building?

Well, as Arkady just shared, there's a very powerful value proposition that we share with Palantir with regards to how we believe enterprise adoption is likely to take place. So working together, combining the full stack capabilities we have and the tooling that they have. And more importantly, the credibility and momentum that Palantir has in the enterprise. We're confident that we're going to be able to jointly develop opportunity that goes well beyond the existing market adoption in the enterprise and actually will accelerate our overall enterprise business and contribute to the customer diversification that we have at Nebius.

Fantastic. And I guess another one for you, Marc. Obviously, NVIDIA talked about 70% growth in fiscal year '28, and they talked about that being supported by demand across Blackwell and Rubin. I think it'd be really interesting to get your sense about what you're seeing on the customer demand side, both the newer generations of chips, but also the older chips?

Well, like what Arkady said at the with your very first question about the overall demand environment, that holds. That demand continues to outstrip supply. It's actually growing faster than supply. So that might lead you to believe that older chips are being used because people have no choice. There's a little bit of that. But the dominant experience that we're having across older chips is that the workloads that most of our customers have on those chips are actually properly aligned to those chips. They're simpler workloads, they're things like RAG infrastructure or text prediction or image creation and the applications that those customers have built or the services that they're running, they understand the TCO.

They have confidence in the hardware that they're using. It's reliable, it's vetted. They're satisfied with the kinds of reliability and performance that they're getting. And as a result, those businesses they grow, they're growing with us to the extent that we can supply them the older chips. Likewise, we actually have a list of new customers that are looking for any of the older generation chips that come available. So very robust demand for the older ranges.

On top of that, obviously, requirements continue to expand and with them, the interest in next-generation chips are expanding. We're seeing that across the Blackwells. We have a very robust level of interest and adoption that continues to be at a multiple of customers to chips that we have available that actually projects out as well to Vera Rubin interest. So -- while I can't confirm or deny the growth expectation that was shared by Jensen, I can say that it's extraordinarily robust in terms of the demand we're seeing across the full range of chips.

Excellent. And if we perhaps pivot to talk a bit about the landscape in terms of hyperscalers and neo-clouds. One thing that was interesting in the last few months, we saw Meta talking about exploring ways to monetize AI capacity and talking about a cloud offering. Arkady, I would be really interested in your perspective on what that means for Nebius and how you think about potential for large AI customers potentially to become competitors. So how do you think about Nebius positioning? What do we learn from this Meta announcement and how should we think about the market developing as more supply potentially comes online?

You said cloud offering. What's...

In inverted commas.

What SpaceX and Meta -- SpaceX started doing, Meta didn't start doing yet. They might -- they offer -- they see the demand. They see that whatever they build for themselves, for their ecosystems, if they do not use it right now, there's such a huge demand for this capacity they can offer it to the market. When they offer it to the market they actually offer it on bare metal basic terms. It sits on the cloud. That's why they sell it. We heard that SpaceX was selling to Anthropic, to Google, recently to Azure. People who actually take this basic level bare metal, and put the cloud and services on it and then resell it to end users. This is actually what we are doing. So potentially, we could be helping them to have -- to sell their unused capacity as well. Because we would definitely take whatever Meta does not want to use because we're [constrained] on capacity. There's not enough capacity.

If they could put more capacity on the market, it's good for the industry is good for us. We could take it and resell -- repackage and resell it to the end users. And if we talk about total volumes, just let's put it in context, the total amount of capacity came to the market a year ago was in the range of, I don't know, 5 gigawatts and 10 gigawatts, it will be probably 15 gigawatts, 20 gigawatts next year in '28.

What this SpaceX or Meta can add, it's several gigawatts on top of it. SpaceX talks about 6, 8 gigawatt additional. It doesn't change the market radically, doesn't grow the market -- help the market grow 10x. It's another 20% here, it's a 30% there. It's good for the market to have more capacity. And it's -- if we put it in scale, it's not so much. It doesn't change, doesn't kill the market, doesn't change the market dramatically. It's just good to have. So first of all, it's useful capacity, and we could actually help utilize it better through our cloud offering.

Thanks so much for putting in context, Arkady. Maybe one for you, Marc. Since we're talking about the market structure and how that works, I believe you've talked about 3 different buckets of customer types and the different kind of pricing you could garner there. It'd be great if you could just recap that a bit for the audience. Why each of those buckets exist? And how do you think about calibrating between those different opportunities?

So just to be clear, the 3 different types that Alex is alluding to include short term. These are agreements that are typically 3 to 6 months in duration. There are opportunities where we're potentially able to secure premium pricing around specific types of workloads that I'll share a little more detail about in a moment. There's what we call medium-term agreements. This is the 1- to 3-year agreements and actually represent, in our point of view, the lion's share of opportunity in the market and some of the most important and diverse set of customer opportunities.

And then thirdly, there's the longer-term agreements, which are 5 years and longer in duration. These tend to be the agreements that the industry, and we've done historically with like the hyperscalers. When we've done them, it's been with an explicit intent that we're looking for the capitalization benefit in terms of being able to fund the expansion of our core business, which takes us back to our primary objective, which is the medium-term agreements.

Medium-term agreements and the kinds of customers that we're talking about here run the range from the AI native that we have a right to win with. This is the heritage of the company. We are AI natives ourselves, and we've built an experience and delivered the type of technologies that these customers are most likely to adopt from. These are like Higgsfield in media and entertainment or HeyGen media production as well. These are companies that have built their apps on us and they're scaling with us.

There's also the larger scale AI natives that are already producing revenue. They're into scaling out. They're looking for longer-term visibility around the capacity that they need, they're signing longer agreements and are looking for greater cluster sizes. Also in the same range of contracts of 1 to 3 years, we're seeing digital natives like what Arkady mentioned earlier, like Shopify. Companies that are, again, looking for a reasonable duration of commitment to be able to fuel their requirements.

We're also seeing enterprises adopt in this range. It's just the beginning of enterprise adoption, but it's a very important dynamic and important part of our market. And we are actually prioritizing a portion of our execution like we talked about earlier around the Palantir agreement to make sure that we're able to start building towards the enterprise because we believe, ultimately, enterprise will represent the majority of the opportunity in the market.

Fantastic. And maybe one for you, Arkady, on pricing. I think it was fascinating to see at the last results, you talked about this capacity auction. So it would be great to get a sense of kind of what you learned from that. What does it mean for how you're going to price and sell capacity for '27? And how should we think about the durability of the pricing in future years as more capacity presumably will come online?

Well, when you have this imbalance between demand and supply, the best price discovery mechanism is something like an auction in one way or another. Again, we try to do it manually first to raise the prices. It didn't help. The demand is still there. What is the best mechanism to discover what the real price balance should be? We ran our first experimental auction for the Blackwell processors a month ago, several weeks ago. We achieved results where we got 15%, 20% higher price than we ever sold before -- after all our price raises.

We liked the experience. But more importantly, our clients liked the experience, the guys who won this auction, they tweeted that they're happy that they were able to get guaranteed capacity through this mechanism, because it's the only way when there is deficit of resources and you need it. what is the way to get it guaranteed in price competition and win? And they won the auction, and they were happy about it, and we are happy about it. And probably, it's the right way to find this right balance.

We will do more of this. We told that we will put more capacity on the auction. We now will start experimenting with our longer tail clients, self-serve clients. We have hundreds of thousands of clients, which are self-served. Small clients with preemptive mechanisms. And we will go introduce something like the auction there. And probably, this is the right mechanism to actually work with more regular and bigger clients, at least we should use some mechanism like the auction to discover the price. And then based on this full price discovery, we should offer them discounts or upsell from those prices, depending on different circumstances to these clients. But auctioning is the best way to find the right balance in this supply-demand situation, which we are now in.

You touched on competition. So maybe one for Marc, just more from a Nebius perspective. Obviously, at results, you talked about those 4 landmark wins. So it'd be great if you could just remind us a bit about what you won, what enabled you to win those? And then how we should think about the pipeline from here? Is there more to come?

Certainly, certainly. Actually, I realized when I answered the last question, I forgot 1 category. So I'm going to rewind just for a minute. I didn't mention what's going on with the short-term agreements, which I prefaced I was going to come back to later, we are setting aside a small amount of capacity for short-term scale training runs. This is a very unique and finite part of the market, and we're going to continue to make that available.

These tend to go out at a multiple of the ARR per megawatt that we see with our core or medium-term agreements. These 4 wins that we talked about last quarter fit in that core segment of our contract base. These were averaging $1 billion apiece. These were significant wins by existing -- pardon me, by new customers to Nebius that were looking for scale capacity. So they obviously had incumbent suppliers, was highly competitive. Those wins were ultimately won in the POC. It was not about price. It was actually about the reliability and performance that we delivered in the experience.

As a matter of fact, one of the customers actually even called out the fact that the POC, according to their technical team, was the best technical experience they have ever had with any supplier in the industry. So garnering significant confidence in the technology that we're delivering, but more importantly, also the humans that were part of that delivery. The expertise that we're able to provide them, gave them confidence about the kind of experience that they're ultimately going to have.

These wins have led to, obviously, the scale relationships that we have. But more importantly, we're partnering with these customers. They all have additional requirements. We're already talking to them about the next cluster that they have. In some cases, they're looking for additional GB300s. In all cases, they're all talking to us now about their Vera Rubin requirements. So already set up for additional expansion with them. We also have other similar customers in our pipeline that we're pursuing with similar types of opportunities that could ultimately contribute in the same way that these 4 wins did.

Fantastic. Staying with you, Marc, you mentioned there about new types of customers. Clearly, over the last 12 months, 24 months, you've had this period of diversification. So it's really not just about hyperscale at all. Obviously, there's a place for that, but you kind of calibrate all these opportunities. Can you talk a little bit about how the mix has been evolving and how you think that customer mix will evolve as we go forward?

Certainly. Today, as I mentioned earlier, we have the right to win with AI natives. This is a very important part of our business. What we are experiencing is that we're starting to move up to the larger and larger scale AI natives like the 4 deals that I mentioned a moment ago. That experience that we're gaining with those types of customers is actually building a profile and reputation with digital natives. So like the Shopify deal that I mentioned earlier, we're getting more and more access to critically leading digital native customers where they're embedding AI capabilities in their offering, utilizing AI in order to be able to deliver a better version of their service.

It's a very exciting portion of our business because those digital natives are not only scale, but they're also oftentimes supplying the enterprise, which also now is leading to our opportunity to supply the enterprise. There's been a significant uptick in our pipeline with customers that are platform providers to the enterprise. Some of the most important platform players are looking at us as the first non-hyperscaler supplier, and they're looking for 2 reasons.

They're looking for the obvious ability to gain access to capacity when they need it. They're also looking for a partner that can innovate with them to be able to support their requirements across their full needs, not just training, but also being able to support them with inferencing and post-training and recognizing what we're building allows them to have the kind of capabilities that, as a matter of fact, we described a little while ago for Palantir. So this is opening up a market opportunity for us and then projecting out to the enterprise where we're seeing significant traction in key verticals.

We have a very focused set of vertical initiatives around physical AI, around healthcare, life sciences, around media and entertainment, around retail and financial services. And we're just starting to get traction there, and we're excited about the progress that we're making.

Aside from the different use cases and the different types of customers. One thing I think Nebius has talked about, Marc, is how, over time, we're going to have more and more of these value-added services. And I think you've talked about how that can help with margins. So can you sort of help us think about how large that can be over time and sort of the evolution in that sense.

So it's a really important part of our business, Alex. I mean what Arkady mentioned a moment ago in terms of the full stack, a lot of folks look at the industry and they sort of stop at the hardware. They think of the market that we're in is all about just compute. But the reality is when you talk to customers, and especially as you talk to customers that are more and more, let's call it, large scale their desire is actually for a much more full stack experience and even more abstracted services. So today, when we go sell a customer that might start with us on compute, looking for model training capabilities, we are oftentimes also able to position our inferencing.

The combination gives them significant efficiencies and advantages because they are using the same platform where we built the solutions together and they can use the compute in the future between the different capabilities. Likewise, we've added to our platform, the beginnings of agentic capabilities through our acquisition of Tavily, which is an agentic search solution. The conversations that we're having with AI builders, agentic search is a cornerstone technology necessary for them to build the agentic capabilities they have.

We're going to be adding to our agent capabilities. Right now, we're delivering blueprint, but we'll be building out more and more solutions to support agentic requirements. As we project out into the future, the vision that Arkady has us becoming a hyperscaler. We have the unique front row seat of being able to watch what's happening with the customers that are building our platform in terms of things that they're doing that could potentially be turned into products and the things that aren't being built that they're turning to us to help them to build in order to bring them to market more efficiently. So it's a unique situation where we're confident that we're going to be able to project into adjacencies that have more value creation and obviously opportunity for us to do value capture and margin -- and add margin to our results.

Talking of value capture perhaps another question for Arkady. It was really fascinating to see the announcement about your new asset-light model, where I understand you can actually partner with other infrastructure players. Can you talk us through how that works? And how does that sort of expand the opportunity and what you see as the benefits when we think about growth and profitability?

With this demand, the main impediment in supply is how much physically we can build, how fast and then how much we can finance all this build. Because one thing is to build the data center. Another is to fill it with racks full of GPUs and tens and hundreds of billions of dollars in our scale. So we develop our own network capacity of data centers, which we build. And this is our natural pace of growth. And recently, we thought that probably we could grow even faster if we partnered with people, there's a lot of -- a whole new class of people who came from electricity companies, from data center companies who are trying to go up the stack, understanding that there is margins up the stack. And they actually need help.

They -- what they have, they have physical capacity, plots, electricity, and many of them have access to cheap financing. And we realize that if we partner with them, we provide them with our knowledge. They don't know how -- what to do with it. We can give them -- bring them the knowledge how to build the data center. We can build them -- we can give them our racks. We can give our software stack. And most of all, we can give them all of our go-to-market organization, we have so many customers for that. So it could be potentially a very good partnership with companies who have access to capital and who have their own physical plots.

We announced this line of business, this partnership program a couple of months ago, I think. We've got a long list of companies wishing to deploy to try to do something here. It's -- all of this -- each line in that huge spreadsheet is a separate big project. We are talking about capacity coming online in 2027 and later. We're working with several projects like this. And this will be sure -- this will be a non-linear way for us to grow.

One thing is we are building ourselves. And in addition to that, we'll probably partner with companies who have access to capital and physical resources to build it. Which is we will be helping them to build and use our client base.

Great. And maybe I think we've got time for one last question. I think you talked about the physical limitations. And obviously, power is really interesting as well as land. Just curious how you think about where you effectively deploy your incremental capacity. And within that, how you think about behind-the-meter opportunities versus perhaps other types of power.

We work with all kinds of plots and power we can get. Anything which moves we will utilize. It's never enough. Again, demand is so huge. We built -- sometimes we built on the plots with electricity, which is ready. Sometimes we need to get connected to the grid. And sometimes, we can just use our own generation. We buy generators. We buy -- recently we started partnering with a company called Bloom, which is hydrogen and co-generation, much more healthy. And so we have -- just like companies going up the stack from the ground, we have to go down the stack from software to racks to data centers and now sometimes to building our own generation, including our recent thing with Bloom, but not only.

So whatever we can build as fast as we can build, everything will be sold. So this is the limit for our growth. So that's why the more we build the more we partner with somebody to build the better for us.

Brilliant. Well, on that note, Arkady, Marc, thank you so much for a fascinating discussion. And thank you for joining us.

Thank you, Alex.

Thank you.

Welcome to Nebius Group's Q2 2026 Earnings Conference Call.

[Operator Instructions] I will now hand over to Gili Naftalovich, Head of Investor Relations, to start the call.

Hi, everyone, and welcome to Nebius' Second Quarter 2026 Earnings Conference Call. Joining us on the call today, we have our CEO, Arkady; our CFO, Dado, and the broader Nebius management team.

Before we get started, I'll quickly cover the safe harbor. Some of the statements that we make today regarding our business operations and financial performance may be considered forward-looking. Such statements are based on current expectations and assumptions that are subject to a number of risks and uncertainties. Actual results could differ materially. Please refer to our Form 20-F for a list of our risk factors. We undertake no obligation to update any forward-looking statements.

During this call, we will present both GAAP and non-GAAP financial measures. A reconciliation of non-GAAP to GAAP measures is included in today's earnings press release. All earnings-related documents are distributed and available to the public through our Investor Relations website, which can be found at nebius.com.

And now I'd like to turn the call over to Arkady.

Thanks, Gili, and welcome, everyone, to our call today. I would like to provide our investors with a better understanding of our business model.

We had an excellent quarter. The demand for what we are building continues to be enormous, and we have the right business model to capture it. We build capacity ahead of contracts and use our multi-tenant cloud and our software stack to support our customers, AI-natives, Agentic AI leaders, neolabs and some of the most sophisticated enterprises.

Our strategy is working. We choose when to sell, to whom we sell and on what terms, and how we finance everything. This flexibility allows us to meet the needs of independent AI builders and to support an open, diverse and competitive market. We address customer needs with 3 types of deals. Each has a different duration, pricing and role in our business.

First, for our core AI cloud business. We have midterm contracts of 1 to 3 years with the world's most ambitious AI companies. This quarter alone, we closed 4 landmark deals, with Reflection, Cohere as well as with a scaled U.S. new lab and a large U.S.-based quant trading firm. These deals were for an average of more than $1 billion each. They represent a yield of $20 million to $25 million per megawatt, with upfront payments that cover 50%, 60% of the associated CapEx. But most importantly, we could sell today our entire 2027 capacity on these terms if we wanted to. But we are not doing this. We see that we can achieve higher value by retaining some capacity to serve shorter-term and immediate client needs.

And here we come to the second type of deal, this is shorter-duration capacity, typically for up to 6 months for customers with an immediate time-bounded needs with high-value requirements. For this, they are ready to pay a significant premium. We're negotiating deals for $40 million to $50 million per megawatt range, and sometimes above, under this model. One of such deals has been signed just recently by the way.

Both deal terms I just described are all coming online later this year. So they will not have a material effect on our 2026 revenue guidance. However, there will be -- they will definitely affect our 2027 revenue and beyond.

The third deal type, as we noted many times in the past, is our long-term contracts with investment-grade customers. They serve an important purpose. They help us to finance our build-out faster and more efficiently. The secured debt facility we raised in July was on the back of one of these deals. And with $40 billion in contracted backlog, we will do more of this.

We are constantly building, innovating and developing our software stack, technological offerings and services. But today, I would like to focus on different types of innovations from this quarter. We launched our first capacity auction. It was very successful and cleared at the highest price we have seen for the Blackwell generation of chips, 15% above the highest price we ever charged before. This gives us a strong signal on the value of this capacity in the market in real time. This was an innovation of sorts on the go-to-market side.

On the capacity side, we're also innovating in how we build. For example, our asset-light partnership model, which we introduced this quarter, provides an additional way for us to scale. This model addresses the 2 constraints of our industry, which is capital and capacity. Partners finance, build and operate the facilities, whereas Nebius brings the full-stack platform and the demand. We provide value-added services that sit on top of our partners' infrastructure, which delivers us with high-margin revenue and requires minimum balance sheet capital. This model has the potential to unlock new capacity for us in 2027 and beyond.

We continue to build our future capacity pipeline through our own and colocated sites, and today raise our year-end contracted power target to 5 gigawatts. Our future capacity pipeline effectively makes Nebius one of just a few companies in the world able to build more than 1 gigawatt of new capacity a year. And we plan to do so in 2027.

So to close, everything we set out to do this quarter, we have done. And in most cases, we have done more. Landmark deals in our core market, on better terms than we expected. Capacity rising to meet demand and the platform meeting the needs of the industry. We are not just reporting a strong quarter behind us. It's much more than that. We see the demand, we see the supply. We see deal terms for 2027 and beyond. Given our future capacity plans, including for '27, we cannot be more excited about the future.

And on this optimistic note, let me hand it over to Dado.

Thank you, Arkady. We entered the second half of the year with strong momentum as we successfully executed our strategy in the first half. We again delivered triple-digit revenue and ARR growth in Q2, even before most of our 2026 capacity comes online in the second half of the year. This was coupled with a significant expansion in adjusted EBITDA margin while strengthening our funding position, which was supported by substantial customer prepayment terms and expanded sources of our capital. This start is strong and keeps us on track to achieve both our strategic and financial objectives for the year.

I will touch on the progress we achieved in Q2, share details on our financial results and conclude with our annual guidance, which we are reaffirming today. Please note that all comparisons are year-over-year unless noted otherwise.

Let's start with revenue and ARR. In Q2, we grew Group revenue by 454% to $582 million, up 46% from last quarter. Our Nebius AI business grew even faster, increasing 514% from last year to $575 million and generating 98% of group revenue. Annualized run rate revenue reached $3 billion at the end of June, up 598% and increased 56% (sic) [ 58% ] from $1.9 billion at the end of March. Revenue was driven by capacity added in Q1, higher utilization and high-margin revenue from our new asset-light business model, Token Factory and our recent acquisitions. Higher utilization was driven by improvements in our underlying infrastructure efficiency. And once again, we sold out of capacity because, as fast as we bring capacity online, we can sell it.

Turning to profitability. Group adjusted EBITDA was $236 million, compared to a loss of $21 million a year ago and $129.5 million last quarter. Group adjusted EBITDA margin was 41%, up from 32% in Q1. Our Nebius AI business generated adjusted EBITDA of $286 million (sic) [ $236 million ] at a margin of 50%. The delta between Group and the Nebius AI business margin reflects our investments in Avride and TripleTen, both are still early stage companies. And we expect the Nebius AI business to continue to represent the significant majority of group adjusted EBITDA.

The increase in profitability is a result of higher revenue and the early contribution of our asset-light model as well as Token Factory and our recent acquisitions. Even as we continue to invest in growth, we see a clear path to capturing more margin, not just this year, but in 2027 and well beyond. We have visibility into pricing and expect capacity coming online from our own data center to begin improving margins in the second half of next year.

Now turning to our balance sheet. Prepayments from our customers reached an all-time high in the quarter. Roughly 70% of the deals we closed in Q2 included an upfront prepayment. In total, customers' prepayments will bring in more than $9 billion of upfront funding this year, directly reducing the capital we need from debt and equity. Operating cash was $2.3 billion in the quarter, and we ended the period with $8 billion in cash and cash equivalents.

In Q2, we also tapped our at-the-market equity program. We issued 12.7 million Class A shares at a weighted average price of $224 per share, generating gross proceeds of approximately $2.8 billion. As of June 30, 12.3 million shares remain available under the program. We view the ATM as a flexible funding tool, and it remains an option in our arsenal rather than a commitment.

In July, we announced our first asset-backed debt facility for $775 million. This facility is secured against contracted cash flows and priced at a modest spread over benchmark rates. Its rate today is equivalent to a mid-single-digit percentage. And we have more than $40 billion of additional customer commitments at similar terms based on the strategic deals we have entered into with investment-grade companies.

We will continue to diversify our funding sources while maintaining a balanced mix of debt and equity. We are progressing on additional asset-backed financing and continue to evaluate corporate-level debt and additional financing instruments.

Turning next to CapEx. In Q2, capital expenditures were approximately $5.7 billion, driven primarily by purchases of GPUs, GPU-related hardware and our data center expansion. Our capital investments to date position us to achieve our capacity plans for the year.

Let me turn to outlook. The solid progress we have achieved in the first half of the year bolsters our confidence in our outlook for 2026. As such, we are reaffirming our full year 2026 guidance across all metrics. We continue to expect annualized run rate revenue of $7 billion to $9 billion, group revenue of between $3 billion and $3.4 billion, group adjusted EBITDA margin of approximately 40%, and capital expenditures of $20 billion to $25 billion. And we remain confident in our ability to accelerate capacity deployment in the second half of the year and expect capacity deployed late in Q2 to begin contributing to revenue in Q3.

As Arkady said, we are now building capacity for 2027 demand, supported by customer commitments already in place and providing near-term visibility into the revenue associated with this investment.

And finally, we are actively pursuing opportunities to generate incremental high-margin revenue from our new models, our auction, our short-term capacity deals as well as our asset-light business. In summary, our strong Q2 performance reflects our rigorous execution to scale the business profitably. We delivered solid revenue growth. We increased profitability. We opened a new capital-efficient path to capacity, and we expanded our funding alternatives.

As we look ahead, we will continue to scale rapidly to capture the opportunity in front of us while remaining balanced, disciplined and focused on delivering long-term value for our shareholders.

And with that, I will turn the call over to Gili for Q&A.

[Operator Instructions] I'll now hand back to the Nebius management team for the Q&A session.

Thank you, operator, and thank you, Dado. Our first question comes in from [ Ryan Lantz ] from Morgan Stanley. With the Vineland, New Jersey data center site hearing adjourning without a vote, can we just get an update on the site? How is capacity tracking and with no vote taking place yet, how could this impact plans to ramp this site?

Tom, can you help us answer that?

Yes. Sure. Happy to. I'll start by touching on the public hearing side of things. And I guess maybe just one quick sort of general comment. Obviously, there's a broader set of issues and debate around data centers in the U.S. These are things that we track very closely. And we've generally found that we have an effective approach to moving into new regions, which is based on coming in early, being very open, transparent about what we're doing, engaging the community, and actually answering the questions that come up. And we find that this is an effective approach, and the same approach that we've been applying in New Jersey around Vineland. So specifically around Vineland.

So overall, we're on track with our delivery, and I'll let Andrey speak a little bit more -- in a bit more detail about that from kind of the construction and build-out side. But with respect to the public hearing, so what's happening right now? DataOne is seeking final approval for the amendment to their originally approved site layout plan. And the amendment to the site layout plan was a result of the decision to switch from the project's power source to Bloom.

I'd say generally that the switch to Bloom, we think, significantly enhances the project, including from a community perspective. It's an on-site power solution delivering reliable power, quietly and ultra-low emissions.

I'd also say these types of hearings, these are all part of the normal process, and we very much build them into our schedule with these steps in mind. And so with respect to this hearing, on the layout plan. So we're confident that the layout complies with all the applicable local, state and federal laws and regulations. And we're optimistic that once the public has been heard, this will move quickly to approval. And we'll definitely keep you updated as we have more news.

But with that, maybe I can ask Andrey to add a couple of points just from the build-out side.

Yes. Sure, Tom. Hello, everyone. Well, we delivered all the tranches that we were required to deliver under the contract up to date. And we have all the reasons to believe that we will continue to deliver the remaining tranches as required by the contract.

And probably to add color, construction of the building itself finished earlier this summer, and engineering fit-out is progressing fairly well. And Bloom fuel cells deployment should be fast. And overall, the switch to Bloom has been a valuable and a good pivot for the project, we believe, and with no significant impact expected on the project time line.

Okay. Thank you. Our next question is -- we're getting a lot of questions actually on this from the portal. Marc, maybe you can help us talk through our landmark deal announcements. Can you characterize how those processes run and why these customers chose Nebius?

Certainly. Thank you for the question. And we are very excited about these wins. Closing $1 billion-dollar deals in our core AI cloud business is an important milestone in our go-to-market journey. It confirms our fundamental belief that we can build a diversified customer base at scale. All 4 were competitive wins following a similar pattern. Our scale, performance and reliability were the differentiators. Also, each customer sees us as a long-term partner that they can grow with.

In all cases, the customer already had an existing supplier, in some cases, hyperscaler, and they were looking to partner and expand with next-generation capacity and a platform. As a matter of fact, let me tell you about one of these wins. They were introduced to us by a key strategic partner late in Q1. The customer was looking for a large-scale, contiguous GB 300 cluster, also flexibility for future expansion as well as strong technical support and a long-term strategic partnership. When we closed with them in May, they cited our responsiveness, our transparency, the white glove support they received, and our ability to provide their current U.S. deployment needs as well as future sovereign expansion as key differentiators. We are now in discussions with them for more capacity and they're exploring our inference solution, Token Factory.

All of these wins were earned wins, not inbound walk-ins. We had multiple engagement cycles before closing. The customers validated our technology through hands-on POCs. One of them told us it was quite literally the best POC they have ever had. We are already in discussions with several of them on post-training and inference to support their revenue plans and growth, and with all of them on significant additional capacity and next-generation chips, including Vera Rubins. And yes, we have more deals just like them. Our pipeline generation stepped up again in Q2 and includes multiple opportunities over $1 billion, spanning AI-natives, neolabs and enterprises.

Thank you, Marc. The next question we have comes from Alex Duval, Goldman Sachs. Andrey, a large share of our deals we signed are tied to capacity arriving in late 2026 and through 2027. Given the sentiment around gigawatt-scale data center build-out, what gives you the confidence in our capacity ramp? And can you provide an update on our development time line? .

Thank you, Gili. Well, we made a very good progress this year with our contracted power, and we're exceeding our end of the year projections much earlier for 2026. And we will be raising -- or we are raising actually the contracted power to 5 gigawatts now by the end of 2026. Almost all of that power will come online over the next 3, maybe to 3.5 years out of the contracted power that number that we mentioned. And we are not stopping actually.

This has been achieved by different region expansion and the mix of grid power and behind-the-meter power. We're also working on secured to have the access to the hundreds of megawatts of the generation behind the meter power already as of today.

We're also excited about our partnership with Bloom, which allows us to unlock and expedite a lot of our sites. And also, I wanted to mention that the vast majority of our contracts is the cloud contracts, where we have the flexibility where within the region, we can deliver it, reducing the dependency on any one location. So over-provisioning the sites, over-provisioning the capacity is our highest priority for today, and I believe that will be the case for tomorrow as well. And general approach is we deploy as much capacity as we can and we try to build in advance as much as we can as well.

Thank you, Andrey. Arkady, we've had a number of questions come in on our strategy. The market is moving quickly. You mentioned prices are increasing and you plan to expand capacity by more than 1 gigawatt per year. You also announced a lot of new initiatives this quarter. Help us make sense of how these fit into our broader long-term strategy.

Yes, you're right. The market is evolving rapidly, actually even more rapidly than ever been expected. And the good thing about our business model and our platform as well is that they both allow us to evolve rapidly with the market. I will give you a couple of examples from just this quarter.

First, as we all see, prices are changing, changing fast, they're growing. And our model was built to build capacity in advance, but not to presell it in advance. And that is why we have this available capacity now that can allow us to allocate it to shorter-term and much high-margin contracts.

The second example would be, as I mentioned, we also launched our first auction this quarter. This is the auction for the midterm contracts. Why were we able to do so? Because again, we had free unallocated capacity, and we had our multi-tenant platform. And we're just realizing the benefits from it.

And we have similar flexibility on the capacity front. And again, it's a strategic advantage of the platform we have. The fact that our platform is universal and able to work on any third-party capacity gives us an opportunity to grow even faster. And this is what is the -- our asset-light model is all about.

So we see there are multiple ways to monetize the universality of our full-stack platform, on both software and hardware part of it. And the way we think about all of this is simple. This market is going to continue to move quickly and to grow quickly. And we will be sure that our business model and platform will stay flexible to allow us to capitalize, to use all this growth.

Thanks, Arkady. Our next question comes from Arsenije Matovic from Wolfe Research and was emphasized by many others from our portal. It relates to our new $775 million asset-backed financing. Debt markets have been volatile and all-in costs have generally moved higher. Are you still comfortable leaning on debt from here? Or should we expect greater use of equity through the ATM or perhaps a convertible? Dado?

Thank you, Arsenije. Well, our approach is not about making a call on the markets. Really, our approach is to match the right financing instrument with the right assets, while remaining disciplined about 3 things: our cost of capital, minimizing shareholder dilution and maintaining a strong and healthy balance sheet.

Our first sources of capital remain customer prepayments and operating cash flow. We expect more than $9 billion of upfront customer prepayments in 2026. Clearly, this directly reduces the amount of external financing we need as we continue to scale.

Beyond that, asset-backed financing is an important part of our strategy. The $775 million facility we completed in July, priced at SOFR plus 250 basis points was backed by deployed GPU's infrastructure and contracted cash flows from an investment-grade customer. It demonstrates that even in a more volatile market, there is a strong demand to finance these contracted cash flows on attractive terms. And look, with more than $40 billion of committed backlog, we believe this is a highly scalable and repeatable financing model. Obviously, we expect to continue accessing this market as we deploy additional capacity.

And also beyond asset-backed financing, we have significant flexibility. We currently have almost no corporate-level debt, so that represents an additional source of capital that we can evaluate as the business scales. Equity and equity-linked financing, such as the ATM or potentially convertibles, offer additional sources of funding, and we evaluate those alongside all of our other funding options as we source of capital. We are actively considering further equity-linked and asset-backed financing options and continue to evaluate corporate-level debt and other financing alternatives.

So overall, we remain very comfortable with our funding position. So we do have multiple sources of capital available to us, and we will continue to optimize across them with a focus on cost of capital, limiting dilution and maintaining a disciplined balance sheet.

Thank you, Dado. Our next question comes from Tyler Radke with Citi. And Marc, this one is for you. How do you think about the mix of allocating 2027 capacity between short-dated capacity versus multiyear deals? Is there a threshold of ACV per megawatt you are waiting for on the longer-dated deals to sign into those?

Thank you, Tyler. Before I dive in, let me just make sure and clarify the deals with a little bit more nuance, the type of deals that we're doing, with a little bit more nuance. Our long -- our medium-term or midterm contract durations are most -- for the most part, are AI cloud customer agreements, and that's our core of our business. And that's where we're lock-in in strong unit economics with some of the world's most ambitious AI companies.

And then separately, as you called out, we have the shorter-term opportunities and durations. And these are premium deals. And that includes, as an example, the auction that we described today, as well as the ability to capture sort of the shorter-term scale contracts that we mentioned.

In practice, what we strive to do is to optimize across customer type, price, payment structure, duration and deal size, rather than strictly looking at a single variable or a compiled metric. Our current emphasis is taking care of existing customers, followed by new logos and then terms, terms in order of priority, including price, then upfront prepayment, then duration. So a bit wider than purely ACV per megawatt.

We have also, at the same time, tactically shortened how far in advance we sell capacity. So in effect, selling closer to deployment, which, at the end of the day, has improved the pricing that we can derive while preserving our agility. We're deliberately allocating a portion of capacity for short-term and immediate needs because that is where we currently see the highest potential for combined realized value.

To summarize, as we deploy capacity looking towards the latter half of this year and into next year, we will allocate across both medium- and short-term deals using this with our customer and terms lens that I mentioned earlier.

Thank you, Marc. This next question comes from James Kisner of Water Tower. xAI has begun selling compute at premium prices. What does that say about market pricing for AI capacity? And are you seeing similar strength on new contracts and renewals? What does this mean for Nebius?

Probably I will take it. It's Arkady. Definitely, we don't see any change here from Nebius or our plans to change. We are playing on the -- basically on the same market as the 3 big hyperscaler clouds. The whole AI cloud market is growing massively from hundreds of billions of dollars per year to probably a trillion, some people say more. And there's no doubt that the main existing players will continue to grow with the market, right? But there's also, definitely -- there is an opening for the new entrants for the independent players like us.

And yes, we want to build a gigawatt capacity per year. Yes, there is not so many companies doing that. The numbers look great, big, but the market as a whole is growing by tens of gigawatts per year. It's like multiple of order larger. And the hyperscalers are those who build a lot of this. But even then, they definitely cannot build all of it. And somebody needs to help them to build. And that's not us. Somebody -- because somebody needs to build the rest of the market, and that will be us. This is our part. And this is where we see the biggest opportunity for us.

So to -- coming back to the question, new players coming to the market don't change the market for us. They actually just, I think, validate the market.

Thank you, Arkady. Our next question is from Rob Oliver from Baird. Is your strong positioning in open-weight driving increased inference among clients?

Roman, this one's for you.

Yes. Thank you, Gili for it, and thank you for the question. First of all, we believe that customers, businesses and even society benefit from competition and diversity. We were committed to an open AI ecosystem from the beginning of the company and providing open infrastructure without lock-in. So customers get flexibility, control over their data and models, freedom in how they deploy.

And at scale, the economics matter. Companies at the frontier of adoption of -- AI adoption are seeing the cost of token consumption and asking, can AI not only solve the task, but do so with economics that allow it to scale?

And another important question is how to extract value from specialized knowledge and data held within enterprises and by their people and convert it into better-performing AI systems. Everybody is saying that data is the moat, and maybe the only moat.

All of this is driving companies towards more specialized models. Open-source models are valuable, not simply because they are available, but because they can be tuned, trained and eventually optimized for a particular use case of a particular customer, a particular business.

What has changed is that the quality of open models is improving rapidly. At the same time, the industry is moving towards the flexibility and control that we built from the start. So Token Factory provides day 0 support for frontier open models. And the velocity is just crazy. If you look just in the last 4 weeks, we got Nemotron Ultra, GLM 5.2, Kimi K3, New DeepSeek Flash, Minimax 3. Today, new Nemotron Lightning was released. But availability alone is not enough. We are building the capabilities to serve these models without compromising quality, cost and performance.

Let's take GLM 5.2 as an example. Some people called the moment when this model was released the DeepSeek moment 2. Our implementation achieved 100% quality score and leading performance validated by independent benchmarking and Artificial Analysis. That positions us in a great spot to grow and drive the demand to open source models that we see at the market now.

Thank you, Roman. A follow-up from Rob. Can you provide any early color on the asset-light offering, early customer engagement and how the economics look for you versus your core?

Arkady, do you want to take this one?

Yes. First of all, we're growing globally, but we're still a startup, and we should be very -- to be very mindful about where we invest our own capital. We can't go into all markets instantly and simultaneously. That's why we are very open to working with partners, and our platform allows us to do so.

After we announced our asset-light model, we have had dozens of inquiries from potential partners who have significant capacity and enough capital but don't know how to build and how to sell it, with GPU increasingly becoming an investable asset, look at the recent news. We expect more and more companies who will want to come to the market, but will need our help to deliver this capacity to customers. By having the right tech platform and access to the market, we can help them.

So this is actually exactly what we can easily offer to them. And that's why this asset-light model. The model is still in very early stage. But we are very much encouraged by early signs and think that it has a great potential.

Our next question is from [ Ryan Lantz ] from Morgan Stanley. Dado, you said you could sell your entire 2027 capacity today. How should investors frame 2027 across capacity, pricing and revenue? And will you -- when will you formalize an outlook?

Thank you, Ryan. This is probably one of the more interesting questions for investors on this call. You should expect the deals we closed in this quarter with more than $20 million per megawatt deals and less than 2-year payback period to start coming online from late Q4 and onwards. This can serve as the baseline for pricing early next year as we could have sold out our planned capacity already today. However, we are choosing not to do so, reflecting our confidence with respect to future pricing dynamics.

But also capacity, obviously, is a major part of this equation. We are going to deploy significantly more capacity in 2027 than what we have already done in '26 and what we are going to do still in '26. The dynamics we expect across both capacity and pricing make us extremely excited about 2027, although we will provide a formal guidance later this year.

And don't forget that in addition, we expect our asset-light model, along with high-value services such as agentic and inference solutions, to contribute an increasing share of revenue while supporting even higher margins.

Thank you, Dado. Our next question is from Brett Knoblauch at Cantor. Do we still expect to end the year with 800 megawatts to 1 gigawatt?

I see -- Andrey, do you want to take that for us?

Yes, absolutely. Yes, we still -- well, we still expect to meet this guidance from 800 megawatts to 1 gigawatt of connected power this year. And -- but -- and also, there was a part about the Vineland, concluding in 2027. No, the Vineland is the part of 2026 capacity and connected power.

What I wanted to say is that connected power represents the data centers, and there are a few steps coming from the data centers to the revenues. You have to commission the data center, build the network, build the clusters, deploy the platform, then onboarding the customers, and then the revenue generation starts. So that takes a few months. And depending also on the generation change of the GPUs probably. And so there is a gap between the connected power and the revenues coming.

And in terms of our guidance from 800 megawatts to 1 gigawatt, yes, that's more -- that's true about the connected, but I would think that about being active in -- throughout the first half of 2027.

Thank you, Andrey. We also have a question from BNP Paribas from Stefan Slowinski, specifically around our $40 million per megawatt monetization in Q3. Is this reflective of initial pricing from selling Vera Rubin capacity?

Andrey, do you want to talk about the ramp of Vera Rubin capacity?

Yes, absolutely. So we have Vera Rubins in our labs for quite a while already. And we already are getting the results -- the expected results with the Vera Rubins. So we -- first of all, I wanted to mention that step between the Grace Blackwells and Vera Rubins is somewhat easier from the technical perspective than a step from the previous generation to the Grace Blackwells. And we expect to start deploying the Vera Rubins late this year or early next year and continue throughout the full next year.

Yes. And to the first part of that question, and we're getting a lot of questions from the portal on our new go-to-market motions. So Marc, I think that, that will be a good topic for you. You've highlighted both capacity auction and short-term scale training deals since quarter-end. Walk us through the strategy behind these motions, what they told you about pricing. And are they still take or pay?

Thank you, Gili. By now, I think we can all agree that the market is incredibly dynamic and moving rapidly. And we are constantly looking for ways to discern signal and understanding in order to deliver a better and more complete understanding of our take-or-pay business.

Now we launched these initiatives to, first and foremost, learn and validate while building our customer relationships in a disciplined way. What we described earlier as short-term scale training capacity deals are specifically customers that are looking for dedicated scale, in this case, GB300 clusters, for 3- to 6-month engagements at a premium. These customers have specific requirements, like time-boxed large-scale training runs ahead of a model release or reinforcement learning post-training sprints.

In the model world, we hear over and over again, weeks matter, and so does access to reliable performant AI compute. These customers are willing to pay for speed and certainty for a controlled span of time. Now on the other hand, the auction is all about clear price discovery. In today's environment, the price in the market can change literally as we move through a sales cycle. The challenge is to get a solid gauge on the fair value of our offering at any moment in time. The standard reference points, competitive pricing, analyst opinions and even prediction markets are all over the place. In a market where we have several buyers for every GPU, we let the market tell us directly.

The auction result was a price 15% higher than we've ever seen before and 20% higher than our pipeline for Blackwells. Also, the winning bidder is thrilled with the experience. They called out specifically the combination of validated pricing and the certainty of access to the compute they need. Also, they told us they plan to participate in future rounds.

So what you should take away is that these are both customer value-confirming and independently lucrative. We are using a small portion of our overall capacity to drive price discovery and value-validating initiatives. And will help -- these will help us to evaluate our capacity and the market as we move into '27. They will both have a broad impact across our pricing, packaging strategy, deal negotiations, and they inherently have faster close to deployment cycles.

Thank you, Marc. We're getting a lot of questions around the deployment of more than 1 gigawatt of new capacity starting in 2027. Dado, can you talk us through how we're planning on funding this CapEx at scale? And how do we prioritize our funding sources?

Absolutely. So what is important to understand is actually that we have multiple pools of capital available to fund our growth, and we are very comfortable with our ability to finance the capacity we plan to deploy in 2027 and actually beyond. Clearly, the first source is operating cash flow. We are already generating positive operating cash flow and expect this to increase meaningfully as the business scales.

Then we have our customer prepayments. The terms we secured in Q2 cover approximately 50% to 60% of associated CapEx. But going forward, one of our objective is to increase this coverage even further.

Third, our long-term contracts with investment-grade customers give us a very strong base to raise asset-backed financing on attractive terms. We currently have approximately $40 billion of committed backlog that we can borrow against, and we began utilizing this source of financing already in Q2.

But beyond that, we have additional flexibility through corporate-level debt and equity-linked financing, which we have not yet meaningfully tapped. We will continue to balance these sources in a disciplined way to maintain a strong balance sheet as we scale.

But we also see financing opportunities that are additional to those emerging around GPUs as an asset class. We have seen increasing activity and interest in financing GPUs as stand-alone assets, and we are strong believers in the potential of this market. Over time, this could provide us with another attractive source of funding.

Overall, we are really very comfortable with our balance sheet today and confident in our ability to finance our growth in '27 and beyond through this diversified funding strategy while maintaining financial discipline.

Thank you, Dado. The next question is from one of our investors on the portal. Roman, what are we seeing in the market around token maxing? And is that having an impact on usage or adoption of Token Factory in Tbilisi?

Yes. Thank you for the question. First of all, I wanted to start with a reminder about our first inflection event that we held in June, and probably that was the first time when we publicly described our product strategy where we built layer-by-layer, the AI cloud to meet developers where they need us from scaled bare-metal infrastructure to the agentic layers. And the obvious trend is that AI systems are moving into the production. Coding is the most visible example, but this is not limited to coding. We see long-horizon agentic workflows in financial services, such as use cases at Revolut and MasterCard, e-commerce applications, improving the customer experience and process at Shopify and many more. For example, Sword Health in health care or Higgsfield in marketing automation and other verticals.

We are also, as always, customer zero for ourselves. Echo, our own infrastructure agent shipped at the last cloud release runs an open source model served by Token Factory. The challenge that we see for the customers is scaling complex systems that combine multiple models, inference engines and tools. We help solve this with our suite of services, including Token Factory for reliable high-performance inference and post-training, and Avride for grounding, particularly as companies move from closed ecosystems with built-in search towards more open ecosystems.

Eigen AI and Clarifai teams are fully integrated into Token Factory effort and already shipping inside our road map. We see day 0 support for the major open model launches, measurable performance optimization post launch and independent benchmark, as I said earlier, continuing to rank us among the leading inference platforms.

Q2 was also Tavily's first full quarter within Nebius. Its developer community grew to more than 2.5 million developers from 1 million in February. And it launched keyless paper search built for autonomous agent consumption. And it achieved the set of certification for enterprise deployments.

We also see more customers bolstering their own models, creating additional demand for inference and grounding throughout the development cycle, not only in production. Reinforcement learning rollouts, evaluations, synthetic data generations and ground and training workflows all require significant inference capacity and reliable access to external information. These trends validate our strategy of building a vertically integrated platform that delivers attractive total cost of ownership and supports diversified workloads.

We want to serve customers across the full AI lifecycle, from training and post-training to inference and grounding. This is where our infrastructure and software come together as a strong offering, and we are only getting started.

New workloads are bringing new requirements to the physical infrastructure as well. And again, we can address them as we build the full-stack. And not only bringing new generations of GPUs that Andrey mentioned, agentic orchestration, tool calling and data preparations are CPU-heavy. So we are adding ARM and CPU deployments alongside the GPU fleet.

Thank you. And that concludes our earnings call for today.

This concludes today's call. Thank you, everyone, for joining. You may now disconnect.

Perfect. So I have to state my name at the beginning for the record. My name is Tal Liani. And I am very happy to host Roman Chernin, who is Chief Business Officer of Nebius, which is basically, as far as I understand your position, it's a product manager. It's a head of product. It's a product guy.

It's different things, but I like to say that I do what needed now for the growth. And sometimes, it's more on the product side. Sometimes it's more on like go-to-market side.

Yes. Got it. Okay.

It's a privileged position.

And I just mentioned it, so we're not going deep into the numbers. I want to talk to you about strategy. I want to talk to you about basically where your product is heading, what's your advantage in the market. And let me frame the discussion first as I normally do in kind of this meeting. So selling data center capacity now is selling water on a hot day in the desert, in middle of the day. So if you want to make it kind of even more dramatic. And the question is, what is the value that Nebius brings to the market? Meaning, is the growth today sustainable longer term? The question is, do you -- is there a differentiation between Nebius and CoreWeave and maybe the data center of Oracle and Microsoft and all the hyperscalers. So I want to ask you about basically, it's the sustainability of the growth, but it's coming more from the product side. I want to ask you about your differentiating factors. So maybe with that, this is kind of an open statement just to frame the discussion. Maybe with that, you can talk about, what -- how do you craft your product strategy? Meaning what are you trying to be in this market? And we'll take the discussion from there.

Yes, it's a very broad question, I think, but let me start from the customers because I think that -- you said like selling data center capacity. Like in reality, we don't sell data center capacity. We sell product that built on top. And like we build product for the customers. And to discuss it like first, we probably need to structure how the market looks like. And we look at the market and the product that we built in the following way. So there are new customers, like everybody talk about, hyperscalers or super labs that need a lot of compute, but they need only compute. They literally don't consume almost any additional services. They need scale deployments. It's not easy to deliver those scale deployments, but at the end of the day, it's most like basic, not differentiated service. Then on top of that -- and this is like what people call bare metal compute. On top of that, there are much more -- much bigger population of the customers, let's call them AI-Native Labs or Neolabs, hundreds, maybe thousands of customers that need infrastructure, but they prefer to consume it in a managed manner because they don't have all the full stack of their own software, and they want to just focus on their research, they want to focus on their training tasks mostly. And for them, we built -- that was actually the first -- probably the first category of the customer that we serve, and the build out of what we call multi-talent cloud. So it's -- again, it's managed infrastructure. But then there is the next layer of the customers. They -- probably, it's much bigger population again. People who don't want to deal with the clusters. They don't want to talk in terms of Kubernetes clusters, GPU hours. They consume models as a service. And these are the people who build product. you can call them vertical AI products like Cursor's of the world, like in coding, I don't know, Harvey or Legora in legal, Gamma in content, Clay in CRM and so on so forth. So those are the customers that even don't think in terms of the cloud, they think in terms of the models they consume and most of them started with the closed models, then for many reasons that we can discuss, they diversify their consumption towards open source or specialized models. And they need the next layer of the product, and we build a managed inference platform called Nebius Token Factory for them. But probably, this is not the final stage of the market. Now we see that a lot of people are building agents. And the agents is a new type of the application, if you want. If developers who came for tokens, they take all the orchestration, all the burden of building kind of end product on them. People who build agents, they even don't want to consume tokens. They want to get the final results of like agent execution, the outcome of the agent. And probably like -- it's a little bit like speculative, but probably they will consume the AI computing their way. They will not choose the model. They will not compare tokens from this model and tokens from that model, they will have some new level of kind of obstruction or new primitives, how they consume. And so answering your question, how we build the product? The essence of what we built is the AI infrastructure. If you want, in the simple words is the AI compute. Obviously, it's not only compute, it's not only GPU compute, you have storage, you have like CPU compute and so on. But the essence is infrastructure. But we think about the product in a way to follow the customers' segments and customers' workflows and be always relevant for the next wave of consumption. So we could just have bare metal compute. But then we -- our addressable market would be limited to a handful of big customers. We could stay on cloud level and provide managed infrastructure, but then we could serve hundreds of whatever, first thousands of customers. We could stay on the level of inference, but then maybe it's like whatever, tens of thousands of customers. We believe that along the way, along the adoption to AI, they will be the market of tens of thousands, hundred of thousands, we don't know, developers, builders who will build on their level of level of obstruction and our product strategy is to meet them there.

Right. And you talk about the full stack here, basically from compute to software and then deployment and integration. And -- is it what's driving revenue growth today? Or is it more in the future? Meaning what you're seeing today, did the market start with just pure compute capacity and then everything else you talked about will come in the future? Or does it start already from [indiscernible].

Yes. Absolutely. It's absolutely happening. And talking about full stack, it's important to talk about full stack in upstream and full stack downstream. Downstream, it's about how you build the infrastructure and how you actually control your supply chain and cost structure and upstream is how you evolve your product offering, right? And definitely, we already see that, for example, inference is the fastest or fast-growing segment. And we definitely see that agentic workloads like [ starting ], and we can expect that they will continue to grow. So I think that even though that the big part of the market is still like sitting in the training, the opportunity to serve inference workloads, the opportunity to follow like the new growing customers gives you much more flexibility and gives you much -- give us as a provider and the platform, give us much more flexibility and optionality on how we build our customer portfolio, how we build our contracts portfolio, and how we benefit from the motion on the market, like obviously, we know that the prices are growing, and obviously, more flexible workloads like inference let us benefit from that. So yes, it's already a significant impact on the business, positive impact on the business.

Is there a difference between what you're offering and what the other Neoclouds or hyperscalers are offering?

Yes, I think so. And again, it's important to define the categories because people call Neoclouds quite different animals, let's call it. There are almost pure data center operators. There are people who have data -- don't have data center and just aggregate -- don't own infrastructure and just aggregate compute. There are people that provide some software layer on top and people who do nothing and just like do the bare sales like wholesales of bare metal. I think like if you look at the market of the Neoclouds, it's like -- we don't like to speak about others. We like to speak about us. But I think it would be fair to say that from this full stack approach, both downstream and upstream, we're probably one of those who are the most sophisticated. And that gives, again, downstream, it gives you a lot of advantage on how the economics work and upstream, it gives you a lot of optionality how you can work with the customers and what customers you can serve. And eventually, it actually gives you the same economical advantage because I think it's like told even publicly that we have 3, 4 customers for each GPU competing together. And you can think about it that the more competition from demand side you have, the more lucrative kind of business you can do because you can pick the customer that actually value what we deliver more and have the better economics and create more value for the customer, which is also important. There are customers, again, there are customers that just need data centers. Okay, it's not just not our business. But there are those who want to focus on what they built in the product, and they want much more value from this -- from the provider. And this is probably the customer that will value what we do.

So if I generalize and I'll say -- the larger are customers probably that's the hyperscalers probably want the lowest value that you can offer. And enterprises probably want to have the highest value. How do you balance between the two? Because at the end of the day, you're selling to both. You're not selling to only 1 group.

So first of all, I think we told many times that in different occasions that we believe that our long-term business is in a diversified portfolio of the customers, AI native enterprises, start-ups and grow up and more established companies. Even though we appreciate the chance and we cannot really learn a lot and gain a lot of working with the biggest customers of the market, which are hyperscalers. And from the business perspective, those customers drive the growth for us because making the -- making business with customers like Microsoft or Meta actually open up like much more opportunities to finance the rest of the business for us. And if you think what are the drivers for our business, this is obviously demand and demand is there. This is the capacity, like how fast we can build and bring online compute. This is the product that we speak a lot. And this is the capital. And the capital is a very important component, and we by growing through the large contracts and large engagement that we have with hyperscalers, we have ability to grow faster and finance more aggressively in the rest of the business. But eventually, our goal is to have as much of the business in the like diversified kind of real cloud business and not wholesale business like of the large blocks. And then within that kind of part of the business, we try to build a very, again, diversified portfolio. I say it many times, the more diversified. But I think this is like a real philosophy of what we do because we want to have like a lot of optionality. And we diversify the customers from their architypes. We diversify the customers from their type of workloads, and we diversify the customers like from their terms. We have long-term deals, we have short-term deals. We have some spot capacity that we can sell them a premium because it's available right now, we can sell something in advance. So like -- and again, it's not only driven by our willingness to have diversified portfolio, but the fact that we have different offerings for the market, let us have this different customers and different contracts.

Right. As the workloads move from training to eventually inferencing, how does it change the economics of your company?

First of all, I think like no doubt that this motion is happening because training is one-off investments to build the product and inferences is that -- it's a recurring part. What is really exciting is when you're engaging with the customer, with the partner and you support their inference needs, you align with them because the better business of our customer, the more they grow, the more business we have with that. So it's like much more aligned -- a much more aligned business model. Then how it changes economics of us is it's a good question. Actually, again, if we come back to what I said about different layers of the product, when people come for the training infrastructure, it's infrastructure sale. So even though we provided in the cloud in a managed manner, most of the customers know what they want. They come for the particular GPUs for the particular time of the time spent and so on.

In inference, it's much more flexibility on our side that we can extract value through the software. For example, we can optimize different workloads for different types of the chips. And we are -- we not necessarily commit customers for a particular cluster or for particular even type of the hardware, and you can abstract it through the software. And on the practical side, for example, it led you actually extend the lifetime -- the valuable lifetime of the CapEx investments because the new chips come, you can sign the first contract for example for the most frontier training jobs. And then when those chips going out of this like we have a first contract, you can still utilize them for the inference workload. The famous Anthropic SpaceX contract is one of the illustration that SpaceX move their training workloads, the more advanced cluster. But the chip they procured like in the first cluster, they are still useful for the inference. And this is just the illustration of the life cycle, if you rent.

I understand. When you go to agentic AI, how do you monetize agentic AI, when you go to agentic AI, explain how your business evolves with agentic AI deployment?

Again, it's -- you can think about it as yet another layer of obstruction on top of compute. So again, bare metal, you sell megawatts. Managed infrastructure, sell GPU hours. Tuck-in factory inference, you sell tuck-ins. Agentic you sell agentic [ cutoff ]. You want to task be solved. Then what's happening under the hood, you consume the ton of compute through inference, through the sandboxes, through other calculations. But you abstracted from the end customer and developer. And so how you monetize it, still selling infrastructure, but it gives you as a platform, another layer of optimization capabilities because now, for example, like I told about inference when customers not choosing what GPU to use, I can optimize for the GPU. At this layer, not customer choose what model to use, I can optimize what -- from each model, how many tuck-ins to extract. And so it gives you the flexibility. And this is actually people ask, why do you build software? Do you monetize software? You not necessarily monetize software directly but you deal the software to better -- like to unlock the new use cases and give you more lever of optimization for the customer as a result for yourself.

Got it. You made 2 acquisitions recently, made acquisitions, Eigen, hopefully, I pronounce it properly, Eigen and Clarifai. Can you take us through them? What was the rationale behind the acquisitions?

Yes. The rationale is quite simple. So these 2 are great teams, quite rare talent in the market, both actually in the area of inference optimization and model post training. And they both from different angles will accelerate our journey with our inference platform, first of all. So Eigen is the team here in San Francisco, very research kind of driven. The founders are MIT PhDs, and they did a lot of work around how you take model and extract more value from one GPU, in a simple word.

Then Clarifai is another team, the core team on the East Coast, and they more focus on inference as a system. So you can think about it as like one capability is you have a model, you have one GPU and how much tokens you can generate. But I imagine you run scaled system, you run on thousands of GPUs for millions of users. And then you're not only optimizing how one computer works, but you have entire system optimization. How your cash works? How you orchestrate compute? How you efficiently after scale, like when the spike workload comes, how quick you can get new nodes up? And then when this spike goes down, how quick you can scale down and so. So the Clarifai is more team that works on inference job on an inference problem, but as a system. And so combining it together and actually combining with our in-house engineering capabilities, we believe that we now have pretty strong maybe one of the best teams to build inference as a big system. And again, it converts directly to economics because the customer needs tuck-ins with the best quality, best price and best performance. And these teams will just improve the product and economics and accelerate the growth for us.

None of you, meaning you, the neocloud companies, none of you speak about unit economics, but -- so I'm not going to ask you directly about unit economics, but I want to ask you differently. Do you think your focus on software? Do you think that focus on token factory and all the other services, do you think that it translates in reality to higher margins for you versus competition?

Yes, absolutely. And again, it's very -- again, like a very simple kind of thinking. Let's even put aside, which is not true, that for different workloads, people are ready to pay different cost because of like different value that we can deliver. Let's think very simple. We are in the world of supply/demand balance. I spent quite a lot of time in advertisement. And advertising business is very simple. The more hot options you have, the more prices you have. So imagine you have a product that can serve 10 customers in the world, and you have a product that can serve 10,000 customers in the world, probably, that will drive your margins up just because you don't have better supply-demand more like for us as a supplier, the supply-demand situation. Even we've put aside that when people buy inference, we can do much more optimizations under the hood and sell the same GPU with the best economics for customer and for us, by the way, even forget about it. Even forget that like on large deals like hyperscalers like probably have more buying power than some -- just if you target much bigger population of the customers. Just only that gives you -- you can assume, gives you a very significant push of your market margins.

Got it.

And that doesn't mean that we don't have all the other factors.

Right. I understand. How do you manage supply constraints? How do you manage -- even from a contracting point of view, you signed some -- your duration of contracts is much shorter than the competition. You go for shorter durations. But still, when you sign a deal and you need to bring capacity on board, how do you manage the risk of commodity pricing going up?

First of all, I'm happy not to be the supply person in our company. Being on demand side, it's the easiest job in the world comparing to being on supply. I think there are -- like there are a few questions in what you're asking. So the main constraint is still the capacity, like not clusters built, but data centers and connected and ready to bring GPUs online. And I think the most important move that we have there is we announced few -- we have capabilities to build. We are not only renting. Of course, we're renting temporarily because it accelerates time to value for us. But we have the team with the expertise of building super efficient data centers from the ground up from the greenfield. And starting from late this year, next year, very significant portion of the new capacity that we will bring online will be in the data centers that we built ourselves. And building ourselves give us so many advantages, both on the cost structure, but also on the control and flexibility on when we bring capacity online. So this is the first factor.

The second factor, if you think about us, we don't -- we are not dependent on one data center. We are managing -- like our customers are living in the cloud environment, most of our customers. And we think about our data centers as a portfolio diversification again. So we run like a dozen of projects in parallel. And maybe some of them will be delayed, maybe some of them will be even like not successful. It will not impact our ability to deliver to the market because we kind of oversubscribed in a way. This is the second factor.

The third factor, diversification of the workloads gives you more flexibility. For example, inference workloads, training workloads and the biggest contracts even if we put aside the hyperscalers, the biggest training workloads require a lot of compute in one place. You need the large clusters. Inference is distributed workload. So you can manage your -- again, you can manage your portfolio of data centers in different regions, in different like timing, very flexible. So that's what's helping us on the key -- kind of key bottleneck, key driver of being able to supply.

And then on commodity side and like chip side, I think that we pretty much well said. So our biggest contracts are locked and supply for them is locked. And I also wanted to say that, in the current market, actually the -- I think the effect of like on the prices, from demand-supply situation versus the cost structure and like raising some of the components like the supply-demand situation is much more strong factor. They're like, again, the demand responding to the supply constraint.

Got it. We have 3 seconds left, unfortunately. I didn't leave enough time for questions, but we can take probably 1 or 2 minutes from the break. Anyone has any question, please raise your hand. No, very good. Roman, thank you so much.

Thank you for questions.

Thanks for enlightening us.

Welcome to Nebius Group's Q1 2026 Earnings Conference Call. [Operator Instructions]

I will now hand over to Gili Naftalovich, Head of Investor Relations, to start the call.

Hi, everyone, and welcome to Nebius's first quarter 2026 earnings conference call. Joining us on the call today are Co-Founder and CEO, Arkady; and our CFO, Dado, along with the broader Nebius executive management team.

Now I'll quickly cover the safe harbor. Some of the statements that we make today regarding our business operations and financial performance may be considered forward-looking. Such statements are based on current expectations and assumptions that are subject to a number of risks and uncertainties. Actual results could differ materially. Please refer to our Form 20-F, which is a list of our risk factors. We undertake no obligation to update any forward-looking statements.

During this call, we will present both GAAP and non-GAAP financial measures. A reconciliation of non-GAAP to GAAP measures is included in today's earnings press release, which is distributed and available to the public through our Investor Relations website located at nebius.com.

And now I'd like to turn the call over to Arkady.

Thanks, Gili, and welcome, everyone, to our call.

We have had a great start to the year. We're building an AI-native hyperscaler. And I would say we are developing it across 4 dimensions. The first is capacity and scale; second, product and functionality; third dimension is customers and demand; and finally, capital, our fourth dimension. All our focus is on execution across all 4 of these dimensions.

Let me put our results of the quarter in this context. First, on capacity. As you see, we are building big. Last quarter, we told you that we already contracted more than 2-gigawatt of power while targeting more than 3 gigawatts by the end of the year. 3 months later today, we have already contracted more than 3.5 gigawatts, and we are now targeting at least 4 gigawatts of contracted power this year. Today, we announced a new site in Pennsylvania to support 1.2 gigawatt of power once fully live. This is our second owned gigawatt scale site in the United States.

Our platform is most efficient when we own the full stack, and we are building towards that. Our owned contracted capacity now accounts for more than 75% of our total power. But more importantly, we continue to build our full stack platform, and this is our second dimension. What does it mean? It means we don't just offer compute. We offer cloud services, services that span across the AI life cycle from bare-metal to multi-tenancy to inference to agentic and more. And we have made significant progress on that front.

And it's not just developing our platform and launching Aether version 3.5 this quarter. Our 3 acquisitions this year, Tavily, Eigen and Clarifai demonstrate the uniqueness of what we're building. All 3 companies bring industry-leading engineers and researchers to Nebius. Eigen AI and Clarifai strengthened our inference optimization solutions. Eigen was recognized as the #1 speed inference provider by NVIDIA. While Eigen optimizes at the model level, Clarifai optimizes at the system level, and they both strengthen our in-house Token Factory offering.

We also acquired Tavily earlier this year, extending our platform reach to agentic search, an increasingly significant part of the market. This acquisition brought us rarer capabilities of what this new class of developers need. We also expanded our technology partnership with NVIDIA. We again achieved NVIDIA Exemplar Cloud status this time on our GB300 for training workloads. We are among a small group of providers to achieve this status across multiple GPU generations.

At our core, we're a technology company. We have top AI engineers and deep proprietary expertise across every layer of the stack, both hardware and software. We're quickly becoming a magnet for top talent. We're happy with our ability to enlarge our offering through strategic acquisitions. Our clients appreciate the full extent of our offering. This is not common in our market. This is our strengths, and this is our uniqueness. And we believe this is what will enable us to win.

Demand is our third image, and it continues to be increasingly strong, but more importantly, our full stack platform allows us to capture and service a large and diverse range of hundreds of customers, not just several big bare metal offtakers. Our pipeline generation in the first quarter grew 3.5x over the fourth quarter, and this is a record for us.

And the demand is broadening across industries. Today, we typically see several customers competing for every GPU we bring online. We're building to support this demand with scale and discipline. New customers across a number of use cases are using our full range of offerings to solve their most challenging problems. For example, European fintech leader, Revolut, recently began using our Token Factory. In physical AI, 1X Technologies is using our cloud platform to build general purpose robots. In life sciences, our cloud platform is enabling start-ups to build more powerful models that accelerate drug discovery and advance the fight against the leaders in ways that were previously impossible.

And beyond technology sectors, larger companies in industries such as manufacturing, energy, heavy equipment and pharmaceuticals are increasingly engaging with us. Demand is high. Everything we build with is sold. That is what is driving us to build more and to raise our 2026 CapEx guidance to between $20 billion and $25 billion, which is up from our prior range of $16 billion to $20 billion. We increase -- this increase reflects investments in our 2027 capacity that will come online early next year. We expect these investments to contribute positively to revenue in the first half of 2027, where we already have customer commitments in place. Meta is one such customer.

We need to invest to fully realize this. This requires capital, which is our fourth dimension. We're doing a very good job in tapping the market at scale. We raised significant capital this year, more than $6 billion. More than $4 billion of that came from converts and $2 billion from NVIDIA equity investment. This leaves us with a strong cash position of more than $9 billion.

More importantly, we have laid the foundation to raise substantial further capital this year. There are a variety of ways for us to do this. There is our recent Meta contract. First, let me just say that we are very proud of our relationship with Meta, and there is tremendous respect between our tech teams. Formally, this is a $27 billion contract with Meta. But in fact, it's worth a lot more for us. This contract alone can unlock billions of dollars of capital for our own multi-tenant cloud at attractive rates that may not otherwise be available to us.

On top of this, we also have our first contract with Meta and our Microsoft agreement that will provide additional financing opportunities. Obviously, there are many other untapped options for us to finance our public cloud build-out from the significant prepayments we get from customers to asset-backed financing of our payment contracts to corporate debt and so on.

So to close, it has been a great quarter. We are even more focused on what is ahead. We will continue to execute, expanding capacity, building our cloud platform, expanding our customer reach and financing growth diligently. Everything we build, we sell, and we are still in the very early days. I want to thank our team for the incredible work day after day and night after night and to thank our shareholders for your continued support.

And with that, let me hand it over to Dado.

Thank you, Arkady. Indeed, we are off to a strong start to the year with a number of important achievements. First, we accelerated revenue growth during the quarter. We also significantly expanded our margins, and we strengthened our balance sheet. I will touch on each of these, share some color on our results and conclude with guidance. Please note that all comparisons are year-over-year unless noted otherwise.

So let's start with the revenue and ARR. In Q1, we grew the group revenue by 684% year-on-year to $399 million, up 75% from Q4. Once again, we sold out our capacity as demand continued to exceed available supply. Our Nebius AI business, which excludes our consolidated investments in TripleTen and Avride delivered even stronger results. Revenue grew 841% from last year to $390 million, representing an 82% quarter-over-quarter increase and 98% of group revenue. Growth was driven by capacity scaling and was further supported by strong utilization of pricing.

Annualized run rate revenue for our Nebius AI business reached $1.9 billion at the end of March, up over 50% from $1.25 billion in the previous quarter. As we delivered strong top line growth, we also remain focused on profitability. Group adjusted EBITDA was $130 million compared to $15 million last quarter and compared to a loss of $54 million a year ago. Group adjusted EBITDA margin was 32%, continuing the inflection in Q4 and reflecting operating leverage in our model.

Nebius AI business adjusted EBITDA margin expanded to 45%, up from 24% in Q4. This improvement was driven by strong revenue growth. The gap between group and Nebius margin essentially reflects our investments in Avride and TripleTen. Both are still early-stage companies and require substantial operating investments as they scale. We expect Nebius to represent the significant majority of group adjusted EBITDA for the foreseeable future. As mentioned in the past, our intention is to find strategic and financial partners for these businesses and they consolidate them in the future. Net income of $621 million benefited from a valuation adjustment on the back of ClickHouse recent [ funding round ]. This is a noncash item that captures the growth in the underlying value of the asset.

And now turning to our balance sheet. Since our last call, we have continued to strengthen our financial position. In March, we closed a private offering of convertible senior notes, raising $4.3 billion in gross proceeds at attractive premiums and coupons of 1.25% and [ 2.60% ]. In the same month, we announced a $2 billion equity investment from NVIDIA, reinforcing our alignment with one of our key strategic partners. Prepayments from our customers also reached a new quarterly record.

Operating cash flow of $2.3 billion was up from an operating cash outflow of $198 million in Q1 last year. The sharp increase was primarily driven by upfront payments from our customers. Together, these sources of capital increased cash and cash equivalents to $9.3 billion at quarter end.

Now let's speak about our CapEx. As Arkady mentioned, today, we are raising our CapEx expectations to $20 billion to $25 billion for the year. The expansion of our infrastructure footprint remains one of our highest priorities given the strength of market demand and customer activity. We are building for 2027 demand where we have customer commitments already in place. And so we have near-term visibility into future revenue associated with this investment. As always, we will invest in capacity with discipline and rigor, including the capacity we are bringing online in 2026.

In terms of how we deal and how we will fund the capacity in the year ahead, we will continue to leverage a diversified range of funding sources. On the debt side, during the past year, we built our ability to take on debt capacity. For example, with our Microsoft contract and our 2 Meta contracts, we expect to unlock the ability to raise significant capital through asset-backed financing. We expect this to be at attractive terms based on Microsoft and Meta credit ratings, and we will inject this capital into building our cloud business.

In addition, we expect to raise corporate level debt. We plan to start tapping into these financing options in the near term. And on top of that, our financing options include our at-the-market program. We have not utilized this program to date, but we are evaluating change on it. Obviously, we are very focused on generating prepayments from our current and future customers in order to reduce the capital needed from equity and debt financing.

We may also evaluate other financing options, but we'll ultimately pursue whichever vehicles have best the long-term interest of the business to support our expected capital spending in 2026. The bottom line is that as of now, given our strong balance sheet and the work we have done putting in place the various long-term contracts, we have laid the foundation to enable us to access a wide range of potential funding sources.

And now turning to our outlook for the year. While it remains early in the year, our strong Q1 performance reinforces our confidence in our annual targets. As such, we are reiterating our full year 2026 guidance for annualized run rate revenue of $7 billion to $9 billion, group revenue of between $3 billion and $3.4 billion and group adjusted EBITDA margin of around 40%.

3 key parameters will determine our growth profile and margin progression throughout the year, utilization, pricing and capacity. At present, neither of the first 2 parameters is limiting our growth. The third, capacity will play an important role in unlocking our growth potential and driving margin flow-through.

On utilization, we continue to sell out our capacity. and we expect this to be the case for the foreseeable future due to strong market demand and our healthy pipeline.

On pricing, strong market demand is translating into pricing gains in our latest sales.

On capacity, the time line of deploying the new capacity impacts both top and bottom line results from quarter-to-quarter.

We anticipate a nonlinear quarterly adjusted EBITDA margin progression during 2026. We will see this in Q2 given the back-end weighted nature of the capacity we bring online. These investments unlock growth by increasing capacity substantially from Q2 to Q3, leading us to be confident in our adjusted EBITDA margin returning to Q1 levels in Q3 before moving even higher in Q4. Overall, we are confident in our full year targets.

In closing, Q1 was another quarter of rigorous execution across the business. We delivered a strong revenue growth, margin expansion, new business wins and continued capital discipline. As we look ahead, we will continue to scale rapidly to capture the tremendous market opportunity ahead, while remaining balanced, disciplined and focused on delivering long-term value for our shareholders.

With that, I'll turn the call back over to Gili for Q&A.

[Operator Instructions]

The first question from our investors on the portal is from Alex Duval at Goldman Sachs.

To what extent have you started to see the impact of stronger GPU pricing reflecting in your core AI business? Additionally, is there a way for us to think about the share of older shorter-term contracts that could benefit from this pricing dynamic? Marc, would you be able to answer this one for us?

Thank you, Alex. We continue to see strong pricing across both old and new GPU generations as demand continues to exceed our available capacity. We just raised prices again in the latest quarter, and we are still selling out across all chip types at the higher prices. We're in a very dynamic market, and we have built a resilient set of processes that allow us to adapt and respond accordingly in any market environment for both new and existing customers.

The strength is showing up in a number of ways beyond just price. Contract durations are extending with the average duration of contracts growing meaningfully over the past few quarters. Also, average contract values continue to increase across new logos and existing accounts where we are seeing strong expansion as well.

And finally, prepayments are becoming more significant. Customers of all types are prepaying in order to lock in future capacity, including the hyperscalers. This improves our working capital position and gives us flexibility around external financing needs. Our go-to-market model is being built to be agile and adapt to the market and yield outcomes that can best help us continue to scale our business.

Thanks, Marc. We have a few questions coming in on the CapEx guide and cost inflation.

Andrey, can you please discuss how much our raise in CapEx is driven by higher capacity growth versus component cost inflation?

Sure. Thanks, Gili. Well, the increase in this spending is driven by visibility into 2027 and our need to invest ahead of capacity that we expect to bring online. We will add much more capacity in first half of '27 than this year. And that requires more CapEx spend in the -- well, starting from now in the later part of this year. We have been able to secure sites and power and customer commitments for 2027. And so we are ramping up construction activities accordingly.

And in short, the high number reflects confidence in our contracted demand pipeline and our ability to secure the infrastructure that we were against it. It's not the cost pressure. The impact of the component inflation in our 2026 program was quite material, around low single digits as a percentage of total spend also because we secured a lot of 2026 back in 2025 at the previous price levels.

Thank you, Andrey. Next question we have is from James Kisner at Water Tower Research.

Nebius said AI cloud adjusted EBITDA margin nearly doubled quarter-over-quarter to 45% in Q1, while you're targeting around 40% for the full year. What's driving the implied step down? Can you walk us through the adjusted EBITDA margin progression for the year? Dado, please.

Thanks, James. Indeed. As you saw in the quarter, our Q1 margins were really strong. Nebius AI adjusted EBITDA margin reached 45%, nearly doubling from Q4. And that really reflects the underlying strength of the business. On the one hand side, the demand we are seeing in the market, then the terms that we are also able to negotiate with our contracts and the unit economics of the platform itself. So as I mentioned earlier, as we move throughout the year, you will see some quarter-to-quarter variability. And I think this is worth taking a moment to explain the dynamic.

We have made a number of important investments in the first half of the year, hiring across go-to-market and engineering, our recent acquisitions and continued development of new product capabilities. And those investments are already in the cost base today. And we expect them to actually benefit from the business going forward.

On the capacity side, our delivery this year is back-end weighted, and we have a meaningful step-up coming in Q3. We have very clear visibility into both the investments that we have made and also the capacity that we are bringing online. So really, what you are seeing across the quarters is a timing dynamic, not a structural one. The investments land first and the capacity and the revenue it supports come online shortly after.

So given the timing of our investments in Q2 and the timing of the deployment towards the end of the quarter, we actually expect those margins in Q2 to go a little bit lower, returning to Q1 levels in Q3 and stepping even higher in Q4. So for the full year, group, we expect a margin around 40% as we have guided. And on the longer term, those dynamics will smooth over time and our capacity footprint continues to scale and higher-value software solutions will become a larger part of the mix.

Thank you, Dado. The next question is around capacity from Andrew Beale at Arete.

Andrey, maybe I can come to you here. Can you talk about the timing of capacity additions beginning in Q2 and when you expect key sites such as Pennsylvania to reach full capacity?

Thanks, Gili. So Andrew, first, about the Pennsylvania. Pennsylvania is going to have lights up by the end of 2027 with the first around 250 to 300 megawatts probably. And then the schedule looks like adding 300 megawatts each year up to 1.2 gigawatts in total actually and 1.2 gigawatts according to our power contract we have in our concession by mid-2030 or the beginning of 2030 to be more correct, more precise.

But overall, our capacity schedule is just ramping up. This year is heavily towards the second half of the year. Q3 is much -- is a very significant improvement for us in terms of the capacity and going online. Q4, also very significant and then Q1 next year is where our bigger projects like Alabama and probably the first Missouri will kick in also.

Great. Thank you. We'll probably stick with you, Andrey. We have a question from Josh Baer, Morgan Stanley.

Can you address the media reports indicating delays at the Vineland, New Jersey site? Understanding you deliver commitments so far, are there any delays to note for the remainder of the Microsoft contract?

So we delivered all our capacity commitments across our Microsoft and Meta customers. So the first Meta as we already spoke, I believe the first Meta contract was fully delivered in Q1 this year. The Microsoft contract is way more stretched, and we have the delivery schedule up to the end of this year. We delivered the first tranche in November last year. We -- yes, and so we continue to be in the contract schedule. Again, it ramps up starting from the midyear and most of the volumes will be coming in Q3 and Q4.

Great. Thank you. So we have a number of questions around the Meta contract. A question from Alex Platt is, can you provide more details on the recently announced Meta deal? Can you explain how the $15 billion capacity option works? Should we view this as Meta backstopping $15 billion with a set attractive margin? And if you can get a customer with better unit economics on that capacity, would you take that instead?

Marc, let me come to you here to walk us through this.

Thank you, Alex. First, I want to say that we love working with Meta, and we're excited that they chose to buy more capacity from us. This expanded new agreement is to make sure that we all understand this, a 5-year contract for a total of $27 billion, and it is structured in 2 parts. First, there's a $12 billion commitment to dedicated compute capacity with delivery starting in early '27. And then second, as you pointed out, there's another $15 billion of additional capacity that we, at our discretion, can either allocate to Meta or sell to our AI cloud customers as it comes online for the duration of the 5-year contract.

Let me explain this in a bit more detail. Meta is committed to buy up to $15 billion of any capacity in these clusters at our option during the entire 5-year contract. This commitment will likely allow us to finance the clusters with asset-based -- asset-backed financing at attractive terms, while selling them to, as I think you pointed out, to our AI cloud customers at potentially higher market prices.

The unique combination of being able to sell at a premium, along with the commitment by Meta to purchase any capacity during the contract should provide us with higher margins, less risk and more visibility in our revenue. If the market remains strong, we should generate more than $27 billion in revenue from this great agreement.

Thank you, Marc. We have a question from Alex Duval of Goldman Sachs about M&A.

Could you explain the rationale behind your move to acquire Eigen AI and Clarifai? How does this move improve your AI cloud platform capabilities? To what extent does this move mean that you could improve customer stickiness?

Roman, I think we'll go to you.

Yes. Thank you, Gili. Thank you, Alex, for the question. First of all, I want to say that we are super excited with these 2 incredible teams of talented people from Eigen and Clarifai will join us. And to deep dive in rationale, let's start from foundations.

Our view is that we should own the compute stack. That is where our vertical integration, our supply chain depth and our hardware engineering generate advantage and it's also the layer that drives the bulk of our economics. But the compute stack, we built the full cloud solution and software plays the role of enabler. By the way, we partner where partnership is the right path. And as you see now, we use M&A selectively where it accelerates our road map, brings in proven developer adoption or as capabilities complementary to what we are building.

Acceleration is the key lens we apply to every potential transaction where we can find rare talent or proven adoption. This is, by the way, the example of Tavily that has incredible developer adoption that would, in general, take us meaningfully longer to build organically, acquisition is the fastest path. And we evaluate every potential deal against the clear criteria. Does it deepen customer engagement, increase lifetime value, unlock the new category of the customers or use cases we can address and in general, strengthen our position as a full stack AI cloud.

Thanks, Roman. We are getting more questions on M&A, so we'll likely stay with you here.

Several participants are asking on whether Token Factory and software more broadly are distinctly different from the infrastructure layer and training. I would love to get your insights around agentic monetization and the opportunity there.

Yes. Thank you for the question. As I said, we look at the software as an enabler. So it's not that we build the software to generate a separate revenue stream. The software first of all plays the role of unlocking the new capabilities for us, unlocking the new opportunities in the types of the workloads that are growing on the market and the types of the customers that we can address.

Software changed the shape of the customer relationship. Every layer of the software unlocks another group of users and the customers. We want to meet customers where they need us and let them consume our vertically integrated solution in the way that they need, and it might be a different way for different types of the customers.

Customers come to our platform for different needs. In essence, they all need to run AI at scale that need -- which means that they need compute. But for example, people who use our multi-tenant cloud, they -- big extent come for large training jobs. People who -- and these are like research-driven, data scientist-driven workloads. People who come to Token Factory, they build vertically integrated vertical AI products or apply AI in their enterprises. And they come for the tokens. And moving forward, we'll see new ways to consume infrastructure at scale that will be the agentic -- end-to-end agentic workloads.

Thanks, Roman. We have a question from Tal Liani at Bank of America.

How do you plan to finance this additional CapEx? And are you considering disposing some of your noncore holdings? Dado, over to you.

Happy to take this question, Tal. Well, look, our balance sheet is strong. At the end of the quarter, $9.3 billion of cash and cash equivalents, and this was supported by $2.3 billion of operating cash flow, which was generated in the quarter, right, mainly coming from upfront payments from our customers. So currently, more than 90% of the CapEx range that we projected in February is already secured by cash and contractual commitments.

The incremental capacity reflected in our raised $20 billion to $25 billion guidance will be funded through additional financing. And as we have -- as I have mentioned in previous calls, right, so we have a wide range of sources available to us. On the debt side, we expect to use asset-backed financing against our contracts with Microsoft and Meta. And we will -- we may also raise corporate level debt.

On the equity side, we have established an at-the-market program from up to 25 million Class A shares. We have not utilized this program to date, but we are evaluating the program regularly. In any case, as we have done today, we will apply consistent guardrails on cost of capital and shareholder dilution, while maintaining a disciplined capital structure.

Thanks, Dado. We have gotten a few questions on pipeline. One from Nehal Chokshi at Northland Capital.

Your pipeline is up 3.5x quarter-over-quarter in 1Q '26. Does this pipeline include hyperscales like Meta -- like the Meta deal? Can you also provide more details on what this number represents? And how likely are you to convert pipeline to revenue? Marc?

Thank you, Gili, and thank you, Nehal. The referenced pipeline growth of 3.5x a quarter, which -- 3.5x quarter-over-quarter, which we're very proud of, is for our AI cloud business, and it does not include any strategic hyperscaler deals like the Meta deal. It does include qualified opportunities across our core AI cloud and Token Factory products as well as across all of our key customer segments, including AI natives, software vendors and enterprises.

What we can share about conversion is that we have maintained our solid win rates at the same time as we've accelerated our sales cycles and increased our average selling prices. And you can see this with some of the strong wins that we have, such as Sword Health in healthcare, life sciences and Rhoda and 1X in physical AI and core automation in one of our AI-native model builders as well as Revolut and monday.com, new customer wins for Token Factory.

What we are doing is enabling our go-to-market teams to have a consultative conversation with our customers about their plans and for current and future workloads, including, as an example, what they're thinking about with regard to Vera Rubin's. We're also focusing and scaling our go-to-market and success teams to help customers to realize their plans, which turns into durable revenue for us.

Speaking about scaling, by the way, we have a number of recent appointments, including key leaders for the Americas, Dan Lawrence, who is our SVP and GM for the Americas; and John Haarer, who is joining -- who has joined as GM for Asia Pacific and Japan; and Raja Agrawal, our VP for the Middle East.

Thanks, Marc. Another question that we have from the portal is saying that you emphasize the momentum in your software stack. Where are you seeing the most momentum across the stack today? Why are customers choosing Nebius? Roman, over to you.

Thank you, Gili. I think it said so many times by different people that now the time of inference, and we see the same. Inference is the fastest-growing segment, new segment in our stack. And we see a very lucrative place where Nebius is positioned. We have the winning combination with capacity and customers need scale. We have the strong software stack, and we invest in-house and with the new announced acquisitions to be on the top performance of supporting the most popular open source models and specialized models.

We can provide the best total cost of ownership and cost of tokens for our customers through the full stack optimization of the stack. And of course, we care a lot about the developer experience. So we think that in a way, we combine the best from different worlds of specialized inference platforms, the scale of AI specialized cloud -- the scale of hyperscalers and specialization of AI specialized cloud. So Token Factory is our primary inference product now, and we are seeing good product market fit.

If we look on the next layer, on agentic, it's still to be defined what is the final shape of the product and who will be the winners. We expect that Nebius will play the same role of foundation for people to build at scale and will provide the set of tools and platforms to optimize workloads in agentic world.

Thanks, Roman. We have a question on customer concentration. Marc, how do you think about concentration risk given how large your contracts are with Meta and Microsoft? What does the rest of your revenue base look like in terms of customer diversification?

Thank you, Gili. As a reminder, I think we say this over and over again, but I think it's important to recognize that our priority is our AI cloud business. As such, we are very intentional about how we are pairing these key strategic relationships with the likes of Meta and Microsoft with a diversified core of our AI cloud customer base. We do not take these big strategic deals lightly and only take them when we see terms favorable for our core mission, again, serving our AI cloud business.

We also very diligently capacity plan, and we're always looking to add capacity to best serve our core AI cloud customers from developers all the way through to enterprises. Our AI cloud business is experiencing strong traction across all of the products that we're offering as well as customer segments and the verticals that we're chasing. The diversified AI cloud book of business as well gives us both customer and use case visibility that helps to fuel our go-to-market and drives our pipeline and revenue diversification overall.

Thanks, Marc. We have a question from James Kisner at Water Tower Research. On the $2 billion investment from NVIDIA and extended collaboration on inference and agentic software around Token Factory. What concrete deliverables should we expect over the next few quarters? And does the partnership affect the timing or scale of your Vera Rubin deployments in the second half of 2026? Andrey?

Okay, Gili. Thanks, James, for the question. So first of all, the NVIDIA strategic investment is meaningful across several dimensions beyond the $2 billion of equity and line of sight to 5 gigawatts of capacity commitment by the end of 2030 that we've done. It really gives us a multiyear partnership with our most important hardware supplier at the moment when access to GPU supplies competitive advantage, so to say.

We gain differentiated supply chain certainty on the future Rubin, Vera CPUs and the networking. We also have a close collaboration with NVIDIA for the design and early support of the future SKUs. As of today, this is Rubin's and Vera CPU platforms. And we are able actually to have an early deployment that supports on our cloud platform as soon as they will be publicly available.

Again, it reinforces our position as a preferred builder of AI infrastructure and aligns our road map with the NVIDIA product cycle, which is very critical for the price and performance and utilization and just leadership overall. We are also expanding our software integration. Our announcement around physical AI is one example, and we are very excited about our partnership driving vertical specific advancements.

We are also partnering with them to build software for inference and agentic part and just recently achieved NVIDIA Exemplar Cloud status on GB300 for training. We are very much one of the first cloud providers globally to receive for all the NVIDIA generation where the status is available. Yes, that's it.

Thanks, Andrey. Marc, maybe this one for you, the questions on the portal. Some of your competitors have mentioned they are sold out for most of 2026 and even into 2027. If you are future selling, how much of your future capacity is sold out for this year and next?

Thank you, Gili. First of all, we are sold out again in Q1 as we have for several quarters as demand continues to significantly exceed available capacity. The vast majority of capacity coming online over the next several quarters to 12 months is already under contract or earmarked for our AI cloud customers. We do retain a portion of capacity for self-service to serve those AI builders that Roman mentioned earlier. And then we proactively manage those allocations to keep the segment supply as demand evolves. Separately, we are typically seeing 4 or more customers competing for every GPU we bring online. We have significant expansion planned for 2027, including Vera Rubin's, and we'll start selling that capacity as we move into the second half of this year.

Thank you, Marc. We also have a question on U.S. data center opposition. Tom, can you touch on some of the political opposition here in the U.S. related to data center construction?

Yes. For sure, Gili. I mean, so definitely, this is a big topic. It's something that we pay a lot of attention to. But overall, I think what I would say is that the approach that we've taken so far, we found to be quite effective. And I would basically say there's a few sort of components of that approach and how we think about this.

Number one, I think, first of all, not all companies that build data centers build in the same way. We're not alike. And I think you've heard kind of Andrey and team talk about how we build the efficiencies that we're able to achieve, what we do around create interesting technological ways of heat reuse and so on and so forth. So we built very efficiently and very effectively, and I think that's an important part of our story and what we talk about when we come into new regions.

But of course, that's not enough. I think that we also -- the second thing is that we take a very transparent approach to what we do and how we talk about ourselves. I think that's not something that's necessarily universal in our industry. But it's -- from the very beginning when we're looking at a site and we're engaging strategically who we are, we engage very actively in communities and talking about what our plans are, what we do, how we build, how we contribute. You can see us showing up at community town hall meetings or I'm looking actually right now at Andrey Korolenko across the table in Amsterdam it has just blown in from our event in Independence, Missouri yesterday, where we're engaging with the local government and communities. So we're trying to just be very clear and transparent about what we do, how we do and what the benefit that we brings -- that it brings.

And I think the last thing is that, look, when we come into a new region to build, we don't just build and then move on to the next city. These are long-term investments. And so therefore, we have to look at these relationships with communities as long-term partnerships and relationships. So we think very much beyond what we do in terms of the building, but where else we can contribute through it, whether it's through our Nebius Academy, academic offerings, working with local universities, helping to train, reskill, retool and so on.

So we view this very much holistically as a long-term partnership. And so far, we found that this approach resonates well. But there's no room for complacency here. So we continue to pay attention and make sure that we're doing the best we can to be a positive contributor to the local ecosystems.

This concludes today's call. Thank you, everyone, for joining. You may now disconnect.

All right. Before we get started, for important disclosures, please see the Morgan Stanley research disclosure website at www.morganstanley.com/researchdisclosures. And if you have any questions, please reach out to your sales representative.

My name is Josh Baer, software analyst here at Morgan Stanley. We are thrilled to have the Nebius leadership team here today or part of it, Arkady Volozh and Marc Boroditsky. Thank you so much for joining us. The CRO...

Thank you.

Thank you.

Maybe to start it off, Arkady, for those newer to the Nebius story, could you walk us through some of the history and what you're trying to build?

Well, that's not the first project of our team. We have some history. The team of engineers who started this company 18 months ago, the previous 25 years was running one of the largest IT companies in Europe. Infrastructure-wise, it was 200-megawatt of different data center equipment. We were the largest NVIDIA partner outside of U.S., China. So we had some experience.

And then a lot of geopolitical events happened, and several hundred engineers actually went to Europe, and we decided to start a new company. We left in 2022. And then several months later, this ChatGPT moment happened. And we realize that probably there is something for us to do here because we know the area very well. We were in machine learning, not even AI yet for many years. We know how to run such systems, and we were one of the big tech companies ourselves.

And we realize that there is a space in this new era, there will be a lot of -- all the big tech companies, all the big tech systems will go there, but also there will be a lot of independent companies and enterprise market, which will go into this area. They all will need experienced infrastructure, people who can provide them with experienced infrastructure and who actually can make them competitive to the big tech companies. And we thought that this is probably an area where we could be useful.

We started with the company just on summer '24. When we started, we had just one data center with us. That's all we had, 25 megawatts. We managed to sell our assets for several -- $2.5 billion. It was a big -- much bigger company, but whatever we managed to sell, it was nice to start with. So we started with 25 megawatts, 18 months ago. Today, we're running 10x of that, almost. We publicly announced that we will be running another 5x on that. It will be almost 1 gigawatt of active power this year, 800 to 1 gigawatt. And we also announced that we are building much more now. We have a pipeline already reserved for 2 gigawatts, and it will be probably close to 3 gigawatts this year.

But what we are building is not just gigawatts. We're building a real platform for AI developers, a real cloud, full stack cloud. We build the basic layers starting from the ground, land power shell, we build our design and build our own data centers. We fill them with our own design racks. So on the first layer, we are like companies, I don't know, Iron, Applied Digital others who build the data centers. We have it internally. Then we build the racks, it's like, let's say, Supermicro, Dell, they build their racks, we build our own racks.

Then we integrate everything and provide the basic layer, bear metal layer, it's called to big players and whoever wants it. And on top of it, we have a product, which is our own multi-tenant cloud. This is the same kind of cloud as the hyperscalers have. The big names, AWS, Azure, GCP. These are examples of multi-tenant cloud. We have this full stack. We recreated the full stack. We -- again, we didn't inherit any IP rights. We had to recreate it from scratch. And that's why this all cloud is very modern, very fresh with no incumbency.

And this is our new stack. This is a new cloud with a lot of services on top of it. It's not just multi-tenant cloud, which means basic cloud services, multi-tenancy billing, security, storage, managed databases, it's all there. But on top of it, there is a lot of services for -- specific for AI builders like our Token Factory, inferencing platform or agentic search services, which people who build AI need, or just recently, we announced human experts in the loop there. Human experts who actually make -- perform tasks given to them by agents. It's agents running humans today.

So that's examples of services which are on our platform. And on top of that, we have a sales machine, of course, go-to-market, which is verticalized in geographies, verticalized in industries like pharmacy, robotics, retail, whatever. So this is a full stack from ground, to racks, to clouds, to services and the sales machine at all. This is the full stack cloud. And we build it in like, two-dimension, this functionality and scale. And in scale, as I said, it's gigawatt scale. We are -- if you compare us to the big hyperscaler, we are probably the smallest today. But we are the same -- like our cloud today provides hundreds of thousands of GPUs. How many GPUs each of the hyperscalers provides? It's not millions, it's also hundreds of thousands, maybe more hundreds, but the same class.

So we are in the class of -- by scale, hundreds of cloud, full-scale cloud with hundreds of thousands of GPUs in there. We'll go to millions eventually when the gigawatts start materializing. So it's full stack and full scale.

That's a great overview and such an interesting story for a relatively new company, start with a lot of cash, hundreds of engineers with a longer history of expertise to move fast and build this full stack cloud.

Marc, I want to bring you into the conversation, if you could talk a little bit about some of the different customer segments that you're prioritizing, and really what gives you confidence in addressing different -- those different groups?

Certainly. And as you all should have heard, Arkady just laid out a very bold and broad vision in terms of the tech stack that we're delivering and the intent to actually meet the hyperscalers competitively. As a complement, we are also implementing a very broad distribution strategy. Today, we have the privilege of servicing thousands of developers that come to us to build on our platform as well as having contracted with hundreds of customers. It ranges from individual AI researchers all the way up to some of the biggest companies in the world like Microsoft that could spend billions of dollars on our platform.

You could say that because of our AI builder orientation, we have a right to win with AI start-ups. So this has been a big part of our success in the recent past. We do that, as Arkady mentioned, by meeting them where they are with the technical capabilities that the AI builder is looking for. That same readiness has actually won us opportunities with major software vendors that are embedding AI in their offerings. So companies like Shopify and Cloudflare. And then even more recently, we started to see opportunities with the enterprise. We believe in the fullness of time, enterprise will represent the majority of the TAM. Today, it's relatively narrow. So we have a very focused workload-specific set of initiatives where we're pursuing specific verticals, including health care, life sciences, physical AI, media and entertainment, retail and e-tail as well as financial services.

Excellent. Arkady, you laid out a pretty ambitious plan as far as capacity. Can you talk a little bit more about that expansion? I think yesterday, you announced your approval for your first gigawatt site. What needs to occur to reach those goals of 3 gigawatts and beyond?

Yes. As I said, we started with just one data center, which came -- which was built outside of our core market in Finland. That was our first. Then we started quickly leasing data centers because for -- to save time, you have no time to build, you need to lease. The first year, we were leasing everything we could, and we launched locations in France, in the U.K., in Iceland, in the U.S., of course, and in Israel, that was our first year.

And today, we would build -- we mostly built ourselves. And this year, we are launching sites, which belong to us, which we built from scratch. And these are much bigger sites. This was the first example, Independence, Missouri. It's 800-megawatt, expandable to 1.2 gigawatt. Just got approval from the local communities. We have more sites coming like that. We build -- continue building in Europe. Majority of what we build now is in U.S., but we build not only in Europe and the U.S., we are now opening our offices in Middle East. I don't know if we can mention this today. But yes, that was outlined last week.

Singapore and Korea and Japan there as well. But what we build today is sites, we happen to build the largest -- currently the largest sites in Europe. For us, it's not so big, it's 240 megawatt in France. But for today, it's the largest in Europe. In U.S., we are building much more, it's hundreds of megawatts, going to gigawatts side. This is the type of sites we are looking for. And we have dozens of them. Today, we have 16 something locations announced. Again, it's a pipeline. It's like 10, 20 is already announced, and it's in, I don't know, a dozen of different markets. it will be more. It will be a network of data centers. Having your own platform allows you to use your own big sites, but also integrate anything you can find. If you found 20, 40 megawatts data center somewhere, just plug it in, it works immediately get the software platform on top of it. All the sales machine works for it. And for the customers, it's just a cloud. So it allows us to connect small sites, but we're building much bigger sites.

And to clarify, at last earnings, you had line of sight or a target to 3 gigawatts. Does the announcement from yesterday, is that incremental to that 3? Or is that a step toward that progress of hitting that target?

This is actually an example of what we -- I think independence was included into our announced -- because the contract there was closed, means everything was signed and binding before approval by the authorities. It's another step, approval. There will be more steps, always it's a long process. And again, and it goes in dozens of different locations. If one of the location is delayed, fine, we have others going, that's how we derisk ourselves.

This stage yesterday was just passing a very important stage of local approvals. So it was included into 2 gigawatts, but there will be more coming, and we said that we probably will be able to announce at least 3 gigawatts this year. It's a pipeline. It's the table which our guys have hundreds of lines there with sites, potential sites. Of course, the bottom 100 will never materialize, but the top 100 of sites eventually dozens of them will come into life.

That's helpful. Marc, so we've been talking about the scale of what you're building. How does the demand -- what does the demand look like? And what gives you confidence in that there'll be demand to align with this build-out?

Well, I'm sure that over the last couple of days, you've heard from a lot of executives that demand is extraordinary right now. Demand far exceeds our supply. The opportunity for us is actually to optimize our model to be able to build a high-margin, durable long-term business to take our abilities to generate expanding pipeline and drive the future capacity decisions that Arkady just mentioned a moment ago is the vision. Today, we have so much demand that we have to actually prioritize. So we're taking the initiative to shift away from a transactional mindset to a value-based and solution selling approach. Our sellers spend time with customers, learning about near-term and long-term requirements, understanding how our capabilities tie to their business opportunities.

We use that in order to be able to prioritize customers that actually align against our business model. And as a result, we are now driving sale of future capacity at an increasing rate. We're also delivering longer contract terms. As a matter of fact, the last quarter, we shared that quarter-on-quarter, the number of year-long contracts doubled. We also saw a 50% expansion in ASP, and we saw prices go up even on older generation GPUs.

Now we're also -- and we haven't shared this yet, but we've added it to our set of priorities, we're now securing upfront payment from customers, up to and including, in some cases, customers paying 100% upfront. So based on this principles-based approach, we're able to prioritize the kinds of customers, the kinds of deals that we're looking for that support the overall business model that we're going for.

Now you asked about demand. What really excites me about demand right now is actually how we grow with our customers. Customers that have landed on our platform to start with a small project and then expand with us over time. Some customers are doubling quarter-on-quarter. They're expanding the use cases that they have on our platform. But what really gets exciting is when they actually move to revenue generation. And then we're actually winning and growing with them at the same time. and that is a durable opportunity for us that actually will extend into the future.

So when you're talking about securing upfront payment, you're not just talking about Microsoft and Meta, you're talking about the broader customer base?

Correct.

Great. And to your point on sort of growing with customers over time, I think that ties pretty closely to training versus inference, what are you seeing as far as the mix on your platform? And how does that evolve going forward?

Well, first of all, I think we all know this, but I want to say it anyways, for AI to deliver on the commercial vision, inferencing needs to outstrip training period. That's the inferencing at the end of the day, is the commercial realization of the functionality somebody is delivering. Inferencing is growing faster than training on our platform. And we're seeing it along three key vectors: customers that are using our platform to do their own inferencing; customers that are actually inferencing platforms that have been built on Nebius; and then our own inferencing solution, Token Factory.

Interesting vector of demand, I think some of this has been reported in the technical press, is the fact that we're now also seeing software vendors who have built functionality on the foundational models that are looking for improved performance, quality and cost that are now diversifying their models and utilizing open source or even training their own models. And that's opened up opportunity for us to sell our inferencing solution in the software vendor sector. We believe this line of demand is going to continue into the enterprise and be a key way that we're going to potentially be able to service enterprise requirements.

Excellent. Arkady, you have some really large customer contracts and customers, but you also have this very clear focus on the broader AI stack, for general customers and a broader tail. So how do you balance the two, particularly thinking about the build-out of dedicated large GPU clusters versus servicing this broader market opportunity?

So yes, true. Everybody talks about our big contracts, about Microsoft contract, about Meta contract, billions of dollars. And those are great contracts, and those are great customers. But don't get distracted. Our main business model is not there. We are building a full-scale cloud for the rest of the AI builder market.

Big tech companies are great. There's a lot of experience. The teams respect each other when they work together, but our focus is multi-tenant cloud for the rest. We go into these contracts because there is a huge demand from their side they need to grow faster than they can, and they outsource this growth to companies like ours and many others. But we used to be on their side before. We understand how they're thinking. Eventually, in several years, they will be building their own capacity. They just cannot cope with the speed now. That's why they're outsourcing.

But we treat this business for us is temporary. We're just helping them to grow now. Why because for us, we need to grow our part. To grow our part is hugely capital intense. We need tens of billions of dollars to build it. Where we get it, one of the source in those contracts. The contracts themselves bring us revenues. There's many other terms in the contracts which help us to finance our main part. So today, I would say it's only roughly half and half, half of our capacity is serving these huge customers, but they buy just basic services, bear metals. They have their own stack. They don't need our services, higher than just bare metal.

We serve them. We're happy to have them. I hope we are a good outsourcing for them, but we use this, their resources, cash, credit ratings and other things, which we use, utilize to build our own. And this is our main focus, to build our own capacity, which goes to multi-tenancy to the rest of the market to all these AI start-ups becoming corporations, as they grow, all the enterprise market going into AI and building into their processes. This is our main market. And we're building for them.

So today, yes, we're serving these big clients. There's a great contracts, profitable, allow us to finance very good clients. But don't get distracted, our main business is in the second part. And that part further down will be large and large.

That's very clear. So you have these ARR targets, Marc, for $7 billion to $9 billion in ARR exiting 2026. I think the Microsoft and Meta deals get you about halfway. So could we talk a little bit about the other half? Like how much is already booked in your current run rate? And what gives you confidence that, that gap can be -- that you can achieve that through these broader customer relationships?

So you're right, Josh, that the combination of existing customers and our big deals gets us about halfway. What we're targeting with the execution this year is the other half. And in order to deliver that, my confidence is very high, first and foremost, in the demand that we're seeing, so what we already described earlier. And that's with a very modest market coverage. So what gives me greater confidence about our ability to deliver against that is the incremental results that we're going to generate as we expand our coverage this year. So we are dramatically expanding our overall go-to-market footprint across all the regions.

Third is the expansion that I mentioned earlier that we experienced with customers as we partner with them. By aligning against their longer-term requirements and selling future capacity, we're actually setting ourselves up based on their journey towards expanded use of our platform.

And then the fourth, which is selling the platform, as an example, selling Token Factory, having gone from helping model creators to now supporting inferencing and then now more and more developer tooling, we've expanded ASP and opened up TAM and increased the routes that we have into accounts. So across those four vectors, we have clear initiatives and programs to be able to deliver the results that I mentioned.

I also want to make sure and be clear, though, this all lies on the fact that we have this big vision of capacity that we're bringing online. And as I mentioned earlier, when we talked about demand, it's really a supply-driven marketplace. So we have a great opportunity with the initiatives that Arkady mentioned a little while ago.

Excellent. I wanted to come back to a couple of elements on the cost side that you have been speaking to throughout. One, the self-build and the other building your own racks. Could you talk a little bit about the total cost of ownership advantages that you have?

As I said, we are a full stack company, just like hyperscalers. When you talk about the size of the industry, you see the revenues of all the participants of all the companies of each layer, companies who build revenues to build data centers, revenues to build racks and equipment, revenues for different services on top of it. You see the same revenues actually several times. You need to look at the margins and like value-added tax. You see where the value should -- is created and how much in each layer?

So in our case, we are working through the layers. It means that on the sites which we are building ourselves versus leasing. Leasing is great because you get it already now and you can grow fast, but you pay their margins. You can design your own racks or you go and buy racks. You can buy racks, fine, it's faster maybe, but you pay the margins on top. The same on each layer. When you build everything through the layers, of course, we save on all these layers. If you have your own capital, another layer actually of value creation because capital is one of the most expensive layers, so to speak, Here.

We're happy to start with some cash. We are a public company, we were ready to get money through different instruments. So far, we don't have any asset-backed financing involved. We will, but it's not. Whatever we build now is built with our own resources for now.

So we save on all these technological layers. We save on capital. And through all this stack, we hope that our TCO, total cost of ownership, is probably not the highest on the market. And plus we built additional services when people buy just compute, it's not bare metal only, sometimes it is, but for other customers. It's not. It's other services, they buy storage, they buy different tools and services on top, and they pay separately. And it all creates layers of additional value, which we added to our revenues.

So we think that building this vertically integrated platforms is what -- this is where the competition will be because the major players of this market, the three hyperscalers, they all are vertically integrated. They all outsource, of course, now, but basically, it's all theirs. And we are playing in the same league. And we're trying to play not just vertically in the same league, not only through all the services, but also in scale.

To round out the conversation talking about some investments, first in go-to-market, and then we'll cover CapEx. So Marc, thinking about go-to-market, where are your priorities? I mean you sort of previewed that in one of your answers. But if you think about your investments looking ahead and building out go-to-market, what -- where is your focus area?

First and foremost is to expand market coverage. I think I mentioned a moment ago that we've got a modest level of coverage and the opportunity is actually to meet the customers where they are today, with a priority to focus on the AI natives that are out there to win as many of them as possible. But we're going to be building an AI builder community. We already have a substantial one around the company, but our vision is to bring 1 million builders into our community in the next year. This is a great opportunity for us not only to gain access to AI start-ups, but it's also a way into the entire AI ecosystem. So vendors and solution -- pardon me, integrators and software vendors as well as enterprises. I've seen this play out in prior lives, and it's a phenomenal flywheel.

Speaking about enterprises, the second area of investment is enterprise readiness. As I mentioned earlier that we think the enterprise market is going to kick into gear in the coming years. And we're going to be set up to make sure that we can arm the enterprise sales team that we're hiring now with the tech readiness. So we're building already the table stakes functionality from a security and compliance standpoint as well as arming them with the playbooks that are vertically oriented to support them against workloads that we know that are there today.

The third area of investment is up-leveling our positioning. The team has done a fantastic job of entering the market as disruptors. But based on the success that we've had to date and with our sights set on the enterprise, we're going to reposition against the broader opportunity in the market. And you're going to be seeing things this year that support that.

And then fourth is to be able to make sure that we are set up to sell the entire platform. So making sure that the teams are supporting execution against a broader solution-based sell. So being able to align to our customers' near- and long-term goals and sell the entire software stack.

Really helpful. With regard to CapEx, there's a $16 billion to $20 billion guide for this year. Could you -- you have talked about sort of the sources of funding that CapEx. Could we review? And I'm also wondering if your comments before about securing upfront payments, if that changes the equation at all?

It certainly helps. I mean, upfront payment is not only a way to ensure that we have the capital to be able to support the deployment of capacity, it also aligns us with the customers' longer-term requirements. So we're now having a conversation and a commercial arrangement that goes beyond today's transactional requirements.

In terms of the balance of CapEx requirements, we have a number of different ways that we're pursuing support for the balance of capital that we need this year. I think we shared on the last earnings call that 60% of it is already supported by existing opportunities.

Okay. That's helpful. And have you broken out that CapEx for this year as far as like how much is self-build, building future sites versus going straight to GPUs that will be generating revenue near term? Any frameworks for thinking about the mix of CapEx?

As we said, we have our own sources of capital, which we deployed already. We have this big contracts, which help us to finance the rest of the build. And finally, we still own all of our revenues, which is an important difference since we are not paying back to the banks since we own the stack and all the margins are ours, revenues which we generate is also a source of our financing.

So I don't know, maybe...

We do know what the requirements are. So the -- and the allocation between a data center implementation and then the build-out is a lot -- just about 20% of the total capital requirements are for the building of the data center and about 80% of the capital requirements are for filling it with GPUs. And we have a series of projects in order to be able to support the capital requirements that we have this year.

Excellent. You've recently made an acquisition, I think, Tavily. What are you seeing in agentic AI? And how are you positioning for that? And what should we expect from an M&A perspective going forward?

Well, Tavily is a specific domain in agentic search. Agentic search is a very demanding function. As you know, up to 20% of all the requests to the models to LLM actually need checking the results on live web-based, mostly data. And they all utilize some kind of web search, and you need a global index to do it. Until last summer, the major global indexes were available through APIs to many players. It was Google's API, it was Bing's API. But then eventually, they were closed or almost closed. All the big players started developing their own functionality in search. They build their own indexes.

But the rest of the market, the longer tail, they don't have this access. And having this on our platform to give them -- make them competitive to the big guys in the industry. That's why we want to have this functionality on our platform. It's one of many which we have, but this is very important for today. That's why the Tavily acquisition, but it's not only that. We had some experience with building global indexes before. So we hope we know what we're doing here.

Excellent. We're out of time. Arkady and Marc, thank you so much for the conversation.

Thank you. Appreciate it.

Thanks, Josh.

Welcome to Nebius Group's Q4 2025 Earnings Conference Call. The presentation will be followed by a Q&A session. [Operator Instructions] I will now hand over to Neil Doshi, VP Head of Investor Relations, to start the call.

Thank you, and welcome to Nebius Group's Fourth Quarter 2025 Earnings Conference Call. My name is Neil Doshi, Vice President of Investor Relations. Joining me today are Arkady Volozh, Founder and CEO; and our broader management team.

Our remarks today will include forward-looking statements, which are based on assumptions as of today. Actual results may differ materially as a result of various factors, including those set forth in today's earnings press release and in our annual report on Form 20-F filed with the SEC. We undertake no obligation to update any forward-looking statements.

During this call, we will present both GAAP and certain non-GAAP financial measures. A reconciliation of GAAP to non-GAAP measures is included in today's earnings press release. The earnings press release and shareholder letter are available on our website at nebius.com.

And now I'd like to turn the call over to Arkady.

Thanks, Neil. And thanks to everyone for joining this call. Well, 2025 was a very strong year for us. Our team did an outstanding job scaling capacity and delivering it to customers with speed and reliability.

We also made great progress in developing of our own hyperscale AI cloud. But before going into more detail around our 2025 results, I want to take a step back.

It's hard to believe we only launched this company 1.5 years ago. At that time, we have built the foundation of a global AI cloud business growing at exponential scale. And we have been operating with very high intensity, building and scaling at extraordinary speed.

Today, we are already 1 of the world's leading and most reliable AI cloud compute providers. We have attracted a diverse pool of clients, who value the quality, performance and flexibility of the platform we built from the ground up. And we have an amazing team who every day seem to do the impossible. And I know they will continue to do this to do so. With this foundation, we are proud of what we have achieved and confident about what we will be delivering.

Now back to 2025. As I said, it was an excellent year. We exceeded our financial targets, and we significantly exceeded our capacity plans, setting the stage for the next phase of scale. Demand remains robust, and our pipeline continues to grow substantially. We sold out of capacity in Q3 and Q4 last year, and we're already now of 2026 also sold out. Even before we bring capacity online, it's often sold out. As a result, the average contract duration of new cloud customers grew by 50%, and the prices of GPUs didn't fall even on previous generations of GPUs as the industry may have expected.

Customers are demanding more compute and increasingly sophisticated solutions to run their AI workloads. AI start-ups are quickly evolving to real enterprise scale customers with revenues from their products resonating with real customers. The demand quickly grows from hundreds of GPUs to tens of thousands. This is already a range the market saw from the biggest customers last year. We are seeing the same trends in the enterprise who are utilizing AI for more and more vital business processes.

Meanwhile, access to compute remains constrained. This imbalance creates a favorable environment for us as we secure and deploy more capacity into a robust demand environment. To capture this opportunity, we are accelerating our capacity plans. Just today, we announced 9 new data centers across the globe. Last quarter, we said we would secure a 25 gigawatts of contracted power by the end of the year. We are already at more than 2 gigawatts now in February, which is why we are raising our forecast for 2026 to more than 3 gigawatt. And we are well on track to deliver 800 to 1 gigawatt of those as available data center capacity just like as we spoke about last quarter.

Capacity is 1 of 2 dimensions of our growth. The other is product. Our main strategic focus is on -- is to scale our poor AI cloud business, which is our multi-tenant air cloud. We will expand our platform both organically and through targeted acquisitions that can enhance and accelerate the development of this platform. We recently launched Tokin Factory, and this is an example of a tenth testament, actually, to all in-house capabilities. And we're also very excited about our recent acquisition of Sabine, which adds genetic search capabilities to our customers and also brings almost 700,000 developers to our ecosystem.

Based on the strength in demand, and our strong execution, we remain confident in our plans for the year ahead. We are reiterating our annualized run rate revenue of into billion by the end of 2026. When we announced this target 3 months ago, there were questions about our ability to get to this range. Over the last few months, our conviction in this range has become stronger. Why? Because we exceeded the high end of our 2025 ARR guidance and showed more than TRY 1.2 billion. Because we already contracted more than 2 gigawatts of capacity and are on track to exceed 3 gigawatts this year. Because we have already delivered all of our capacity for the metal contract. Because we are on track to deliver the capacity for Microsoft through the course of 2026 exactly as planned. And lastly, the demand for our AI cloud continues to be strong. Pricing is strong. We are seeing more and more customers coming into the platform and committing larger and larger and longer contracts. Everything we build, we sell.

We're in the very early days of 1 of the biggest industrial and technological revolutions in history. That's what we believe. And Nebius is quickly becoming the high cloud provider of choice.

I want to thank again all of our employees for their dedication and for work as well as our shareholders for your trust and support, and we will continue to deliver.

Well, now I would like to hand the call over to our CFO, Dado Alonso. Dado, please.

Thank you, Arkady. Indeed, 2025 really was a fantastic year with great execution and delivery. We exceeded the targets we set, outperforming our ARR guidance and achieved positive EBITDA at the group level. We also raised significant capital to fund our growth during the year.

Now let me provide some color on the results, discuss our financing plans and then I will conclude with 2026 guidance. In Q4, we delivered group revenue of $228 million, representing year-over-year growth of 547%. Revenue grew 56% from Q3 to Q4. Annualized run rate revenue for the core business stood at $1.2 billion at the end of December, exceeding the high end of our Q3 guidance range of $1.1 billion. The results of our core AI cloud business were even more impressive. Revenue grew 830% year-over-year and 63% quarter-over-quarter. This was driven by high utilization strong pricing and strong execution. As Arkady noted, we sold out our capacity once again in Q4 as demand continued to significantly exceed available capacity. Even as we delivered such strong growth, operating leverage and spending discipline enabled us to achieve notable progress on the bottom line. Group adjusted EBITDA inflected positively in Q4 and consistent with our guidance, driven by the strength in our core cloud business, where adjusted EBITDA margin expanded from 19% in Q3 to 24%.

Turning to the balance sheet. We ended the year with $3.7 billion in cash and cash equivalents and generated $834 million in operating cash flow in Q4, which was primarily comprised of our strong payments from our long-term agreements. These early cash flows will continue over the course of the year as we execute against these commitments, providing us with significant visibility into future cash flow. Our cash on hand, projected operating cash flow and strong balance sheet position us very well to fund our capacity build-out plans for 2026. I will share more detail about our capital plans in just a moment.

Now I will turn to 2026 guidance. As Alkali mentioned, we remain confident in our ability to generate annualized run rate revenue of between $7 billion and $9 billion by the end of 2026. For the full year, we expect to achieve between $3 billion and $3.4 billion in revenue. Starting in Q4 -- Q2, we expect to begin bringing online some of the new sites we announced today, with the majority of the planned capacity to be deployed in the second half of the year. As of early February, metals capacity was fully deployed, and we are now fully servicing their contract. After delivering the first tranche of our Microsoft commitment on time we expect to continue to deliver the remaining tranches throughout the year, with the majority expected in the second half. We expect Microsoft to begin contributing to revenue at the full annual run rate in 2027 once we have deployed all of the stages.

On adjusted EBITDA, we expect group adjusted EBITDA margin to be approximately 40% for 2026. We expect EBIT to remain at the launch in 2026 and as we progress against our capacity expansion plans, deploy GPUs and invest in R&D that will significantly enhance our technology stack and our future AA product. We believe the return on these investments are attractive and will remain committed and remain committed to our medium-term EBIT target of 20% to 30% with the potential to go higher.

Starting in Q1 2026, we are updating our depreciation schedule from 4 years to 5 years to reflect what we are seeing in the market and in our current utilization commitments. This approach is aligned with accounting-based practices, and we continue to be conservative on this front.

Now turning to CapEx. In order to capture the large and growing opportunity that we see for the future, we plan to invest in CapEx in the range of $16 billion to $20 billion in 2026. We already have about 60% of the capital needed for this range from our balance sheet, existing operations and commitments. We evaluate several funding options available to us on a consistent basis and will deploy 2 guardrails when we look at capital alternatives. First, we will focus on raising debt relative to our business needs and will be prudent with respect to the cost of capital. Second, we will be mindful about the shareholder dilution if we choose to issue equity.

Given our balance sheet and minimal debt, we are fortunate to have many additional options to finance our growing business. For example, we are currently exploring adding corporate debt and asset-backed financing to a balance sheet. Our at-the-market equity program, which we have not used the door to date remains an additional alternative for opportunistic capital. In addition, our equity stakes in ClickHouse and other businesses such as AVride can also be future sources of capital. We are excited about these holdings, especially as the market recognizes their significant value. As an example, it was recently reported that ClickHouse valuation was approximately $15 billion in the most recent funding round. So the wide array of funding options available to us allows us to fund growth in a way that is balanced, disciplined and aligned with returns that rather than committing to a signal path.

In conclusion, 2025 was a year of strategic progress for Nebius and we executed with focus and discipline across our business, generating strong momentum as we enter 2026. In the year ahead, we will continue to scale rapidly to capture the meaningful market opportunity in this once in a generation moment in our space.

And with that, I will turn the call back over to Arkady.

Before we turn the call over to Q&A, I want to provide 1 more update. After this earnings call, Neil Doshi, our VP of Investor Relations, is moving into a new strategy function role, and we would like to thank you for all his great work in building out of our IR function to date. And I'm also extremely pleased to welcome [indiscernible], who will be our new VP of Investor Relations. [indiscernible] brings with her deep research and strategic finance experience from Goldman Sachs, UBS and [indiscernible]. And we're very excited to her from [indiscernible].

With that, let's go to the Q&A.

Right. We are just collecting some questions from the portal, and we will begin Q&A in just a moment right. The first question from our investors on the portal as Nebius is moving quickly to both bring demand online and to build strong foundation for the future capacity. What are you seeing in the market that gives you conviction that the demand for AI will continue to justify these investments. Let's get back to Arkady.

Well, First of all, we all look around and see what's going on, practically everywhere, how we change our habits and everyday primate lives, what happens in our corporations. In our company, for example, how much of workloads are now utilizing AI capabilities to quoting. And actually, we now see all the industries actually changing. You look at a recent example, quoting or look at the move industry, look at the research whatever you do now, you do it with AI. But again, these are just general notions, but our condition is mostly based on what we're seeing directly in our business. We see very signals in all sectors of our business. First of all, in the large accounts. All large clients are talking to us and not fully to ask the whole market about renting more capacity and more GPUs, because their businesses are growing. There is the ongoing [indiscernible] actually lead will lead to more contracts everywhere. But we are, as you know, much more focused on our AI cloud business. And in AI cloud, we have these 2 major sectors AI start-ups and enterprise and look at native customers, what happens to them in 2025. Some start-ups appeared, but some of them are becoming the real companies real enterprises of the future. And they started getting traction. The products became more and more the more and were used by their customers, and they're scaling quickly scaling with real revenues, real demand. And what we see such customers, such companies who used to order hundreds of GPUs, tons of GPUs, now the order in tenths of GPUs. And this is actually -- is the magnitude of what we saw in the largest consumers last year. Plenty of models. We're ordering tens of thousands of GPUs just a year or 2 years ago, now it's yesterday's start-ups at this level. Again, this is real business. These are real customers of them paying real revenues. So the traction is visible. I can -- there's a lot of customers from the sector, starting with [indiscernible] molecular different sectors, and they all have the infrastructure and becoming as we speak, becoming real companies, actually, the future enterprises. And on the other hand, the enterprise clients who involve more -- we're actually switching most of the everyday business process to AI and generate new profits through that AI implementations. We see the growing number of such customers on contracts from each of such a customer, the number of GPUs is growing. The duration of the contract is growing, as I said, the average generation for new customers grew 50% last year. So from the signals we see from all the sectors of the market from big clients, hyperscale labs from AI start-ups becoming enterprise clients for enterprise going everywhere we see positive signals, and we just need to build more for them or data centers more tools. And if we could build faster, and even more than we have today, we will do it. So we are building because it's clear growth.

Great. Thank you, Arkady. The next question is from Josh Baer from Morgan Stanley, who asks about our CapEx financing plans, especially now that we have to -- $16 billion to $20 billion of CapEx guidance out there. how are we thinking through to meet these expectations for CapEx? Ophir?

I am Ophir Nave, the COO and Board member of Nebius. Usually I'm not participating in these calls but given the importance of this question, both for our business and well to the investor community that will take this one. Basically, this question has 2 parts. First, what is the optimal capital expenditure for our business in 2026? And second, how is how this CapEx is going to be financed. And obviously, these questions are connected related.

So let us to take to your second question. On the second part, how are we going to finance the CapEx. So obviously, we will first finance it from our cash flows. Our cash on hand, we have cash iterative generated from our core business. But most importantly, I would say, we have a significant amount of cash that we received and will continue to receive in 2026 from the favorable terms of our long-term contracts. We are talking about significant more difficult. This cash flow -- we actually finance the majority actually around 60%, maybe even more of all of our CapEx needs in 2026.

So how we're going to finance the remainder [indiscernible], the rest? So we all know we have a very healthy balance sheet. By the way, not by coincidence because we are working very hard. We have a very -- wouldn't -- if plant healthy balance sheet. As of today, we don't have any corporate a little bit. We don't have any asset-backed finance center, even though we have multibillion dollar revenues from long-term contracts. We don't have any bank work. And we didn't have to date by choice.

But moving into 2026. And as we all know, an occupant capital structure in our business should include also that this should obviously be changed. And we plan during 2026 to use some of the tools that I mentioned in order to move towards a more optimal capital strategy.

So having said all that, from the financing point of view, we believe that the $16 billion to $20 billion CapEx and make to offsets. And we will be able to finance it while keeping a very healthy disciplined balance sheet.

But moving to the first part, the first question, what should be our optimal CapEx for 2026. I'd mention -- actually Arkady said it time and again, given that we are fully vertically integrated company. Our CapEx is basically divided into 3 parts, less than 1% due to security power. And given the importance of power to our business and given the importance of secure power for our future hyper grip. We are moving full forth in order to secure as much power, if possible, given relatively low-cost fleet. And I said that on this call, [indiscernible] mentioned that we already secured a significant amount of power, and we will continue to do so.

The second part of the CapEx building the data set of this is approximately 20% of the total CapEx. And we invested in building data centers, we invested this moment. And it turned out as we expected, that it became 1 of the best adjustments that we made. Why because when we are lettered almost ready where we can deploy GPU in a very short period of time. we can see this capacity in a very attractive and favorable country. We have done so to date and we plan to do it also in the future. We see the demand, we see the interest. We have a clear visibility on the demand for the second half of 2026 and for 2027, both from our cloud clients as well from AI Labs and hyperscalers and we are positive in these investments with the interest that we are getting from all these plans will play out again very beneficial for our company. And is it done so in 2025.

The remaining part of the CapEx is to deploy the GPU. Again, this is a significant part of the CapEx, the beauty that we deploy the GPUs in a short time period when we have a great visibility about the demand and about the prices and about margins and about and we had, we don't view it as a very risky to be. So having said all that, we believe that $16 billion to $20 billion CapEx for 2026 makes the most sense for us. First of all, it is based on preferability on the demand for 2026, 2027. It's based on the interest that we are getting from various 12 months in our cloud platform and labs and hyperscalers. It will enable us to meet our $7 billion to $9 billion ARR in 2026. But more importantly, it will also put in the front petition for our hyper growth in 2027 behind that.

But speaking about CapEx and investments I think that it was while also to address additional bonds, one is ATM. As you all know, we launched our ATM program last long term. As of today, we didn't use it in the Actually, we don't have any concrete plans to use it also in the near future. But however, it's another tool in our tool book. that will enable us to use the -- it will make the most sense for our business as well as our shareholders. And this is another tool that will enable us to keep a prudent, balanced disciplined in balance sheet. And another part that I think worth mentioning is obviously a non-call businesses. We are fortunate enough to have over 25% stake in click out and our ownership in every right. This noncore businesses, this takes a lot many billions of dollars as of. This is great, but it's less interesting by itself. And most importantly, we truly believe that this takes will significantly increase in the midterm. And when they do they will enable us to continue growing our business in a higher based moment as we plan 2027 and behind, while using this capital and continue keeping a very disciplined balance sheet. We're really happy with this potential capital injection for the future. [indiscernible] right and overview of the CapEx, the needs and the ways to finance.

That was very helpful, Ophir. All right. Next question comes from Alex Platt, DA Davidson. Can you help us bridge to not only the 800 megawatts to 1 gigawatt of connected power guidance, but now to the 3 gigawatts of contracted power guidance by year-end. Andrey?

Yes. Thanks, Neil. Hello everyone. So what we can see now is we are accelerating the build-out and deployment of the capacity in 2026 greatly. And we expect that the acceleration will continue in 2027 and beyond that. So what we are doing is we are, as Ophir mentioned, we are doing the investments in building the foundation for the 2027 and for the other years. We are well on track of achieving our goals that we mentioned about 2026 of this 800 to 1 gigawatt goal around year end. And we do that by launching a lot of sites with a mix of smaller and larger projects. And some of our bigger projects starting to come online in the end of this year. And in addition to that -- in addition to the size that we mentioned today, we have a multiple going projects. And we expect that some of these additional sizes may materialize this year. So basically, we expect that the collections will help us grow faster certainly than Q2 this year and ramping throughout the year. And our own projects, which are substantial in size will really start to ramp up in 2027 and beyond. And that relates directly to the contracted power.

Great. Thank you, Andrey. Andrey, maybe just staying with you, Andrew Beale from Arete is asking about an update on the New Jersey data center site.

Yes. With the New Jersey, we are very pleased with the progress there. We delivered the first tranche to the Microsoft on time, and we are well on track to be the remaining commitments on time as well. We believe that our partners secured the components in the supply chain, and they're working extremely hard to get everything online. Well, we also have a some safety margin buffer times in our projects. And while all projects can have some flotation, but we have built our plans with a large margin of safety and we have a high confidence of execution this.

Great. Thanks, Andrey. All right. It looks like we have a question from Alex Duval from Goldman Sachs. [indiscernible] came in light on revenue versus consensus for Q4, but ARR was ahead of expectations. So which is the more meaningful metric. And are there any timing considerations. And then just more broadly, lumping this with the ARR question. Can you help us understand the difference between ARR guidance of $7 billion to $9 billion and the revenue guide for 2026? So Dado, maybe you can take this.

Thanks. Let me take it 1 by 1. On the first question, of course, we were very pleased with our ARR of $1.2 billion, which exceeded our ARR guide. Our revenue came in the middle of our guidance, which actually was what we anticipated. Look, as we are in a hyper-growth phase, ARR is the metal metric. Now on the next question around the difference between ARR and revenue guidance. Let me say that first and foremost, at Nebius, we are freely taking a prudent approach to our revenue guidance as we communicated. In 2026, our guidance is $3 billion to $3.4 billion in revenue. The difference between ARR and revenue is logical, our revenues and ARR reflect the deployment schedule of our capacity throughout the year, with the majority of this capacity being installed in the second half of the year, as Andrey mentioned. We need also to consider that our largest enterprise partnerships are also still ramping. So the revenue guide simply reflects the ramp-up in capacity coming online.

Great. -- thank you, Dado. Question from Alex Platt from D.A. Davidson. How should we think about your progress against the billion of ARR guide. And are you really dependent on the large hyperscalers to get to that range. Marc, why don't you take this?

Thank you, Alex. First of all, let me clarify our 2026 ARR target is not dependent on any new mega deals. As we bring on our planned additional capacity, combined with the already strong pipeline and our go-to-market plans, we are very confident in our ability to deliver our '26 ARR target. Our continued success with AI natives and the early progress we've seen with ISVs and enterprises set the foundation for capturing the market this year. Based on the traction we are experiencing and our extensive research on our total market opportunity, we are leaning into the verticals we've already laid out, health care life sciences, media and entertainment, physical AI and retail, which gives us ample runway to capture share and grow our business. While we are happy to service large strategic customers like hyperscalers. We will remain opportunistic with such large deals as we look to balance the opportunity with the long-term positioning we plan to achieve with our AI cloud.

Great. Thank you, Marc. I'll just take a question from the portal. Can you provide an update on where you stand as it pertains to the delivery schedule with Microsoft and Meta and remind us again about how those 2 contracts layer in throughout the year. Maybe on the first part, we'll have Andrey and then Dado, you can take the second part on the contract layering.

Yes. Thanks, Neil. As I mentioned earlier, the first tranche to Microsoft was delivered according to the plan in November and the remaining capacity is on schedule ongoing. All of the remaining tranches will be delivered through the whole 2026 and more than half of them will land during the second half of the year. about me earlier in this month, we delivered both contracted tranches to me on time, and we're just now fully in service and stage. And Dado, can you comment on the financials?

Of course. Thank you, Andrey. The deployments in Meta, as Andrey just mentioned, went live early February. As such, we expect to recognize 12 months of revenue for the first tranche and roughly 11 months for our second tranche. As for the Microsoft deal, we expect revenue to ramp over the course of the year in line with our plans to deliver the capacity tranches, which as Andre mentioned also, will happen throughout the year with the majority expected in the second half. So Microsoft deal will begin to deliver a full year revenue starting in 2027. And as we execute on these commitments, we expect them to contribute positively to our medium-term EBIT margin target of 20% to 30%.

Great. Thanks, Dado. Another question from the portal. What drove the upside in the December 2025 ARR? And as you look into Q1, what are the demand trends that you're seeing in the market today? Marc, why don't you take this?

Certainly. The upside in December ARR came from solid execution and strong pricing and utilization. We continue to make great progress in adding new logos and expanding with existing customers. On the second point, we are seeing very strong pricing across all families of GPUs, and we are at full utilization as we continue to sell out all available capacity. On the -- on Q1 demand, it remains extremely robust, and we are seeing the same trends that we shared in 2H 25 carrying into '26. The 3 things that give me tremendous confidence and excitement as we enter '26 are continued pipeline growth, positive deal trends and the progress we're making with our vertical strategy. The pipeline creation trajectory in Q1 is on track to exceed $4 billion. And as we expand our available capacity and add sales coverage, we expect it to continue to increase. In terms of deal trends, they're all moving to the right direction. Deal terms are getting longer and average deal sizes are increasing. In Q4, we saw nearly twice as many transactions completed for over 12 months in duration over what we succeeded with in Q3, while average selling prices increased by more than 50%. As an example, we are sold out of hoppers and those that are coming up for renewal often off of short reserve agreements are getting renewed at 12 months or longer, while we're actually seeing pricing nudging up.

Lastly, we're focusing on customers with premium workloads and use cases, which is resulting in superior terms, including increasing those who are willing to prepay for securing future capacity.

Great. Thanks, Marc. We had another question from the portal. There are headlines of data center equipment shortages in the market. How is Nebius handling the situation, ensuring access to these products? And do you expect it to have any impact on your deployment? I think, Andrey, this would be for you.

Yes. Thanks, Neil. On the sensor delays, while generally building the data center so the is a quite complex task and no 1 can get away from all the risks. But I think we are in a quite good shape of managing this risk. As you saw today, we are well ahead of our contracted power and we are developing 9 contracted sites we announced today. And the main idea, the main strategy is to have the portfolio of sites. So we are not dependent on any specific single data center project to achieve our guidance and deliver our plan. And it's pretty important to understand our differentiation will push that cloud, allowing us to enjoy the flexibility that it provides that we are not dependent on any size, we can move the loans in between, we can provide it from different locations. We just operating to ensure that we have enough capacity. And the second 1 is we already contracted the majority of our long lead items around our own sites to be sure that to ensure the capacity deployment beyond 2026 as well. And the second part of the shortage is the memory storage. The first important point is our largest deals with Microsoft and Meta, we were able to secure the necessary components last year for the full scope of those contracts. And we secured it in second half of 25 before any price increase. And for the remaining, we are confident in our supply chain to get the parts we need to deliver the capacity.

Andrey, maybe sticking with you, a few of our analysts are asking with the announced to 9 new sites for data centers that are going to be a mix of owned and co-locations. How do you evaluate when you want to buy versus lease and how do you expect this mix to shift longer term?

I think we discussed it quite a few times. So generally, we are focused on bringing most of our largest projects as a sole the oil projects. There are 3 main reasons. First, we get much were the total gold ownership with this we have much more control over the execution of the project because we have expertise and experience of building our own site. We also can achieve greater efficiency at scale because generally, we tailor the design specifically for what we need and for our technical requirements, but this takes time. And well, we use leases and partnerships to fill the gaps before we are at full speed with our own one. So just to sum up the preferences, to develop the infrastructure ourselves. But again, we still need the partners and leave us occasionally to support the growth.

Great. Next question from Josh Baer at Morgan Stanley. It's really around the software stack. What are the most common software tools that your broader customer base startups and large enterprises are using and paying for from your AI cloud. And what proportion of your customers utilize your software and services in addition to your compute capacity. And then any color on just how much ARR comes from the software stack today? Marc, why don't you take this?

Certainly. Thanks, Josh, for the question. As I think you know, our AI cloud is purpose built for AI and is battle tested with our AI customer base. At this stage, 100% of our AI cloud customers are utilizing our AI cloud software, obviously. So we have 100% attach rate. We're very excited, by the way, about the new products that we've launched, like Token Factory and the Aether releases, which have opened up TAM and give us an opportunity to expand into enterprise. Our new acquisition of TAL also extends our platform capabilities by providing a genic search for AI developers, creating more routes into accounts. I should also mention that we continue to see demand for embedded storage from customers across key verticals, including physical AI, media and entertainment and health care and life sciences. As a result, we are creating solutions that are vertically specific to meet the demands of these customers. We're really in the early stages of our monetization journey. But our software and services make our platform sticky, which enables us to charge more on a relative per GPU hour. We are exploring other monetization models, including consumption based, such as per token pricing with Token Factory. So Josh, Stay tuned for more monetization in the future.

Great. Thanks, Marc. Let's -- another question from Alex Duval from Goldman Sachs. It's really on the 40% EBITDA margin target that we gave. What gives us confidence in that. Dado, can you take this one?

So thanks for the question, Alex. Let me bring the bigger picture. Nebius is scaling and as we do so will our margins. Now let's come to adjusted EBITDA margin guidance. In Q4, the group achieved adjusted EBITDA margin of 7% and for 2026, we are expecting to reach 40%. We see a tremendous demand for our AI cloud business, and we are investing appropriately to serve this demand. As we reported, well, the margin of the core AI cloud business is actually significantly higher than the group. And as we scale the AEL business, most of our revenue and margin will continue to be driven by the core AI business. So we expect, of course, the other businesses to still operate at EBITDA loss in 2026, but their contribution to EBITDA will be smaller and smaller. So the guidance of 40% adjusted EBITDA margin reflects the current expansion state of the business.

Great. Thank you, Dado. We have a few of our analysts, including James Kisner from Water Tower Research asking about [indiscernible] Table. is the strategic rationale behind [indiscernible]. How is this complementary to your existing offering? And really, is there a lot of demand for this product from your cloud customers. Roman, why don't you take this?

Yes. Thank you for the question. I was afraid we will not come to it. It's quite exciting event for us, and we're super excited to announce our first deal with the acquisition of Tavile and welcome Tavili team and Tavili customers in Nebius family. Tavili is agentic search company. They connect AI agents to the web. And it's very much fits in our strategy to become and to be the platform where all the AI developers from start-ups and enterprises building their AI applications and engines. Tavili got quite a significant progress. They already serve many Fortune 500 customers, and they've got great adoption in developer community and loved by developers. We will continue to evaluate for potential acquisitions of the companies that can deepen customer engagement, stickiness, and increase our lifetime value and strengthening our positions as a full stack AI cloud provider and this game is such large that we obviously will not build everything ourselves. We will be in the strategic opportunities and we'll be in the strategic partnerships moving forward, looking for the companies and partners that got a great product, great developer experience, and similar to our DNA to build a platform that will be loved by AI developers long term.

Great. Thank you, Roman. We have another question just on capital allocation and M&A. In terms of like build versus buy, like Nehal Chokshi is asking this from Northland. Arkady, when you're thinking about capital allocation, especially whether it's investing in capacity or doing M&A, how do you think about this?

Again, what we're building here, we are building 1 of the largest platforms for AI developers to build their applications. Largest the largest platform means 2 dimensions: the scale and functionality product. and we allocate capital between these 2 dimensions. We need to build scale. That's why we need to build all these data centers and buy all those hundreds of sound GPUs. And we need to build the product and we build this product both organically, internally by our own developers, but we don't not cover everything. So we go into acquisitions. And we spent some capital to acquisitions as well to enhance our product, to get more talent and ultimately to develop more even past the delays perfect example of such tons. Hopefully, not the last one.

Great. Thank you, Arkady. And I think we're at the top of the hour. So thank you, everyone, for joining our fourth quarter 2025 earnings call. And next quarter, Gilly will be leading the earnings process. Thank you, everyone.

I'm super excited about this session. As you all know, the emerging cloud infrastructure players are in the spotlight these days. They're growing through some of the most amazing growth trajectories in all of tech. I'm sure many of you were frankly unaware of Nebius 2 years ago. Now they have a $25 billion market cap. Like those stories don't happen a ton. And so I've been looking forward to this one because it's -- well, Neil and I have had a few occasions to chat. I haven't had a chance to meet Marc yet. So Neil, nice to see you. Marc, lovely to meet you.

Pleasure is mine, Karl. It's great to be here.

Thanks, Karl.

Just because this is a newest story for people, you don't need to, Marc, go into the origin story. By the way, for those of you that want to go deep on Nebius, Marc just did a podcast, which I was just joking, I still haven't had time to listen to, but it's a very good hour plus pod running through the Nebius story. But I guess what I wanted to focus on just as part of the origin story is, obviously, the execs, a lot of them engineers came out of Yandex in Russia. What was it about that experience that enabled you to scale Nebius? I'm sure the answer is just partly that you had a lot of amazing software engineers who knew how to scale a business.

100%. It's not the normal start-up story, as you pointed out, Karl. The reality is that Yandex was already a scale and successful Internet company that was operating some of the largest infrastructure in the market, actually the biggest NVIDIA customer in Europe at the time. And when Arkady Volozh, our CEO, who was also the founder of Yandex, started Nebius, he brought with him 1,000 engineers. So software and hardware engineers that had been working together for, in some cases, more than a decade.

So while the company is a 2-year-old or less actually company, the reality is it's a team with deep longer-term experience and that they had built scale infrastructure of the type that we're building today in order to be able to service some of the most important workloads. So we're starting with a bit of a head start compared to a lot of other companies that are in the category given the resource depth that we have in the team and the expertise that they bring.

Marc, do you want to hit on one essential question I get, which is what is the enduring strength of Nebius? Because there may be a misconception that some of the emerging cloud infrastructure vendors are scaling like they are because we're in a compute-constrained environment. But a lot of these investors are looking out 3 to 5 years to a point that will inevitably come when those compute constraints aren't as severe. And at that point in time, Nebius and your peers will have to sink or swim based on your core strength. So what are those?

So like I mentioned, we have this wonderful extremely experienced team of engineers that span hardware and software, okay? These are individuals that have built scale infrastructure in the past. An important part of our offering is delivering the software that AI engineers are going to need. And this is AI engineers of all types and sizes or from all types and sizes of organizations.

So being able to service a start-up that's coming in self-service that wants to put in a credit card and get access to a small cluster, be able to burst up to their needs all the way up to a large enterprise that's looking for tens of thousands of GPUs to be able to service their requirements. That software gives them the capabilities, the tools, the workload management necessary for them to successfully build on our platform.

Second, it's scale. Scale is critically important in order to be able to deliver against the increasing requirements that are on the market. And today, we're one of very few gigawatt scale organizations. And that's pretty significant when you look at bare metal. It's even more significant when you recognize that there are even fewer scale organizations that are delivering cloud infrastructure.

We're not building a GPU-as-a-service platform, we're building a hyperscaler. And that requires a combination of scale infrastructure and scale CapEx necessary to continue to fuel the expansion of that infrastructure.

And then thirdly, we are highly focused on engineering at every layer of the infrastructure, okay? We build from scratch. We can literally start with brown dirt, greenfield projects and be able to put in place the data center, the power, custom cooling, custom racks, we spec out our own servers.

Our team is looking for every opportunity for cost optimization. We're looking for ways to continually improve our efficiency in the way that we're delivering and operating so that we can drive up our margin.

And then lastly, and this is one that's a little bit hidden. And since I joined the company, I've been able to watch it in action, it's pretty extraordinary. We have a support organization that's made up of AI engineers. So the humans that are actually taking the inquiries from customers and dealing with the technical escalations, they are just like the humans on the other side. I hear this all the time from customers. They're very impressed with the skills and capabilities of the engineers that are responding, and they're able to get 24/7 coverage of their needs. So we're building not just a hardware layer that can be sold to customers bare metal, we're building the AI equivalent of a hyperscaler.

So let's dig in on that. And in particular, the full stack vision. So if Nebius's dream is not just to be a great per hour GPU rental infrastructure provider, what does that full stack look like, Marc? How high up are you going to go? Is there ambitions to get like the big 3 hyperscalers are deep into the database layer? Like what's full stack mean to you?

So today, what our stack looks like is the fabric that is stretched across all of the hardware that is providing the ability to manage workloads. So think of this as virtualization and manage Kubernetes. On top of that is an administrative layer, a control plane that gives the AI engineer the ability to manage their security, the operational characteristics and ensure that they're getting the performance they're looking for.

We're constantly adding new functionality to the platform. We're getting the guidance on that new functionality all the time from customers. And the way that we're thinking about that new functionality is as one contiguous vertically integrated platform. So you can go from training to optimization, to inferencing without having to do any extra steps. You basically have the tooling at your fingertips.

Our vision is to continually specialize our platform. So looking to the special needs that, as an example, our customers in health care, life sciences have, which are different than, say, our customers in physical AI. They're delivering very different requirements to be able to service the end requirements they have. And we're trying to put at the fingertips of the AI engineer the tooling necessary.

So in the fullness of time, our expectation is that we will build out an entire set of functionality that services all the data, development, security and operational requirements that a customer has and where we don't feel like it's a top 3 need for us to own it, we will be partnering with ecosystem suppliers so that there's a ready available ecosystem of already integrated capabilities so that the AI engineer doesn't need to leave our platform and instead can actually meet their full needs. So there is no ceiling per se, especially this early in the game. We're looking to all opportunities to ensure that we can service the full requirement the AI engineer has.

Okay. So let's flip from the stack to what the customer base will look like 3 to 5 years out. I think everybody is well aware that for younger companies entering this business that's very capital-intensive, scale-based, you often have to start out with one large customer, scale up with them and then use that as a beachhead to diversify. So we will get to that large customer in a quick second.

But I want to start with what the 5-year out vision looks like, where you've got a number of customers, loosely, maybe you look at it a different way, Marc, but you've got the big model providers. You've got AI natives that are very compute hungry. And then you've got traditional enterprises like UBS. When you think about what your plan is for 5 years out, what do you want that customer base to look like? Where do you think the bulk of compute spending will come from?

Well, first off, our intention is to build a platform that can be consumed by all segments. We're going to pursue the 4 corners of opportunity in the market, okay? It's very, very difficult to say with certainty that this movie is the same one we saw the last go around with CPUs, although there's a lot to learn from that version of experience. And we're taking as much as we can from that experience to deliver on our market development.

And if you look back at the experiences of AWS and GCP, early going was all start-ups and the inevitable enterprise opportunity came much later. And in some respects, they're still making their way through the enterprise opportunity. And if you look at the addressable market, and if anybody has a strong, well vetted, detailed data-driven model, I'd love to talk to you because we're looking for that. But right now, as we do the basic analysis, 2/3 of the opportunity is likely to be in the enterprise.

And the challenge that we're facing is that it's still really early going. And I'm not surprised. I mean, UBS, as an example, is a conservative organization and other companies like UBS would -- there's no upside for taking too much risk. But organizations like UBS are actually doing very specific things with AI, as an example, high-frequency trading or model-driven trading.

And we're seeing those green shoots, those very early use cases in key vertical markets, and we're pursuing those with very focused attention to be able to get those footholds in organizations. And then as they scale and we have the opportunity to pan out, my confidence it's likely that 2/3 of the opportunity 3 to 5 years out is going to be servicing scale enterprises that are looking to ensure their relevance into the future. And it's to be seen, but if that doesn't occur, we're servicing the rest of the market with the same infrastructure.

Okay. Let's now talk about that one anchor customer, where, obviously, Microsoft signed an enormous deal with Nebius. I think with options, it's $19 billion over, obviously, multiple years. Can you -- you're probably somewhat limited in what you can say, but can you say anything about duration? What exactly Microsoft is using Nebius for? Anything that might help the audience understand the nature of that relationship?

It's a 5-year agreement. It's a very scaled single instance, massive cluster that they can be utilizing for any aspect of their business and their demands and needs are extreme. If you think about it for a moment, they've got an established business around Copilot, which, by the way, is doing extremely well. We're a Copilot customer. We have it activated in our environment. People are utilizing it all the time to enhance their experience. And you can see some latency in its performance. They need to put more compute behind it because they've done a good job of selling Copilot to the market.

They've also got products that they're building, models that they're creating. They've publicly said they want to own their own models, and they want to diversify their capabilities. So they're investing in model capabilities. And then they have the Azure business, which competes with us in being able to deliver compute in order to service end customers. So we could be across all of their business. Our confidence is it's probably in the former categories mostly.

Okay. Got it. I want to go into a little detail around Nebius's effort to scale up that infrastructure. But before I do, I want to hit one last thing more on the demand environment. When we listen to you, we listen to CoreWeave, we listen to Lambda, even when we listen to Microsoft, the -- what they're all expressing is that this year, it's hard to pinpoint it, but it felt like 6 or 9 months ago, everybody just started talking about a surge in demand for compute. We saw a ton of huge deals, long duration, something inflected. And I don't pretend to have my finger on exactly what happened. So I'd like to ask you, what on earth happened, call it, the spring when it feels like everybody's tone just pivoted, Marc? Where did that come from?

I wish I had a crystal ball. I was not at Nebius 9 months ago. I joined Nebius 6 months ago, correlation, not causality. But what I can tell you is what the experience is. And maybe this is just more of the same, but I do think it's important to share. We are definitely seeing an acceleration in demand. And this is not isolated, in terms of thinking about revenue and pipeline production, we saw our pipeline. We reported this in the third quarter accelerate by 70% quarter-on-quarter. We generated $4 billion worth of pipe in the quarter, couldn't service it all, obviously.

So we are seeing demand beyond our supply to be able to service customers. And we're seeing it in every category. So start-ups that are building the next generation of whatever category they're in. And oddly, interestingly, there's a fair number of start-ups. I've never seen this before, that seed or A stage have $1 billion on their cap table. So they're actually looking to build the next frontier model or the next specialized model. And that's consuming a fair amount of supply in the market.

We're also seeing the scaling AI companies do something interesting. Those that built on the foundational models are also now starting to specialize open source or even build their own models. Michael Trull at Cursor has been very public about the need to expand their capabilities to ensure that they remain relevant and able to deliver the best autonomous coding capabilities in the market.

We're also seeing software vendors move from just like adding minor tooling to rethinking their entire offering from, going from an AI-enabled to an AI-first capability. Some companies are even like looking at a complete restart in the way that they're thinking about being able to remain relevant and compete effectively against the AI native players that are out there.

We're seeing enterprises, as I mentioned a moment ago, very, very focused, but we're also seeing that focus accelerate in terms of their willingness to take up AI as a solution to whatever area of expertise that they're in. So we're seeing an extraordinary expansion in demand. By the way, we're seeing use cases expand, people going from training to optimization to inferencing. We're seeing modalities expanding. What was largely LLM generative AI is now including voice, video, imagery. So with all of this is coming a greater dependence on infrastructure being available like the kind that we supply.

So maybe this is a good segue for one hot question for you, Marc, and that is the equity markets have corrected in the last couple of months and your stock has corrected on a view that we're in an AI bubble that you, CoreWeave, Microsoft, Oracle are massively overbuilding ahead of demand that won't be there to justify that scale of investment. What's your retort to that?

It's going back to what I just said. The demand is there. The demand is there. I mean we're seeing it in our pipeline. We're experiencing it with our customers. We're watching customers doubling every 6 to 8 weeks in some cases. We're watching the uptake take place.

An interesting anecdotal side note is watching what's going on with inferencing, okay? Just to make sure we're all on the same page, you can create a model, you can optimize the model, you put it in production. And then when your customers are consuming, that's inferencing. So for us, that's actually the realization of the commercial delivery of the capabilities the customer has.

Inferencing is growing at an extraordinary rate. As a matter of fact, there's a guy at the Times that is tracking inferencing. Evan O'Donnell has a blog where he's tracking all the different inferencing suppliers and looking at their scaling compared to CapEx investment. And what he's showing is that inferencing is growing 40% faster on a monthly growth basis than CapEx investment is. So I think we're actually still lagging the demand that's out there.

Okay. Your point from that is that actually, there are outstanding returns on that CapEx so far?

That's correct.

Okay. Let's talk about your need to stand up capacity. None of this is going to happen. You're not going to hit your forecast unless you get those gigawatt-plus facilities up. So maybe you could give us an update on go-live progress on getting capacity online and the extent to which you've run into any unexpected supply bottlenecks.

This is a complicated industry. I just want to say that right upfront. If you go look at my LinkedIn, you'll see I'm a software guy. We don't worry about in software, physical delivery. So there's quite a bit of complexity in being able to orchestrate and deliver the capacity that's ultimately being sold by our organization.

And what we do, actually, going back to a statement that I made earlier, the team that we have is extraordinary. They have a lot of experience at being able to look at a greenfield project, being able to diligence it all the way down to the details. They don't take a cookie-cutter approach. The market doesn't actually lend itself well to cookie cutter right now. You need to be prepared to go to unique locations with unique configurations to be able to actually look at whether or not you can actually build and deliver the capacity.

And our team is able to look at those projects and understand with confidence which ones are going to actually make it through to full maturity. And they do that by managing every single variable, controlling every aspect of the project so that they can confidently deliver the capacity that's being promised so that we can ultimately sell it to the market.

Okay. What do you think is the biggest bottleneck right now? It used to be chips, then we heard power, and then CoreWeave's CEO said a couple of weeks back that it's actually not power, it's -- we got enough of that. For the next couple of years, it's actually just literally building the data center shell. What would your contribution to that bottleneck debate be?

Well, it's a reality that the lead times that go from handshake or signed agreement on a given project to it being delivered, there's some physical limitations in terms of how fast you can do that. So the reality is if you're planning your delivery as one sequence of serial events, you're actually potentially delivering at the most extreme and not necessarily the best, most optimized way. So it's managing a series of projects in parallel that have different likely scenarios in terms of their ultimate delivery and fruition.

And being able to manage that span of projects is a special skill. It's not something that -- again, there's 10 companies that have done this before. Agile infrastructure delivery, that's a new thing. And what we're trying to do is apply the same agile methodology that you use for software to the challenges and discipline of delivering the infrastructure.

Got it. There's probably a role for AI in that process.

It certainly is.

Neil, maybe we'll bring you in as well for a couple of questions. I think the other aspect of these bubble concerns that go beyond just will demand be there is will the financing remain there, where it feels like there's been plenty of funds that are very willing to fund these data center build-outs. But the concern from the Street is, will that be the case tomorrow. So perhaps you could comment on, let's say, the health of the financing market right now.

And by the way, for everybody, I'm keenly interested in this subject given that I'm not a financial guy, but it's become a very important ingredient to this subject. So we actually invited 2 of the bigger participants in the financing side, Blue Owl Capital and Magnetar to come to our event, and I'll be up on stage with them this afternoon. But Neil, do you want to take a shot at that?

Yes. It's a great question, Karl. I think one of our strategic advantages has been our ability to raise capital and deploy that capital effectively. So maybe just helping level set what we've done so far. After the divestment in the summer of '24, we received $2.5 billion of cash. A few months later, we raised another $700 million of capital through a pipe transaction with Excel and NVIDIA. That capital allowed us to really start to accelerate and grow our platform aggressively.

Over the course of this year, we've raised another $5.3 billion in the form of converts and an equity follow-on offering. And recently, we put into place an at-the-market program that will just give us future flexibility to raise capital. So all in, we've raised about $8.5 billion of capital to date. That's enabled us to go from one data center location in Finland to multiple locations around the globe.

We will have expanded our connected capacity from 25 megawatts at the end of last year by almost 9x by the end of this year. And it's also enabled us to bring on marquee customers like Microsoft and Meta. So we've been able to deploy that capital and do that all with a net cash -- with a net positive cash position.

As we think about the near to medium term, we'll continue to look at capital markets. We think that there's opportunity to raise money through debt. We're looking at asset-backed financing. We're looking at corporate level debt, and we'll continue to look at equity. I think what's also really interesting is we have these other businesses and equity stakes that we think can yield us multiple billions of dollars in additional funds that we can use to invest back in our core business.

So Avride is our wholly owned autonomous vehicle platform. They just today announced they've rolled out a robotaxi service with Uber in Dallas. We own a large stake in ClickHouse, which is one of the fast-growing AI database companies often compared in the same vein as Databricks or Snowflake. And then we own a significant majority stake in an AI data labeling business called Toloka. Think of it as a mini scale AI, but we recently brought in Jeff Bezos and Bezos Expedition as investors and some other high-profile investors in that.

So we think over time, as these businesses and equity stakes grow in value, we can use these to actually fund our business without having to tap into excessive debt or raise money through equity. So we want to be mindful of shareholder dilution and be really mindful and disciplined on the capital structure side.

Let's also talk for a moment about what the ultimate profitability of this business will be. Marc, you mentioned a little bit ago about the parallels to, call it, the hyperscaler 1.0 era, the CPU era. That era created gross margins, even though Amazon and Microsoft don't disclose them, let's loosely say, mid-60s. It doesn't feel like any of the GPU clouds are on a path to that, but maybe they are. Can you talk a little bit about what the 5-year out margin structure of this business should be?

Yes, absolutely. As much as we're focused on growth, we are equally focused on margin expansion. We are not a growth at any cost company. If you just think about where we were at the end of December last year, we were still at a group level, losing money on an EBITDA basis. The core business turned EBITDA profitable in Q2. In Q3, the core business posted 19% EBITDA margins. And now we're on track for the entire group, all of the businesses, including the core business as a group to be EBITDA breakeven by the end of this year.

Back in April, we also gave our medium-term outlook. We said we expect EBIT margin, so factoring into account depreciation and amortization, to be in that 20% to 30% range, and we are definitely focused on delivering on that. I think there's a few structural reasons why we think we can get there. First of all, there's scale efficiencies, right? We are not just building a GPU-as-a-Service bare metal product, we are building a cloud business on top of that. That cloud enables us to actually drive better customer retention. We can charge more of a premium price for a premium product. And ultimately, we can deliver higher customer LTV.

We also, Marc talked about, having 1,000 engineers. That was a huge benefit for us, right? That actually leads us with operating leverage. So as we scale our ARR by 7 to 9x in 2026, we don't necessarily have to scale our headcount by that much. So there will be a lot of operating leverage having started with a strong team of engineers and infrastructure guys.

And then on the core infrastructure side, we actually build our own racks and servers, right? This provides us with 20% lower total cost of operations. And then finally, we just have very strong financial management. We take a more conservative view on depreciation. For hoppers, we depreciate those chips at a 4-year depreciation standard versus many of our peers that are doing it at 6 years. It's not because we think the chips have a 4-year useful life, it's just more of a conservative approach to accounting standards.

And then when we also look at these large deals that we are doing, we are focused that these deals will actually deliver in that 20% to 30% EBIT margin range. We are very focused on the economics of these deals and making sure that they are helping us drive towards better economics and better margin expansion over time.

Okay. Thanks, Neil. Thank you, Marc. We're out of time. Really appreciate both of you coming. And for those of you that want to keep this conversation going, walk over with me to Ballroom A, because I'm about to hop on stage with Chase Lochmiller, the CEO of Crusoe; and Mike McNamara, the CEO of Lancium. You may know that Crusoe and Lancium built the Stargate data center down in Texas, and they'll be building out future OpenAI data center. So let's keep this data center conversation going. I'll see you in 10 minutes.

Thank you, Karl.

Thank you, and welcome to Nebius Group's Third Quarter 2025 Earnings Conference Call. I'm Neil Doshi, Vice President of Investor Relations. Joining me today are Arkady Volozh, Founder and CEO and our broader management team.

Our remarks today will include forward-looking statements, which are based on assumptions as of today. Actual results may differ materially as a result of various factors, including those set forth in today's earnings press release and in our report on Form 20-F filed with the SEC. We undertake no obligation to update any forward-looking statements.

During this call, we will present both GAAP and certain non-GAAP financial measures. A reconciliation of GAAP to non-GAAP measures is included in today's earnings press release. The earnings press release, shareholder letter and accompanying investor presentation are available on our website at nebius.com/investor-hub.

And now I'd like to turn the call over to Arkady.

Thanks, Neil, and thank you, everyone, for joining the call today. I'd like to share my thoughts about the demand environment, about our capacity plans and what we are doing in our product.

First about the demand. Q3 demand was very strong. We sold out all of our available capacity. We continue to see a consistent trend every time we bring capacity online, we sell it all of it. With the new generation of NVIDIA Blackwells coming online, more customers are interested in purchasing capacity in advance and securing it for a longer period of time.

Today, we're very pleased to announce that we signed another major deal at this time with Meta for approximately $3 billion over the next 5 years. In fact, the demand for this capacity was overwhelming, and the size of the contract was limited to the amount of capacity that we had available, which means that if we had more, we could have sold more.

This deal comes on top of the Microsoft deal we announced early in September with a contract value between $17.4 billion and $19.4 billion. As we said before, we expect to sign more of these large long-term deals, and we are delivering that promise.

As busy as we are with these mega deals, our main focus is still to build our own core AI cloud business. We made great progress here with AI [indiscernible] start-ups like Prosus, Blackforce, Labs and others. The economics and the cash flow of mega deals are attractive in their own rate but they also enable us to build our core AI cloud business faster. This is our real future opportunity.

Now on the capacity. In order to meet the growing demand, we have accelerated our plans to secure more capacity. And this is actually our main focus for now. Capacity today is the main bottleneck to revenue growth. And we are now working to remove this bottelneck. As we look to 2026, we expect our contracted [indiscernible] to grow to 2.5 gigawatts contracted. This is up from the 1 gigawatt, which we discussed in our previous earnings call in August. Furthermore, we plan to have [indiscernible] connected to our data centers which means fully built of approximately 800 megawatts to 1 gigawatt by the end of 2026 -- by the end of next year.

While we made significant investments in our capacity footprint, we are also investing in our main product, our AI cloud. To extend our addressable market opportunity to large enterprise customers, we released our new enterprise-ready cloud platform version 3.0 called Ether and our new influence platform called Nebius [indiscernible] factory. We believe Ether gives organizations with trust, control and simplicity they need to run their most critical AI workloads. Nebius [indiscernible] factory is a production scale inference platform that enables organizations to run open source models with reliability, visibility and control.

And we have a large pipeline of new software and services that we are continuing to build, which will differentiate us from other cloud companies. Based on the strength in demand that we see and our accelerated capacity growth plan, we believe we can achieve annualized run rate revenue, IRR of $7 billion to $9 billion by the end of 2026.

In summary, Nebius is positioned to win in this large and rapidly expanding AI cloud market. We're just beginning to realize the powerful potential of the AI revolution that is underway. And we are quickly becoming one of the primary cloud and infrastructure providers to support it.

And with this, I would like to hand the call over to our CFO, Dado Alonso. Dado, please?

Thank you, Arkady. While the details of our Q3 financial performance can be found in our shareholder letter, I'd like to provide some additional color to the quarter, discuss our financing options and conclude with 2025 guidance. Q3 group revenue was $146 million, up nearly 355% year-over-year and 39% quarter-over-quarter. Annualized run rate revenue for the core business at the end of September was $551 million. The core infrastructure business, which accounted for nearly 90% of total revenue, grew 400% year-over-year and 40% sequentially. Once again, we sold out our capacity and our revenue growth was limited only by the capacity that we were able to bring online.

I'm also pleased to say that adjusted EBITDA margin for the core infrastructure business expanded quarter-over-quarter to nearly 19%. On financing, in order to support our aggressive growth plans in 2026 and to maintain this stage of growth in 2027, we will be utilizing at least 3 sources: corporate debt, asset-backed financing and equity. We are in the process of raising asset-backed debt, which will be able to secure with attractive terms supported by creditworthiness of our largest customers.

Tomorrow, November 12, we will be putting in place and at the market equity program for up to 25 million Class A shares and a plan to file a prospective supplement. We will evaluate the program regularly based on our capital needs. The program enables us to access equity funding on an efficient ongoing basis. However, we will remain [indiscernible] as we prepare to finance future growth opportunities.

Now I would like to turn to 2025 guidance. As we approach the end of the year, we are tightening our full year group revenue guidance to a range of $500 million to $550 million and we are currently pacing to the midpoint of that range. This compared to the $450 million to $630 million in our previous guidance. The reason we are in the middle and not at the top of that range simply relates to the exact timing of when capacity comes online.

Our current momentum and long-term trajectory remains extremely strong. Our annual run rate revenue, which is a good reflection of our future growth opportunity, continues to expand, demonstrating the resilience and scalability of our business model. As such, we remain well on track to hit our ARR guidance of $900 million to $1.1 billion by the end of 2025, while also paving the way for substantial annualized run rate revenue growth in 2026 and beyond.

In terms of the mega deals, we will begin serving Microsoft and Meta late in the quarter, and almost all of the revenue from these deals will start to be realized on ramp-up during the course of 2026. We plan to give full year revenue guidance for 2026 next quarter.

Turning to adjusted EBITDA. As we have previously indicated, we expect to be slightly positive at the group level by year-end, while remaining negative for the full year. Regarding CapEx, we are raising our 2025 guidance from approximately $2 billion to circa $5 billion. This acceleration reflects our strong conviction in the demand outlook and our decision to secure critical infrastructure, including hardware, power, land and key sites. These investments are strategic enablers of future growth and will position us exceptionally well to capture the opportunities ahead.

In summary, we have a large and rapidly growing opportunity in front of us, and we are executing with focus and discipline to capture it while delivering substantial sustainable growth and setting the stage for strong long-term profitability.

Now let me turn the call over to Neil for Q&A.

Great. Thank you, Dado. We'll give it a moment to collect questions from the online platform, and then we'll begin the Q&A.

All right. Let's start with the first question coming from Alex Platt of D.A. Davidson. Can you tell us more about the new Meta deal? Why did you choose -- why would they choose you? And how should we model the deal? Arkady?

Well, again, as we were happy to announce today this new deal with Meta approximately $3 billion. As I said, the size of the deal was limited only by the capacity that we had available. And if we had more capacity, we could have signed a bigger deal probably. After we announced Microsoft in September, we said that we will have more deals of this kind, more large deals and actually we are delivering on that promise. And we actually -- we are optimistic as these deals -- those deals will arise more and more.

However -- however, these deals are important, these mega deals. It is important to stress that we will remain focused on developing of our own AI cloud, which -- which currently serves not only these big deals, but AI start-ups and enterprises. And ultimately, we believe that these large contracts provide us with great sourcing of financing for us to continue building our core cloud business.

Great. Thank you, Arkady. We'll take the next question from Alex Duval, or analyst from Goldman Sachs. So we provided the updated 2026 ARR outlook of $7 billion to $9 billion. What exactly is in the $7 billion to $9 billion ARR target? And is this based on pre-existing core business plus Microsoft and Meta? Is there anything else in terms of signing up for large deals? Marc, maybe you can take this one.

Thank you, Alex, for your question. And let me walk you through the building blocks of how we get to the $7 billion to $9 billion in ARR. First of all, as we already shared, we had a bottleneck in capacity, and we worked extremely hard over the last several months to unblock this bottleneck. As we shared, we plan to have 800 megawatts to 1 gigawatt of connected power by the end of '26 and 2.5 gigawatts of contracted power.

Second, we see the demand out there from AI start-ups to enterprises to the large strategics, and we see that client demand that we were unable to sell this past year due to a lack of capacity, and we strongly believe that the capacity we are putting in place in '26, will help us to meet more of this demand. At the end of the day, we will allocate between the categories of customers based on the individual economics of the deals they represent.

Thirdly, this new capacity that we are putting in place, together with our current capacity that has already been sold and the long-term contracts that we signed with Microsoft and Meta gives us the confidence that we can achieve the $7 billion to $9 billion of ARR, of which more than half is already booked.

Great. Thank you, Marc. So we'll take another question from Alex at Goldman. Can you walk us through the time line of your infrastructure build-outs for Q4 '25 and '26? And what gives you confidence that you can reach your 2.5 gigawatts goal for contracted capacity? Andrey?

Thanks, Alex. So we are ramping up our capacity as fast as we can to accelerate our growth for the next year and beyond. We are happy to launch now already Israel and U.K. and all the capacity in those regions were presold before the launch. And we are growing with the numbers of the regions where we are present. And we are also bringing new capacity online in the current sites. We are also coming online in New Jersey. We are launching new phases of Finland in Q4 which are also presold, by the way.

In 2026, we will continue to scale in the existing data centers, including U.K., Israel, New Jersey, and we have new data centers already in development both in U.S. and Europe, and they start to come online in the first half of 2026. We also in the process of securing several new life sites, which we believe will add hundreds of megawatts. And some of those will go online by the end of 2026.

So overall, at the moment, we are looking at more than -- around 2.5 gigawatts of contracted power by the end of 2026. And as we said, demand is growing massively, and we are very focused on rapidly building the capacity and the future pipeline to meet the demand in '26 and beyond.

Great. Thank you, Andrey. All right. We'll take a question from some of the folks who have been submitting questions. We're getting a lot of questions on Microsoft and Meta revenue. How should we be thinking about revenue contribution from Microsoft and Meta deals for this year and going forward? Dado?

Well, the Microsoft contract will not have a material effect on our revenue and ARR in 2025 as the first tranche was just delivered. All of our remaining tranches will be delivered in 2026 with more than half of them during the second half. So we actually expect revenue to ramp up over the course of the year. Starting in 2027, we will begin to recognize the full annual revenue run rate of the Microsoft deal. With regards to Meta, we will be concluding the deployments within the next 3 months. So we expect to mostly be at a full revenue run rate in 2026.

Great. Thank you, Dado. Maybe another question from our online audience. What does the overall demand environment look like in Q4 and into the next year? Marc, do you want to take this?

Certainly, certainly. I joined the company about 5 months ago, and I've had an extraordinary experience in these past 5 months. It's extraordinary from the standpoint that I've never seen the kind of demand profile that we're experiencing. It is literally accelerating for Nebius and I believe as well for the broader market. As an example, in the recent quarters before, we saw pipeline generation. This is opportunities by customers that want to buy from us expand. As a matter of fact, in the past quarter, Q3, we saw pipe gen expand 70% quarter-on-quarter, and we generated $4 billion in pipeline in that quarter.

But we were only able to convert a portion of that given to the constraints of our capacity. As a matter of fact, I've learned a new skill, one, I don't think many go-to-market professionals have ever had to experience, and that's learning to say no to customers as we routinely sell out and have to actually let them down lightly and try to convince them to purchase in the future.

As I look out to '26 and I think through the demand profile, the kind of pipeline that we're generating right now has given us high confidence to continue to expand our results and drive towards the ARR growth that Arkady mentioned earlier on the call.

Great. Thanks, Marc. We have a question now from one of our analysts, Nehal Chokshi from Northland. Incremental ARR in September quarter was around $12 million, down from $180 million in the prior quarter and $159 million in the March quarter. Why is incremental ARR down?

Neil, it's a great question. As we've stated, a lot of our revenue and our ARR is really dependent on us able to bring on capacity. And because capacity really has been the bottleneck. That's why we've seen a little bit of that trend. However, as we're bringing on a lot of capacity in Q4, you should see that incremental ARR in Q4 will be significantly higher.

All right. Let's go to another question from online in terms of the CapEx. You have just announced your plan to achieve connected capacity of 800 megawatts to 1 gigawatt by the end of '26. How are you thinking about CapEx? And what is your philosophy on CapEx spending? I think, Arkady, care to take this?

I should take it. Again and again, as we see at least this year, our revenue growth was limited by our capacity and everything will we built was ultimately sold. So in theory, we should try to build as much as we can. In practice, though, we are limited by certain physical [indiscernible] limitations. [indiscernible] cannot grow 5 or 10x a year. We have limitations in supply chain and obtaining permits, amount of capital that we can deploy.

So when we plan for data center CapEx, there are actually 3 stages there. The first stage is securing the land and pulp. The second stage is building the data centers themselves, [indiscernible] physical installation which we call connected [indiscernible]. And the third part is finally deploying the GPUs themselves. And if we look at it from the CapEx point of view, roughly speaking, it breaks into 3 spending blocks. So first stage, securing land and power. It's pretty cheap. It's around, again, it depends on the scale, but it's around 1% of total CapEx for [indiscernible].

The second stage, building [indiscernible] is something around, I don't know, 18%, 20%. And the remaining 80%, the main part is for deploying the actual GPUs. This is the main part of CapEx. So if we want to build as much as our capital will allow us, what should we do? First, we should secure as much capacity as we can because the cost, actually, it's not so -- it's immaterial at this scale. Second, we should build as much as our capital allows. And third, we will fuel GPUs in line with contracted or clearly visible demand. We will need this massive 80% spend will come only when we see real demand. That's why we say that in 2026, we will be securing 2.5 gigawatt total contracted capacity. And we are planning to physically build 800 to 1 gigawatt of connected data centers. This will be done by the end of next year.

Great. Thank you, Arkady. Another question from Alex Duval from Goldman Sachs. You have announced your target is 2.5 gigawatts of contracted power, whereas before it was 1 gigawatt. Is it fair to assume that if you get 2.5 gigawatts, this will equate to over $20 billion of revenue? By when do you envisage you could do this and how? Maybe we'll give this to Andrey.

Thank you, Neil. I guess it's fair to assume. But as Arkady just mentioned, we will -- we are securing the access to the power and the ability to build but we will invest CapEx actually in building out and deploying the GPU in those, keep in mind the constraints that we have with the capital and according to the demand in the future periods. It's just important that we are able to accelerate when it will be needed. So we don't like being blocked by the capacity constraints all the time.

Great. All right. Take another question from online. Is it -- in a situation when you are sold out, is that the same issue -- or is that really an issue with your future growth and differentiation of servicing a broader range of customers? Marc, can you take this one?

Certainly. Thank you, Neil. It's a great question. I mean, in theory, the situation of being sold out is a nice problem to have. But the person asking the question is right for our business model, it's really important for us to be able to not only service large tech companies but also be able to support our AI cloud and a very diverse set of customers. As a matter of fact, servicing start-ups and software vendors and enterprises is not only about delivering on their capacity needs today. We want to build partnerships with these customers and help them to meet their capacity requirements in the future, especially with enterprises because they don't want to actually have a multitude of vendors. They prefer to align with a strategic partner. That's why we are working very closely with Andrey.

And as Andrey mentioned earlier, as we look forward, and think about deploying capacity, it's going to be based on the demand that we're seeing out there. So utilizing the pipeline that we're building and the demand that we're experiencing to work with Andre to identify the capacity that we should deploy. It's a very dynamic model that we're trying to put in place.

Great. Thank you, Marc. Appreciate that. We have a question from Nehal Chokshi from Northland. He's asking going, so you've done equity deals. You've also done equity-linked deals as well, Dado. How will we focus on debt and asset-backed financing for large deals?

Thank you, Nehal. Well, as you know, this is a capital-intensive business. And as we've said previously, funding our growth will require raising a significant amount of capital. In this context, we are actively evaluating a range of financing options today, including asset-backed financing, corporate level debt and equity financing. And we are working on all fronts in order to maintain a disciplined capital structure to maximize our shareholder value.

With regards to asset-backed financing, we believe that we will be able to secure such a facility with attractive terms, supported by the creditworthiness of our largest customers. I would like to reiterate that as we are growing our business, our focus and ultimate goal is to maximize our shareholder value.

Great. Thanks, Dado. And maybe just sticking on the theme of financing from the online portal. Why are you planning to pursue an ATM? You just completed a secondary and this will result in additional solution to shareholders? Any thoughts, Dado?

Any thoughts on perspective? We will be putting in place and at the market equity program for up to 25 million Class A shares, and we plan to file the prospective supplement tomorrow. We want to make sure that we have more tools at our disposal to access capital markets for EBIT. This is a long-lasting program, which will be used along with other capital raise options, including corporate debt asset bank financing and others, as I mentioned in my opening remarks and just before this question. So the program enables us to access equity funding on an recent ongoing basis. However, we will remain dilution-sensitive as we seek to finance future growth opportunities.

Great. Thanks, Dado. See another question from online. How are the early operations of your new U.K. facility progressing? Tom?

Yes. No. Absolutely. So short answer is progressing very well. You may have seen just actually last week, Arkady and a few of us [indiscernible] we had our official launch as we brought the data -- presented the data center to the U.K. market. This is actually capacity that will be coming online really actually in the next week or so, so pretty -- very in the coming days. You might remember actually that in June was when we first announced our intention to launch in the U.K. and actually even in the time since June and now we've already come close to doubling the capacity that we're bringing on stream. And that's just really a function of extremely strong demand that we're seeing in the U.K.

And actually, as often the case with the new capacity that we bring on, even before going live, we're pretty much sold out. So I think they've not already fully sold out with that capacity. So that's a trend that just continues.

I would just say, overall, a few words about the U.K. actually, I mean, we're very bullish actually about the opportunity in the U.K. It's a vibrant AI market. It's probably one of the most dynamics that we see outside of U.S. and China. The government is making a big push to support the growth of the industry and having a reasonable degree of success in this field. So there's a -- we see a lot of AI start-ups. We see environment is strong. You also see some of the kind of the large tech companies establishing regional R&D and presence there. So really, there's a lot happening in the U.K., and we think a lot still to come from Nebius in the U.K. and we're very happy to be there.

And actually, although this specific facility that we have, I think with the capacity once by January, we'll have reached the peak capacity there, we see a lot of other opportunities to expand capacity in the U.K. overall.

Great. Let's see. In terms of -- we'll take another question on capacity. So you mentioned this quarter that you're fully sold out of available capacity, what are your constraints to growing in the near term and medium term to capture more of that demand? And could you also address some of the recent comments in the market around power equipment constraints?

Yes. I'll take it. Yes. Thanks, Neil. As we discussed in most of the previous question, capacity remains our main bottleneck, everything we deploy we sell. And we see the demand that continues to significantly outstrip our supply each time we add new clusters. So in near term, the key challenges to increasing capacity, securing power and the supply chain, and we're addressing this. We have managed these situations in the past, have quite a bit of expertise both building on the data centers and the [indiscernible] those. So overall, we are [indiscernible].

Generally speaking, we are doing quite well actually with the pipeline. And when we spoke last quarter, I believe that we announced that we have secured the road map or the -- to 1 gigawatt of the power. Now we are talking about the number, 2.5 gigawatts. And we are still putting a lot of focus on growing this number and making this number reliable and effective and actually bigger.

Great. Thank you, Andrey. And online we're getting just a few questions about any updates on the New Jersey facility? Andrey, do you want to take that?

Yes, the New Jersey facility goes as planned and the first tranche already was handed over to Microsoft, and we are continuing on the further expansion [indiscernible].

All right. It looks like a question from online. Maybe more of a market question. Are you concerned that we are in an AI bubble? Arkady?

I haven't really been asked this question these days. Well, what we see today, the demand is here, right? And we understand that we are in the center of a once-in-generation [indiscernible]. Much -- no doubt that much more compute will be needed and much more will be built. The situation of unbalanced demand supply is temporary. Of course, eventually demand supply will level up. And what we are doing in addition just to growing this overall capacity. We are building our aircraft which will support real businesses, real industries, real enterprise market, where AI will be creating value. And we believe that AI industry, in general, [indiscernible] sector specifically, it's going to be okay.

Ultimately, we just we need to make sure that, a, we are diversified in terms of customers and workloads. And this is actually what our software is basically doing; b, that we invest conservatively and that we finance our growth responsibly and we are very much focused on this. And also where we're growing rapidly, 5x more a year. We still remain largely focused on meeting healthy margins and a sustainable business model as a whole. In old days, I would say, healthy [indiscernible]. So we are focused on that. And [indiscernible], I would hope that we will be okay.

Great. Thank you, Arkady. Next question is from Alex Platt from D.A. Davidson. How should we think about the lead time between when power is connected to and when it is hooked up to GPUs and generating revenue? Andrey?

Yes. Thanks, Neil. So on the technical side, it also depends if it's in use side or if it's expansion of the current side. But generally speaking, from the connected power and start of the GPU deployment [indiscernible] can go into platform [indiscernible] revenues anywhere from 6 to 12 weeks. If it's already existing site that can be even quicker. But generally, we also have flexibility. That's why we are building infrastructure. We have flexibility when we deploy and we will be [indiscernible] how much we deploy.

Great. Thanks, Andrey. All right. Question from online. Can you update us on your progress with your primary customer segments? Marc, can you help us with this?

Certainly, certainly. We continue to see extremely strong demand from our customers in our core AI business. And we're continuing to expand business overall with our existing customers. As a matter of fact, we added a number of new customers in most notably some very disruptive start-ups like [indiscernible] AI, Blackforce Labs and World Labs. I'm sure everybody's heard of Cursor. We're very proud to be their partner. For those that haven't, they're an extraordinarily popular AI-powered code editor that is helping millions of developers to write and debug and optimize their code faster, and they're making great strides into the enterprise.

Black Forest Labs is an interesting customer that is developing cutting-edge generative AI models, specifically for image and video generation their popular Flex One model helps turn text and images into high-quality media ready visuals. And World Labs is building something they call a large world model, which is able to simulate 3D world, and it gives developers and AI engineers the necessary spatial awareness to build applications for things like media and gaming and architectural design and as well for physical AI and robotics.

We've also, as I mentioned, seen expansion with existing customers. As a matter of fact, as an example, we've seen expansion with our software vendor customers like Shopify. And then also, we've made great strides with our efforts around driving vertical market success, adding significant customers in our health care life sciences part of the business. And we're also making significant advancements in physical AI and media entertainment customer segments.

Great. Thank you, Marc. Maybe a question to -- looks like this is a question for Dado. Few people are asking, any puts and takes that you can provide on your revised 2025 year-end revenue guidance?

Well, our businesses are to scale rapidly. There can always be fluctuations in the exact time of deployments in such a fast-growing company like ours. And our focus remains on building a very large company, obviously much larger than today and significantly bigger than our plans for 2026. This was and continues to be our main focus. In any event, our annualized run rate revenue, which is a better reflection of our future growth opportunity continues to expand, demonstrating the resilience and scalability of our business model.

As such, we remain well on track to hit our ARR guidance range of $900 million to $1.1 billion at the end of 2025 while also paving the way from substantial revenue growth in 2026 and beyond.

Great. Thank you, Dado. See question from the online community, how is your enterprise initiative ramping up? Do you seem to make some good improvements there over the past couple of quarters. Marc, do you want to help take this?

Certainly, certainly. Yes, we are making strides with regard to becoming enterprise ready. As you saw with the launch of Nebius 3.0, what we call [indiscernible], we've delivered a number of AI cloud improvements to support enterprise requirements. As an example, in the [indiscernible] release, we are delivering really important compliance and security certifications. And we did this as a matter of fact, in a matter of months when it would normally take other organizations a lot longer to deliver these types of capabilities. As a matter of fact, as well, we delivered some important functionality that enables enterprise administrators to proactively manage their implementation.

So tooling and controls like identity and access management and dashboards for evaluating the performance and security of their implementation. I think as we all know, the sort of the critical foundation for enterprise readiness is to have these kinds of compliance and certifications in place and the enterprise functionality that enterprise is looking for.

And the third is to have an enterprise-ready sales team. On that front, we are adding a number of key leaders to our organization, and we are expanding the overall sales organization for coverage in enterprise software vendors and key verticals. It will take some time for the sales team to ramp but we are building the foundation between the functionality that I mentioned and the overall team coverage that I think will set us up for a strong 2026 with enterprises.

Great. Thank you, Marc. Keeping with our online investor base. You recently launched [indiscernible] factory, what is the opportunity around this? And will this expand your market or open up new segments? Maybe we'll ask Roman to take this one.

Thank you, Neil. Happy to talk about our new launch. So I will start a little bit from demand evolution. We fairly see now the next wave of AI demand growth. And it's mostly driven by the companies, but the people who apply real-world applications across all industries in B2C and B2B. It's not necessarily a foundational model builders like it was, let's call it, the first wave. And we as Nebius realized that we needed our inference as a service offering to make to make it serve a broader set of customers, including enterprises. So [indiscernible] factory gives vertical product builders, ICs and enterprises a platform to build, we call it flywheel of applying LLMs in vertical use cases at scale. Transforming -- we help them to transform open source models into optimized production-ready systems with guaranteed performance and transparent cost per token.

We obviously leveraged the underlying infrastructure to bring the most efficient and scalable solution to our customers when they can be sure that they get the best total cost of ownership and can confidently grow with us. So as a result, organizations can deploy and scale models such as OpenAI [indiscernible], DeepSeek, Lllama, [indiscernible] and many others on dedicated endpoints with guaranteed performance tuned for the super latency and 99.9% uptime. So in total, I must say we are excited about the opportunity of inference workloads. We believe that all companies will invest in inference to productize AI. And for us, it means like it will require significantly more compute and will support this wave of growth as well as we do for foundational model builders.

Great. Thank you, Roman. Jumping back to online. It looks like we have some additional questions here. What demand are you seeing for new [indiscernible] generation? And how is this demand from the previous hopper generation? Marc, do you want to take this one?

Yes, certainly. Thank you, Neil. Demand remains very strong across all types of GPUs. And as we said, we sold out our capacity in Q3, and that's across all types. And we're nearly sold out with respect to Q4. Talking about the hoppers, we continue to see extremely positive demand for these chips. An interesting set of dynamics that we're experiencing is that as customers come to their renewal, for hoppers or if they're looking to upgrade to, say, Blackwells in both cases, we're typically selling them immediately. And often case -- and often at better pricing than they were previously priced as we're actually in tandem rolling out the black wells. So very strong demand profile for existing offers. We're also seeing very strong demand for Blackwells.

And we're benefiting from the fact that we're one of the first companies to deliver them in the market. In our Israel data center, we launched with B200s. And in our U.K. data center, we launched with B300. And we've essentially presold much of that capacity before these facilities even opened. We're very excited as well that we're launching GB300s. We're the first to do so in Europe which will be coming online or finished data center later this quarter in December.

Thinking about in production capacity, right now, as I mentioned, selling the remnants of Q4 but we're also now preselling new capacity being delivered in future quarters. So we're seeing a very strong demand across all types of GPUs. And as I mentioned earlier and as Andrey mentioned earlier, we're working in close partnership with Andrey's team to make sure that our sales pipeline allows us to drive our model in order to be able to support GPU requirements in subsequent quarters.

Thank you, Marc. Another question from Alex Platt from D.A. Davidson. He is was asking about our strategy regarding larger deals. Do we have medium-term capacity targeted for these deals and customers? Arkady?

Yes, we are very opportunistic here. The demand is there not only for our everyday deals, but for large mega deals as well. And we will enter into the deals, which provide us with the best margins. We're very much focused on margins and profitability, not only in growth itself and all our decisions actually direct from there.

Great. Thank you, Arkady. Our next question from Andrew Beal from [indiscernible]. Can you provide more details regarding some of the greenfield sites? Do you have LOIs for further new U.S. and [indiscernible] locations? Or are you further down the road with these? Andrey?

Yes. Thanks, Neil. Generally, we are making a great progress in [indiscernible]. We have a robust pipeline, both in Europe and U.S. We mentioned that we are on the way of securing 2.2 gigawatts of, well, road map for the power in the next year, we are in [indiscernible] also further down the road as well. But we are not in a position to say more at this stage.

Great. All right. Question from online. Can you provide an update on your facility in Israel? Tom?

Yes, sure. So as you can see, we're growing rapidly. So just last week, we were in the U.K. launching; in just a couple of weeks before that, we were in Israel. And actually, the day center facility that we have there is already fully live. And as I think we've made various references to Marc's mentioned it previously, again, that was capacity that was effectively even presold. And we definitely have opportunities to expand further in terms of capacity. We think Israel is a great market. Again, we see a lot of demand. There's a lot happening in tech and in AI.

And actually, one of the things that's interesting about the market, I mean our decision to go in there was purely based on our own commercial considerations, we think there's a great -- there's a lot of growth for that. But actually, the government is also doing some interesting things, really to stimulate further demand. And so they're actually effectively putting money to subsidize sort of AI start-ups and institutions and helping them to access the compute as a way of getting -- having the growth move faster. So we think it's -- anyway, we're doing great there. We have -- we think it is a great opportunity for us. And actually, the model that might even be a model that we think [indiscernible] other countries might look at that are thinking about building up that domestic demand in AI industry. So overall, going great.

Thanks, Tom. All right. From our online platform, let's see, how do you think about partnering with or buying potential companies that already have secured power or land or consolidating or consolidating other neo clouds? Arkady?

Well, companies -- we secure power and land, again and again, it's all about margins. We are pretty much focused on the margins when we enter into a new contract when we are raising capital, when we're developing new products for data centers, when we designed those data centers, when we build our own racks, software and so, we are vertically integrated, and we are looking on efficiency on each stage. We have looked into potential acquisitions of power/land market. But so far, our approach proved to bring, say, much higher margins so far. So we are still moving further and further into building our own facilities, and we're actually decreasing the share of our own [indiscernible] branded facilities. More and more, we will be our own facilities.

Obviously, we will continue to consider different opportunities. But as you can see, we were able to secure a significant level of contract power organically. So we strongly believe that sooner later, the margins for the infrastructure business will play a significant role in the ability to grow and develop. And we have basically remain very focused on that.

Great. Thank you, Arkady. All right. Let's -- another question, kind of market-related question. Maybe this one will go to you Arkady. Is there any chance that GPUs are oversupplied in the coming year as new suppliers come to the market?

Two things here. First, we strongly believe that the market will still be supply constrained, at least in 2026. Means that data center capacity will be the [indiscernible]. Also, as we mentioned earlier, we plan our capital spend in those 3 stages: land power, building the [indiscernible] facilities and [indiscernible]. And this conservative stage approach keeps us from our spending actually and allows us to maintain a healthy financial position. If there are any changes in the market, we'll be in good shape to [indiscernible] any downturn, we hope.

Thank you, Arkady. Let's see. Couple of ours analysts are asking in terms of any big challenges regarding the completion of the [indiscernible] facility or any challenges to meeting any performance obligations out [indiscernible] Microsoft deal?

Thanks, Neil. I think I already spoke about it. So yes, as of today, it goes as planned, and we already handed over the first tranche to the Microsoft. So we are continuing [indiscernible] according to the plan.

Great. All right. I think we'll end the call there. Thank you, everyone, for joining, and we will speak to you again next quarter.