CoreWeave Aktie

All right. Fantastic. We will go ahead and kick it off on stage with the CoreWeave session at the Goldman Sachs Communacopia Conference. I'm Gabriela Borges, and I'm delighted to have Mike Intrator on stage with me, CEO of CoreWeave. Thank you for coming.

Excited to be here.

Mike, I wanted to start with a little bit of a technical question for you, which is we've all seen the data points from third-party industry. We've, in selective cases, been able to speak with customers. And they'll consistently tell us that CoreWeave is able to deliver a level of performance on GPU training that is unparalleled in the industry. And so give us your best layman's explanation on why you think you've been able to go from a crypto miner to offering the best performance in GPU training in the industry [indiscernible].

Sure. I'll do the best I can. So one of the things that's important to understand is when we built the company, we kind of -- we had an opportunity to kind of break it down to first principles and say, if you were going to build a cloud, specifically built to address the needs of artificial intelligence, parallelized computing, how would you go about building that? And the clear answer was you don't go about doing that by retrofitting a legacy environment to go ahead and create the solution. It's the -- like we jokingly talk about it internally, it's like the minivan versus the F1, right? It's just a -- it is a tool that was specifically built in every decision from when we founded the company in order to optimize for the specific use case, not to provide generalized computing. There are lots of great options for that.

And so what you have is an environment that has these compounding incremental thousands of decisions that have been built around making the environment as performant as possible for this specific use case. And we select our clients carefully in terms of finding folks that need this use case to be able to be supported and you find a software stack and infrastructure stack, all of those things really built to be able to provide that environment in the way that we do. And so it's great to hear that the clients that you're speaking to are taking advantage of the efficiencies that the tooling that we built provides them.

The next question I have for you is taking some of those performance decisions that have been made for training GPUs and then extending that advantage to something like post training and something like inference. Talk to us about why those decisions that optimize for training also give you a performance advantage for post training or for inference.

Sure. So one of the things that we kind of believed very early on was that the space was going to evolve very, very quickly. And when you're in an environment as dynamic as that, you want to build your infrastructure to have the most amount of optionality embedded in it as possible, right? And so we don't really think about our infrastructure as being, oh, this is our training infrastructure, and this is our inference infrastructure. We think of it as AI infrastructure. And we build to a standard that allows our clients to use that infrastructure in whatever way they deem to be most productive for their businesses, and that's how we built it.

And so all of the tooling that we have within our Mission Control and all the other observability suites and all the tooling that we have provides for an environment that allows for incredibly performant environment for folks that are using the compute. And we can see when people are using it in different ways, the consumption profile will look different. All of those things will follow with their use case. But to varying degrees, different parts of our stack will provide for different advantages depending upon what their use case is.

This all makes sense. I want to pick your brain on a couple of industry dynamics that we get asked about all the time that are very core to where CoreWeave sits in the stack. So the first one is supply-demand, where CoreWeave has consistently said that we might not be at supply-demand equilibrium until the end of the decade. And so my question to you is tell us a little bit more about how the demand environment has evolved this year because of agentic? And where do you think it goes from here? What do you think is being underappreciated about agentic demand over the next 12 months?

Yes. So the -- we have been incredibly consistent in the way that we talk about demand. And that has been driven by our ability to talk to an incredible cross-section of compute consumers, right? And that's everything from the hyperscalers through the AI native, the frontier labs, people that are productizing and delivering agentic solutions, all of those things. And we have never flinched from our assessment that the world's capacity to deliver compute has been and will continue to be wildly overwhelmed by the delivery and demand for intelligence, broadly speaking. And so we've been righter than wrong on that one. And I've got a bunch of really smart folks around me that come in and yell at me every day that we don't have enough compute. So I'll take their word for it.

The -- one of the really interesting, I'm not sure this exactly answers your question, but one of the really interesting phenomenon that you're seeing in the market right now is the large consumers of our services of compute delivery of the software stack that we've built on top of that. Those consumers are demanding ever larger amounts of resources, right? It's -- every -- we have the 9-month rule at CoreWeave. We look back at whatever we've done and we -- 9 months later, we think it's cute. And that's kind of how we've -- that's the dynamic that we've seen in the market. But that is being compounded right now because of new entrants into the market. And new entrants are enterprise clients coming in. And I talked about this a little bit on my last earnings call where a company like Caterpillar comes and starts to use us. And it's a great example of enterprise consumer coming to us and saying, hey, the way that we're going to participate with our company in artificial intelligence, the way that we're going to train our models, the way that we're going to serve our models is going to look different than it has historically. We're going to build our own clusters. And that is transformational, right? That is really allowing CoreWeave to establish itself as an AI cloud participant, a hyperscaler within the AI system. And that's been really, really exciting for us.

I'm curious what you think changed that now has enterprise customers wanting to build their own AI clusters in-house versus outsourcing?

I think it's a lot of things. I think part of it is the experience of building cloud-based systems over the past 20 years has given companies some perspective about how important compute was going to be. And then with the advent of artificial intelligence and the degree to which that kind of raises the stakes, there is an interest in having more control over that resource that they consider to be fundamental and important to their businesses than perhaps they did historically. It's the boiled lobster problem, right? It's like the cloud kind of came out of nowhere and people began to use it and people were using it before they realized how incorporated and embedded it was going to be within their critical systems.

And this time, they're going in with eyes wide open, which really provides us with a wonderful opportunity to go to them and say, hey, we can really assist you beyond what has been an oligopoly of 3 massive companies that have delivered this service. We have some alternatives. We have a different way of looking at it. We have a different way of providing the compute that will be most performant to you. And that drives down cost. It does all kinds of great things for you.

Very good. Okay. So my second industry question is world models. When do you think world models start to have an impact on the demand curve to CoreWeave? And I know you've already announced customers in this space.

Yes. So look, it's the same but different is the way I look at that. The value propositions that a company like CoreWeave provide to that space are really quite similar to the value propositions that we provide across the entire space. Yes, their data is a little different. All of those things are true. But I do think that really when you're thinking about the world models or agentic, like it really comes down to have you built your infrastructure, have you built the software rails and the stack to provide an environment that is going to be incredibly performant, incredibly well controlled, incredibly cost effective from the amount of compute that you're able to squeeze out of every dollar you invest. And all of those things are -- that is the North Star for CoreWeave. It's how we built our business from day 1.

We're about to go into midterm elections here, and we've seen on both sides of the aisle, a number of debates around data centers that, in aggregate, have made ease of scaling and securing a data center harder than it was a year ago or 2 years ago. How do you navigate at CoreWeave, the stuff that's within your control? How do you plan for what's not in your control? Where do you think this debate goes over the next year?

Okay. So somehow in a relatively short period of time, I turned into a bad guy according to my son because of the data center dialogue that's out there. Look, we have a method and approach to how you go about building and scaling data center capacity that is really built around entering the markets early, really working with the local communities, being sensitive to their needs and building infrastructure that is useful to our clients while not being unreasonably irritating to the local communities. And that has been a very productive way to approach building infrastructure.

The world is definitely evolving, right? There is increased resistance to data centers going into communities. There are pieces of that debate that I think are completely nonsense, and I think there are pieces of the debate that are -- that warrant real introspection by the industry. And so water usage, these are closed-loop systems. And as more of the data centers come online that are able to support this AI infrastructure that is so power-intensive, they must be closed-loop systems. They don't function unless they are. And so I think that it is at a -- going into the elections, it's going to be a maximum volume. I think that the approach of not allowing data centers to be built is kind of a fool's errand in some ways because I think the demand for the compute doesn't change. It just moves in terms of where it's going to be built and how much it's going to cost to build. But I don't think it changes the overarching demand cycle.

We have been really, really forward thinking around the idea of diversifying our portfolio of data center infrastructure. And once again, I talked a little bit about this on my earnings call because it's such an incredibly high-profile issue. But like I talked about the idea that we have over 1 gigawatt of power contracted outside of the U.S. as you kind of look to build a portfolio of infrastructure to be able to support your clients with. And I think that's the right strategy, and I think we'll continue to kind of focus on making sure that we can build both within the domestic U.S. but abroad as well.

Maybe I can bridge this to a horizontal and vertical integration question. So we'll talk a little bit more about your ability to go up into the software stack. What I want to ask you is why now you feel the need to also own the underlying data centers. I think you have your first data center coming online by the end of this year. What's driving that decision on vertical integration beyond the margin stacking argument?

So this is not a new issue for us. We've thought a lot about this. We've spent a lot of time thinking about which parts of the vertical integration we want to be involved with. There are 2 reasons that you want to vertically integrate downward. And both of them are very important. The first one is that by vertically integrating down into the physical data centers, you can recapture some margin, right? And I would say that's the lesser of the 2.

The second thing is this stuff is difficult, right? It's not, hey, I've got some megawatts and I'm going to turn that into a supercomputer. There's a lot of steps along the way there. And entering into an environment where we have more control, a deeper understanding, a better sandbox to work on the flywheel of new innovations that we want to bring to our own data centers, that makes us stronger, right? More control over pieces of our infrastructure make us better. And so it's not a new part of our strategy. We've thought a lot about it.

Data centers take time to build. And so you're starting to see the first 2 or 3 of our self-builds spin up, which we're incredibly excited about. But you will see a continued effort on our part to control the physical infrastructure, both up and down the stack. We think it's important. We think control is important, learning is important.

Let me stay on this point on bringing more capacity online is very hard operationally to do. We've seen so many announcements from new entrants in the past several months where we'll get data points on companies bringing on 10 gigawatts over the next 2 years, 7 gigawatts over the next 2 years. How do you think this all shakes out from an industry structure standpoint?

I think it's hard. I think building infrastructure is difficult. I think that some will be more successful than others. I think that it is a capital-intensive industry and capital-intensive industries will tend towards periods of proliferation and they will tend towards periods of consolidation. I think that during a time where the demand is so incredibly overwhelming, it is easier to launch a company.

But the difference between launching a company and talking about a contract you signed and actually being able to operationalize and deliver and maintain that infrastructure, the more experience you have with it, the more you realize how difficult it is to do over time. And so I think there's going to be some interesting outcomes that are going to be associated with delivering this new infrastructure.

Absolutely. Okay. Let me ask you the supply-demand question in a slightly different way. So we've established that we're in a period of incredible supply constraints. At some point, and we can debate when that is, the industry will come back into supply-demand balance. Right now, CoreWeave gives us these amazing data points on contract renewals for H100s and pricing dynamics. How should we think about longer term, what happens when supply-demand normalizes? And the bear case we hear from investors all the time is, well, you have hyperscaler customers who can simply then pull back what they've "outsourced" to CoreWeave and retrench back in the stack. And so maybe just would love your thoughts on the longer term for CoreWeave in a normalized supply-demand environment.

So I think my last answer is the most like politically correct answer I could have given to that question. Look, CoreWeave has been leading the AI cloud space now since its inception. I would argue that largely CoreWeave has really built the space. What started out as a cost-plus environment where we were able to win contracts because we would go to large consumers of compute and say, hey, we'll build this for you and they would say, well, we're willing to pay cost-plus for it. That kind of built our business and allowed us to scale. And with scale in this business comes all kinds of incredible ways of driving economics and driving margin, which we're really excited about, and the market will see that continue to accrete to us over time.

The other piece of it is just these companies, they entered into contracts with us. And when we IPO-ed, they were like, oh, you're never going to do that again. It was just a one-off. And then they came back and did it 5 more times with us, and then they did it across the industry. And the argument continues, oh, they're just going to pull it back. And that doesn't really -- those 2 things, right? It doesn't really match with the fact pattern, right? They build data centers and they could have pulled those back, but they didn't, right?

They go through these periods where they build their own data centers and then they go through these periods where they go to third-party data center providers. So that's one thing. We do not believe that they are going to pull back, right? We just don't think that, that is the way that it's going to play out. But the second piece of it is what these contracts did for us was provide us with time, right? And with that time, you have seen us build a cloud that is specifically built to be able to support AI use cases. And with the software stacks, with the infrastructure, with the reputation for delivering this incredibly high-quality product, we have begun to win a broader and broader universe of clients.

And so some of those contracts may come back to us. Some of those contracts may not. But the size and scale of the business, our ability to support enterprise, our ability to support governments, our ability to support our own products internally, our managed inference product, all of those things, those internal demand things, those are incredible paths to building a sustainable cloud. And we believe that is a necessary part of exiting this period of disequilibrium in a way that allows you to be successful and a hyperscale provider of the infrastructure that is required by the world for artificial intelligence. And we're really well positioned to do that.

Let's talk about building the cloud. So there are a number of pieces that I think are interesting. We could talk about managed inference, for example. There's a CPU pull-through piece to this as observability. Maybe just take a step back for us. What are the pieces that you think are most strategic to going from AI neo-cloud to full-blown long-term hyperscaler winner?

So the answer to that is we meet our clients where they are, right? And so when we go and we work with the client, they tell us what they want, right? And we can be extremely flexible about integrating, hey, we've got this particular type of storage we want to use. Okay, we'll build it into our system, not a big deal. So like it really depends on who the client is, what the workload they're trying to support will define what we bring to market. The attach rates, and I think I spoke about this maybe 2 earnings ago, like 75% of our clients are using 3 different services, right? 3 different silos within our product suite are already being consumed, right? That's exactly what you want as you're building the cloud. That is exactly the way that you measure how people are going to consume what you're doing.

And I would argue that the percentages are even higher than what they look like because some of our clients are these really, really early-stage ventures that are not up to product # 2 and product #3 yet. They're still kind of like working it out. And so we really think that your -- you've got to build a suite of software services, whether it's storage or memory or networking and all the different things that people are going to consume and then go ahead and put it in front of them and say, look, we can help you build your company. We can help you deliver your service. We can help you be successful because of the infrastructure that you are going to be able to run your products on. And that has been an engine for us.

Absolutely. In terms of road map, anything you can share on where customers are pulling next?

We are -- we've always talked about the fact that we are client-led, right? And being client-led means that you spend time with your important clients and you talk to them about what they're trying to accomplish and you let them kind of help you build your products. So a lot of the storage solutions that we have were built specifically to help folks that had a very specific problem. When you look at the way that our road map has led us internationally, right, to ensure that we are both resilient from a capacity factor, but also able to serve different markets with low latency infrastructure.

I mean like all of those things are really manifestations of our clients and what they need, and that really does inform the road map. I mean sitting at the center of the flywheel, having the conversations with all of these incredible entrepreneurs or existing companies that are integrating artificial intelligence into their workflows. And that is really, really fascinating, right? So just absolutely fascinating to see how that works.

Maybe the perfect case study is actually someone like a Caterpillar, which you were talking about earlier. Give us a sense for the flavor of the enterprise customer conversations that you're having today where customers will say, look, we have previously used 1 of 3 hyperscalers. And we're now incrementally moving net new workloads or even brownfield workloads over to CoreWeave. What does that look like in practice?

So what it looks like is when we bring a customer like that over, we try to wrap our arms around them and support them through the process, right? And so it really depends on who the customer is and what they're trying to build and what they're trying -- and we try to make it so that all of their resources are being focused on integrating AI to make their company better, faster, stronger, whatever -- however you want to talk about it.

What we really are trying to do is simplify the process of ensuring that they have the most performant infrastructure to be able to train internal models, right, that they're able to serve their inference calls effectively. All of those pieces have to come together for us to be viewed as a viable alternative to one of the hyperscalers. And so the breadth of the services, like it is -- it's not endless, right? We really do say, hey, this is our lane and our lane is to be able to serve all the components that are required to drive artificial intelligence. And then we don't wander too far from it, right? Like the world doesn't need another solution for X, Y and Z. What it needs is the best solution for the AI workflow, and that's where we're focusing our energy.

Mike, you've already touched on the diversity of your customer base. Do you have a view or do you -- how do you think about the mix between frontier models and open source, open weights? Does that have a second derivative impact either on the health of your customer base or the health of your business?

So we're in a position -- when we think about like closed source versus open source, believing that this is not going to be a binary breakdown, right? There are going to be use cases for open source. There are going to be use cases for closed source. As a supplier of the cloud that is required to run these use cases regardless of whether it's open or closed, we're sort of indifferent, right? We want to make sure that our infrastructure is able to serve both sides of that fence equally well, equally as efficiently. And that's sort of how we've approached the problem.

I think there are other people that have to spend more time deciding which models are appropriate for which workflows within their organization. But for us, it's the idea that we will provide the computing power that you need in order to be successful regardless of how you choose to allocate the infrastructure in terms of whether it's open source or closed source model. So we're sort of a cop-out, but we're sort of indifferent on that, right? Like it's just more demand for compute is better for our business.

The other part of the strategy that I wanted to ask you about is the relationship with NVIDIA on selling through pieces of software. Maybe bring us up to speed on some of your ability to take proprietary CoreWeave software and sell it outside of the immediate CoreWeave stack.

Yes. That was pretty cool. Yes, it was. It really was. We -- NVIDIA designated our software stack as a reference architecture. And that's super exciting for us because it really is a recognition of the quality of the solutions that we have built, right? And we're -- that was great. We had some really excited engineers that have put their blood, sweat and tears into building the infrastructure. And so it was a wonderful recognition that they thought the quality of what we have built is high enough to be designated a reference architecture. And so that was really great.

What it means in reality is that the ability to provide third parties access to our software solution on their infrastructure allows us to get some leverage on other balance sheets. And that's a great thing, right? Like we can go into companies that want to own their own infrastructure, and we can provide them a software layer. It's called Omni is what we call it. But it also allows us to build infrastructure in jurisdictions that we maybe are not comfortable owning or delivering an asset-heavy solution to for whatever reason. And once again, that's a great way to make use of other entities' ability to take the risk around the physical infrastructure while still being able to generate returns and to expand our TAM for the products that we bring to market. And so that's been a very exciting part of our portfolio. We closed a deal on that a couple of months ago, and there's a whole bunch coming down the pipe that really look like that for various reasons why we would use that model instead.

I want to spend a couple of minutes here on capital structure because the appetite of the debt markets to own neo-cloud businesses ebbs and flows over time. Tell us a little bit about your plans for capital raising. How do you think about the optimal balance sheet structure for a company like CoreWeave?

Yes. Look, we took a strategy around building our company that is, in many ways, part and parcel for how you go about or how the world has historically gone about building infrastructure and capital-intensive business, right? If you're SpaceX and you're valued at $2 trillion and you can sell some equity, you sell some equity and you can go ahead and build whatever you need to build because $2 trillion is a lot. But if you're building organically, like we were, the question is, how do you raise the capital to build at a scale that allows you to be of relevance. And the answer to that was clearly the debt markets. And I think that the market struggles to understand what we have done in the debt markets.

And the largest engine of our borrowing is occurring at an SPV level, where we are providing transparency through to the credit of the offtake. And that is why we were able to build structures that allowed the credit market to underwrite what we were doing because they were basically looking through and saying, okay, it's Microsoft on the other side of this transaction. The money is going to flow into the SPV and then pay back the debt before it goes back to CoreWeave. And that was an incredibly effective engine for raising capital at an order of magnitude that has rarely ever been done by a company as new to the market as CoreWeave. That was a wonderful way of enabling us to drive the scale that we needed to be able to serve our clients.

We also raised some debt at the ParentCo. We raised convertibles up at the ParentCo. We have a very -- not a lot of religion around this stuff, right? Like we look at the space and say, what is the most effective way, the cheapest way, to raise the best possible capital to be able to execute on our road map. And that has been the defining North Star around the capital market structures that we use. And we have been the tip of the spear around the innovation. We did the first GPU transaction. The -- just kind of looking back in history, the last debt structure we did was the first time anybody has been willing to take -- or any of the lenders have been willing to take renewal risk.

Like that's an important step forward to being able to access capital markets to be able to take advantage of the short-term contracts because now the lenders are taking renewal risk. The one before that, we borrowed -- we were able to get the [ venture rated ] so that we were able to borrow at A- credit, like that's unbelievable, right, to be able to borrow for a company like us to be able to borrow at such a low cost of capital was kind of a crowning event within the group that builds these capital structures. It was incredible. And so it puts our borrowing really at par or close to at par with many of the hyperscalers that are out there raising capital, and that allows us to operate more effectively, more competitively.

Mike, congrats on all the milestones. Please join me in thanking Mike for his time.

Hello, everyone. Thank you for joining us, and welcome to CoreWeave's Second Quarter 2026 Earnings Call. [Operator Instructions] I will now hand the conference over to CoreWeave. Please go ahead.

Thank you. Good afternoon, and welcome to CoreWeave's Second Quarter 2026 Earnings Conference Call. Joining the call today to discuss our results are Mike Intrator, CEO; and Nitin Agrawal, CFO. Before we get started, I would like to take this opportunity to remind you that our remarks today will include forward-looking statements. Actual results may differ materially from those contemplated by these forward-looking statements because of factors that are set forth in today's earnings press release and in our quarterly report on Form 10-Q to be filed with the SEC. Any forward-looking statements that we make on this call are based on assumptions as of today, and we undertake no obligation to update these statements as a result of new information or future events.

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 an accompanying investor presentation are available on our website at investors.corweave.com. A replay of this call will also be available on our Investor Relations website. And now I'd like to turn the call over to Mike.

Good afternoon, everyone, and thank you for joining us. Q2 was an exceptional quarter for CoreWeave. We outperformed our plan across the board with the operating leverage we have been building beginning to show up clearly in our results. Extraordinary execution across the organization, drove record financial performance, rapid capacity growth, broadening customer demand and continued platform innovation. We generated record revenue of $2.6 billion up 112% year-over-year. Increased revenue backlog to $104 billion while driving rapidly expanding enterprise adoption. This figure does not include the over $25 billion of net new customer commitments added in the early weeks of Q3.

We continue to execute on our power strategy, reaching 1.5 gigawatts of active power, adding nearly 500 megawatts, more than any quarter in our history and more than tripling year-over-year. We remain firmly on track to reach at least 8 gigawatts by 2030. We grew adjusted operating income to $128 million, with margins expanding meaningfully as our scale increasingly translates into operating leverage, and we continue to broaden our technology stack, delivering 7 new AI platform capabilities. and achieving multiple industry-first milestones that enable customers to build, deploy and operate AI faster and at greater scale. Our incredible progress is a testament to the entire organization, and the business is only getting stronger. In Q2, the customer contracts we signed came with contribution margins we expect to be 5 to 10 percentage points above those added in recent quarters. This is more than a collection of milestones. It is evidence that the AI market is developing in line with the convictions on which we build CoreWeave.

We believe that the AI era is here and will ultimately touch every part of the global economy, that every organization is being transformed, creating an opportunity to reinvent established markets and create entirely new ones that the future will be led by the pioneers who seize that opportunity, both AI native companies at the frontier and change makers inside established enterprises. That learning and iterating at light speed are now table stakes for AI leadership. And we believe those pioneers need a new kind of platform to unleash AI's potential at scale. These beliefs are the operating assumptions that drive our strategy. They guide our product road map, our capital allocation, our partnerships and ultimately, how we serve our customers.

Today, I want to discuss how our vision is translating into 4 areas: one, broadening demand and customer adoption; two, the continuous AI development tools we provide on our platform; three, the performance and economics enabled by our AI native architecture; and four, the power and supply chain foundation that will support years of growth. AI is transforming every organization. The debate around future demand for AI cloud infrastructure will likely continue. But what we know with certainty is informed by our customers' actions. Demand continues to intensify as the market broadens across sectors, geographies, workloads and generations of GPU architecture.

AI is no longer confined to frontier model labs. It is becoming embedded in software, industrial systems, financial markets, enterprise workflows and national security missions. We see that breadth in our backlog, in the new commitments we have signed and in the utilization and pricing environment across our platform. Pricing and margins for our Blackwell and Vera Rubin SKUs are setting new highs, while pricing for prior generation SKUs is at or above where it was years ago. Our near-term capacity remains effectively sold out. That is translating into signed commitments on increasingly favorable terms from a broadening set of customers and is positioning Core Weave to gain market share for years to come.

Every organization is being challenged to rethink what is possible. AI is not simply accelerating existing processes. It is enabling enterprises to redesign core functions, create new offerings and enter markets that did not exist before. Caterpillar is a powerful example. Together, we will deploy NVIDIA's Vera Rubin platform to support Caterpillar's physical AI training and inference at industrial scale. Using CoreWeave's AI cloud infrastructure as its data factory, Caterpillar will train specialized models that enhance the intelligence and productivity of autonomous construction equipment.

Life sciences is emerging as another important growth vertical for CoreWeave, as organizations tackling some of the world's most complex scientific challenges increasingly turn to our platform. We recently welcomed isomorphic Labs as a new customer and are excited to support them in their mission to solve all disease. Financial services remains a major growth area for us as well. We recently added Flow Traders and IMC to our growing roster of systematic trading firms. These customers are using CoreWeave to develop and deploy the next generation of AI models for quantitative trading. They select our platform because of our ability to orchestrate high-performance workloads with the speed, reliability and efficiency those applications require. And in the public sector, our collaboration with Leidos marks an important step in the growth of core Weave federal.

Together, we are working to accelerate the delivery of secure AI capabilities for defense, national security and intelligence missions. The future of AI is being built by a new class of innovators. Some are AI native companies operating at the frontier. Others are change makers inside established enterprises who are willing to challenge the status quo. CoreWeave serves both Companies such as Descartes are using our platform to develop Oasis 3, the first API accessible world model for physical AI. IBM is using core weave to experiment securely with reinforcement learning, agent tool use and model evaluation. And through monolith our specialist field engineers work directly alongside customers, such as Nissan and ZF to accelerate the development of enterprise-ready AI applications.

Demand now extends beyond our infrastructure as well. Through CoreWeave Omni, we are seeing significant interest from sovereign, enterprise and cloud customers alike. In the past few weeks, we signed our first deal, which will begin to scale in 2027. These examples differ by industry and use case, but the pattern is consistent, AI is moving from experimentation into core operations and the organizations that act decisively are creating an advantage. Deployment is no longer the finish line. As AI moves into production, the way applications are built is changing, and the leaders will be those who learn and iterate the fastest. For the last several years, many organizations treated a model like a deliverable, train it, deploy it and move on. Enterprises no longer operate that way. training, inference, evaluation and improvement now for a single continuous loop. Models and agents in production generate real-world data. That data informs a valuation driving new experiments, which improve the model or application before being redeployed into production.

The loop repeats and capability compounds over time. That shift fundamentally changes both the demand curve and the economics of AI. Compute is no longer a onetime requirement concentrated at the beginning of a model's life. It becomes an ongoing requirement that grows with every application in production and every cycle of improvement. Our AI native platform was built for this. It spans cutting-edge cloud infrastructure, a rapidly growing managed inference business, leading developer tooling and agent solutions and a best-in-class orchestration and observability layer powered by mission control. Together, these capabilities give customers 1 integrated environment. Customers deploy applications through CoreWeave inference, using our models or the ones they have customized with our serverless capabilities. monitor performance with weights and biases, evaluate applications in production, experiment with new models, refine their performance through serverless reinforcement learning or sandboxes, and validate every change against quality, performance and cost before returning it to production.

CoreWeave Aria, our AI research and iteration agent accelerates that process further by analyzing thousands of evaluation runs, surfacing insights in minutes and recommending the next experiment. That allows customers to compress the time between an idea and experiment and a production improvement at Lightspeed. In Q2, we introduced 7 new AI platform capabilities and achieved multiple industry firsts. These innovations were built alongside our customers and partners to solve real production challenges. That is why we are seeing such strong adoption. And just this week, we surpassed 1 billion model training runs tracked on our platform. Behind that number are millions of experiments, thousands of research breakthroughs, and a growing community of engineers, researchers and organizations building the next generation of AI.

Our AI development services, which carry higher margins, are also being adopted by a broader set of customers than our core cloud. That is proving to be a natural customer expansion path because the developers building AI applications today are the AI cloud infrastructure customers of tomorrow. By serving them early, we are establishing relationships that naturally expand as AI workloads scale. We have seen an explosion of growth in our managed inference platform in the few months since its launch, with growth constrained only by our near-term capacity. Across serverless offerings and dedicated deployments, CoreWeave is monetizing tokens while giving customers flexibility in how they consume our platform. Companies such as Grammarly, and u.com are moving from experimentation to real production traffic, running AI coding agents, serving their own fine-tuned models and deploying open weight models at scale.

Customers shouldn't have to trade speed for cost. And on CoreWeave, they don't. They choose our platform for the combination of total cost of ownership, quality, breadth of service and performance. That is reflected in our consistent leadership across cost per token and speed to first token on leaderboards like artificial analysis for open source models, including KIMI-2.6, K2.7 code, GLM 5.2 and Minimax, M3. And that leadership is converting directly into revenue. In the past few months since its launch, booked ARR for our managed inference platform has grown from 1 million to more than $100 million. We expect to exit 2026 with at least $250 million of managed inference ARR. Pioneers need a different kind of platform. The continuous AI life cycle cannot be supported by simply adding GPUs to a general-purpose cloud. It requires a new approach from power, cooling and rack design through networking, orchestration, observability, developer tools and managed services. That is why CoreWeave is purpose-built for AI.

Our platform is singular in its depth, breadth and technical capability. In Q2, we became the first cloud provider to bring up and validate NVIDIA's Vera Rubin/NVL72, leveraging our innovations in software-defined liquid cooling and rack management to extend our track record of being first to market. We also set new ML perf records for training and inference with open source models, running on the NVIDIA Grace Blackwell platform. and in our tests, achieved the lowest cost per token for inference. However, performance alone is not enough. Customers need enterprise-grade security and observability reliability and compelling economics. According to Signal 65, CoreWeave delivers total cost of ownership estimated to be up to 47% lower than the average hyperscaler.

Customers also require our platform, which integrates these capabilities with a broader portfolio of storage, CPU and networking services across a distributed footprint of data centers globally. In July, Gartner named CoreWeave, a visionary in its 2026 Magic Quadrant for Cloud AI infrastructure. We believe that recognition provides additional independent validation of our approach to building the AI cloud. These achievements are not isolated technical milestones. They translate directly into faster deployment, higher utilization, better application performance and lower cost for customers. Pairing product depth with best-in-class performance, quality and market-leading TCO is a winning formula for our customers and for CoreWeave.

CoreWeave is the foundation for AI at scale. This market requires a foundation at a magnitude unlike anything that came before. That means securing power, sites, cooling, hardware, storage, networking and supply chain inputs well ahead of need and operating them as 1 integrated system. As I shared at the top of the call, we ended Q2 with 1.5 gigawatts of active power, adding close to 500 megawatts in the quarter alone. To put that in perspective, we added more power in Q2 than any single Neo cloud operates in total today, according to third-party estimates.

Critically, our scale is working in our favor, and the math gets better from here. Each new deployment is landing against a much larger installed base than it was even 1 quarter ago. As that base grows, each new build becomes a smaller part of the whole, while contracted revenue from existing deployments remains in place. This is how we are transforming scale into operating leverage. It is why margins expanded in Q2 and why we expect them to continue expanding sequentially during Q3 and Q4. We are also securing the ingredients required to sustain growth over a multiyear horizon. Contracted Power grew to 3.7 gigawatts in Q2, and since quarter end, we have added roughly 500 megawatts, bringing contracted power to 4.2 gigawatts as of today. These figures exclude more than 1.5 gigawatts of further potential power from powered land we have accumulated. options, we have to expand at existing sites and LOIs we have executed.

Our first several self-builds are already well underway. -- including our first site expected to come online later this year. Powered land forms the foundation for deeper vertical integration, giving us greater operational control and supporting enhanced long-term margins. We are also expanding globally and have contracted more than 1 gigawatt of power outside the United States, including recently entering the APAC region with 360 megawatts in Indonesia, that will begin coming online in approximately 18 months. We expect international markets to become a major driver of growth as we meet customers where they and their end users operate. All in, we have excellent visibility to our target of at least 8 gigawatts by 2030. We expect demand to meaningfully exceed supply for years. In that environment, access to power is only part of the equation.

Just as important is each necessary component required to deliver the AI cloud at scale. Building on our close partnerships with NVIDIA and our OEM and ODM partners. Our recent long-term agreement with Solodyn is 1 illustration of how we are derisking access to the critical inputs needed to serve our customers. Our investments in technology, capacity, vertical integration, supply chain and global expansion, all flow from the same vision. AI is increasingly pervasive. The pioneers who move fastest will lead and they will require a platform capable of supporting continuous learning and deployment at unprecedented scale. Before I turn to Nitin, I want to reiterate that CoreWeave enters the second half of the year with more momentum than at any point in our history. AI is reshaping every industry. the pioneers that are building need more than compute, and that is why they come to Core Weave for an AI cloud designed for the full AI life cycle, serving any workload from frontier training to rapidly scaling inference.

Demand continues to exceed supply across sectors, geographies and generations of infrastructure. We are delivering at extraordinary scale to a diverse set of customers with improving operating leverage and visibility into the power and critical components required to sustain growth for years. The opportunity ahead is generational. CoreWeave is the essential cloud for AI. Our conviction in our strategy has never been stronger, and our execution continues to reinforce it. With that, I'll turn it over to Nitin.

Thanks, Mike, and good afternoon, everyone. Q2 was an exceptional quarter for CoreWeave marked by intense customer demand, significant ramp of our active capacity and continued execution against our product and financing road maps. Perhaps most importantly, Q2 marked the quarter in which we saw margins inflect expanding sequentially as we had discussed over the past several quarters. Before diving into results, I wanted to spend a few moments touching upon how demand dynamics are evolving in the current environment as well as its implications on cash flows and the value of our rapidly growing infrastructure footprint. Demand for CoreWeave Cloud remains exceptionally strong across the entirety of our customer base, with demand from multiple customers for each GPU we bring online. We are being disciplined in how we allocate our scarce cloud capacity. We are prioritizing opportunities that are strategically important adding new customers while deepening long-term existing relationships and delivering attractive returns that, as Mike noted, are expanding further.

The scale of our AI products and services beyond GPUs also continues to ramp significantly as customers consolidate spend with us. These margin-accretive businesses including storage, CPU, networking and software already exceed $400 million of ARR as of Q2. We expect they will continue to expand rapidly. Simply put, customer spending on CoreWeave has gone up as customers recognize the increased value we deliver. And this operating margin improvement came before our July pricing changes which included an approximately 25% increase across SKUs in response to the current demand environment and the increasing ROI our customers are observing from their investments in the CoreWeave platform as they shift to inference. We are also passing through component price increases. In terms of how this translates to cash flows, as we previously discussed, a typical 5-year contract carries strong and still expanding unit economics across its term. But those economics do not arrive evenly. The cost, primarily in the form of CapEx is front-loaded requiring a combination of debt, customer prepayments and other corporate level capital to finance its build out. Once the cluster is delivered, contracted revenue ramps becoming predictable and highly cash flow generative. This is all considered in our underwriting of expected margins before a contract is signed.

The deployment delivers attractive returns, fully repaying asset-level debt used to fund the CapEx while generating significant additional free cash flow. So when an initial contract ends, the cluster no longer has any leverage, and we are free to recontract that cloud infrastructure or offer it to the market. We will have generated an attractive return even before the prospect of further monetizing the cloud infrastructure. Every resale or renewal is incremental on top of the returns already earned within the initial term. What we are seeing today is that the upside of recontracting is real as we remain largely sold out of prior generations of NVIDIA GPUs in addition to the current SKUs. So as our earlier generation fleets roll off their original contract, they offer the potential to deliver strong returns in the subsequent years. We are seeing this across our AMP and hopper fleet.

As an example, we recently signed an A100 contract that extends into 2029 at an attractive price. As a reminder, this SKU was introduced in 2020. Clusters of prior generations of architecture of our installed, energized production-grade compute already running at scale. They come with a proven ROI for customers. In a market where new capacity is supply constrained and costs are rising, AI cloud infrastructure and production is a scarce, valuable asset. While we have built a business whose economics do not rely on recontracting after initial customer term. Increasingly, we are seeing longer utilization at higher prices, offering the potential for significant further upside.

With these tailwinds at our back, we are more confident than ever in the long-term ROI of our product and capacity investments, enabled by our industry-leading AI cloud services. Now turning to Q2 results. Revenue was $2.6 billion in Q2, up 112% year-over-year and 24% sequentially, driven by continued strong execution and customer demand for CoreWeave's AI cloud platform. Revenue backlog ended the quarter at $104 billion, up 246% year-over-year. As Mike noted, this does not include the over $25 billion of net new customer commitments we added early in Q3. Of the existing backlog, more than 50% is attached to a contract where customer delivery has commenced. We expect this figure to reach more than 2/3 of our Q2 backlog by the end of this year. Operating expenses in the second quarter were $2.6 billion, including a stock-based compensation expense of $165 million. The increase in our operating expenses was a direct result of scaling our active power while converting backlog into revenue.

This drove the corresponding increases in our cost of revenue and technology and infrastructure spend. In addition, the increase in sales and marketing was driven by increased investment in our go-to-market organization as we further diversify our customer base and expand into new products and markets. G&A increased driven by personnel cost to support our growth while continuing to moderate versus revenue growth. Adjusted EBITDA for Q2 was $1.5 billion compared to $753 million in Q2 of 2025, doubling year-over-year. Our adjusted EBITDA margin was 59%. Adjusted operating income for Q2 was $128 million compared to $200 million in Q2 of 2025 and up from $21 million last quarter, well above the high end of our guidance as operating leverage comes into a business with scale. Adjusted operating margin was 5%. Margins expanded as we scaled despite continuing to incur significant ramp costs.

Net loss for Q2 was $626 million compared to a net loss of $290 million in Q2 of 2025. Interest expense for Q2 was $640 million compared to $267 million in Q2 of 2025, driven by increased debt to support the continued scaling of our infrastructure and delivery of our contracted customer commitments. We recorded an income tax provision despite a net loss due to valuation allowance on net deferred tax assets. As noted last quarter, absent significant discrete items or a change in circumstances, our tax rate should remain broadly consistent over 2026. Adjusted net loss for Q2 was $567 million compared to a net loss of $130 million in Q2 of 2025.

Turning to capital expenditures. CapEx in Q2 totaled $9.4 billion, slightly above the high end of our guided range. Higher CapEx in the quarter reflects customer deliveries accelerating. Construction in progress CIP, increased to $11.9 billion from $9.6 billion quarter-over-quarter signaling the significant amount of PP&E we expect to deploy early in Q3 based on the large amount of power we received very late in Q2. In fact, in June, we brought on more than 300 megawatts of active power, which makes June itself larger than any full quarter in our history. As Mike noted, the global supply chain remains complex. We continue to navigate these challenges with operational discipline and leveraging our partner relationships including new ones like Solodyn to strategically source required inputs.

Turning to our balance sheet and strong liquidity position. As of June 30, we had more than $6.9 billion in cash, cash equivalents, restricted cash and marketable securities. In Q2, we made significant progress in strengthening our balance sheet and expanding the depth and breadth of our access to capital, raising approximately $18 billion across a combination of debt, convertibles and equity. These transactions included several firsts, like our inaugural Eurobond as well as our first ever delayed draw term loan backed by HPC infrastructure issued in the public markets. Our most recent financing, our second publicly syndicated term loan marked another significant milestone as the first to include shorter duration customer contracts. The transaction, which priced during 1 of the most dislocated weeks for credit this year was met with meaningful interest.

Despite the environment at the time, we chose to complete the transaction at its full size. Our spreads have retraced since. Perhaps most importantly, the transaction demonstrated the credit market's growing conviction in long-term value of NVIDIA infrastructure running on CoreWeave cloud. This financing is significant as it unlocks our ability to serve critical part of the enterprise market at scale while also allowing us to accelerate the ramp of our managed inference platform and grow our exposure to shorter dated contracts that typically come at a higher ASP and margins. These transactions attracted broad and deep investor participation highlighting the significant interest we continue to see in supporting CoreWeave's journey. The combination of these transactions brings us to over $32 billion of debt and equity capital secured to date.

Over the past year, we have reduced our weighted average cost of debt by almost 300 basis points, representing approximately $1.1 billion of annualized interest savings based on our end of Q2 debt load. Turning to guidance. As a result of continued strong execution, we now expect to end the year with more than 1.85 gigawatts of active power, up from our previous guidance of more than 1.7 gigawatts. In terms of how this flows through the second half, we expect Q3 revenue to be in the range of $3.45 billion to $3.6 billion. We expect Q3 adjusted operating income of $200 million to $260 million as margins continue to sequentially expand reaching low teens in Q4. Q3 interest expense is expected to be in the range of $860 million to $940 million reflecting the growth in our debt balance to finance our accelerating deployments.

We expect CapEx to be $11.5 billion to $13.5 billion based on the significant amount of new capacity, we will be delivering to customers. Moving on to full year. Our disciplined execution and the momentum we are seeing across our customer base gives us confidence in raising our full year 2026 revenue guidance to $12.4 billion to $13.2 billion and adjusted operating income to $960 million to $1.15 billion. As a result of our increased expectations around capacity to be delivered to customers this year as well as some of our significant recent wins, we now expect 2026 CapEx in the range of $35 million to $39 billion.

Finally, we're also raising our expected end of year annualized run rate revenue to $18.5 billion to $19.5 billion. The long-term nature and attractive margins underpinning our contracted revenue backlog continue to provide us with clear visibility, and we are confident in the targets we have put forward. In closing, Q2 demonstrated the strength of the demand environment for CoreWeave's full technology stack and the discipline of our operating model. We strategically expanded our customer base to support the next wave of enterprise AI adoption at increasingly attractive margins. Customers are expanding their spend with CoreWeave to leverage the full depth of our AI native platform. We remain on track for our sequential margin expansion through the balance of the year and we have made significant additional progress on our capital structure, reducing our weighted average cost of capital while securing the financing required to support our long-term growth plan.

We look forward to seeing many of you at our annual developer conference fully connected in September, where you will hear from our leadership and customers alike in how our platform is accelerating AI in production. Thank you. With that, we will open up for questions.

[Operator Instructions] Your first question comes from the line of Samik Chatterjee with JPMorgan.

Congrats on a strong overall rent here. Maybe just a couple of topics. One, you did mention the renewal opportunity with shorter-term contracts as some of the older contracts come off expiration to leverage sort of the pricing that we are seeing in the market. Can you just help us think through as you engage in some of that discussion with customers, what you're finding in terms of typical customer intent in terms of contract period? And how much of your installed base of equipment is maybe up for renewal over the next few years, if you can get us -- give us a sense of how to think about the magnitude of that opportunity? And I have a follow-up.

Thank you for the question and excited to spend a little bit of time with you talking about what was a truly outstanding quarter for the company across our infrastructure, across our software, across our solutions, across our sales and contracts with new clients and existing clients. Yes, 1 of the most exciting components of what we are beginning to understand on what we believe the market is providing real insight into right now is that the older generations of infrastructure continue to have significant value for use cases within many of the consumers of AI. And we've talked about this literally for years now that the most bleeding edge solutions that are coming out of NVIDIA that we build into our cloud and deliver to our most demanding customers. That's really important for some of the most cutting-edge use cases.

But within that environment within that ecosystem, there are an enormous number of other use cases that can make use of older, more later-dated SKUs. And the fact that we have been able to go ahead and sell a GPU whose architecture was from 2020 in a contract that was fully priced out to 2029, really provides some insight into what the future is going to look like as this infrastructure comes off contract.

Yes. In terms of the capacity that's coming up for renewal, Samik, is a very limited part of our fleet. And the ASPs on the older generation remain higher or at levels that we've seen about a year ago. The second part of the piece that is very interesting in our business is as these fleets come off maturity, it allows us to have a great product in terms of managed inference to serve for our customers, which, as Mike noted, is a very fast-evolving nature of our business. part of our business, which we expect to continue to grow rapidly and expect to have about $250 million of ARR by the end of the year.

Got it. And so my follow-up there. But in terms of the follow-up, can you talk about the supply chain a bit. You're obviously navigating it pretty well to bring capacity online. But in terms of the agreement that you have now with Solodyn, for example, how are you looking at sort of the need to maybe do something more broad-based across supply chain in terms of longer-term agreements to assure yourselves of more supply as well so that you can continue to sort of execute on the power -- the capacity that you want to bring online?

Yes, it's a great question. So look, End of the day, my job is to ensure that this company has the capacity to deliver the product that our clients require. And in order to do that, we need to aggressively manage a complicated supply chain. And that supply chain includes everything from land power and shell through GPUs and networking through memory, all of which is being challenged by the growth and expansion of artificial intelligence. In order to do that, we have built over the last several years, really long-standing deep relationships with our OEMs, our OEMs, NVIDIA the companies that surprise us with memory, all of them. And what we've done is we've thought about what is necessary to ensure that we have access to the infrastructure and the components and the capital that we need in order to deliver our products at an acceptable price and quality to our clients.

And it's one of the things that's just embedded in the DNA of CoreWeave. That's what we do. It's part of what we do every single day is nurture these relationships and ensure that we have access to everything that we need in order to deliver the product, which is NVIDIA infrastructure delivered through our cloud.

One thing, Samik, to note here is the increase in the value of output of the core bed cloud has outpaced the value of the input increases that we are currently experiencing in the supply chain. And as a result of it, what you're seeing is margins expand, as Mike noted, in his comments around the typical contribution margins that we saw last quarter we are 5 to 10 percentage points higher than what we've observed in the recent quarters.

Your next question comes from the line of Brad Zelnick with Deutsche Bank. Please go ahead.

Great. Congrats on the strong execution. My first question, I wanted to ask about your managed inference offering, which is off to a really strong start. What are your initial learnings? And what are the factors that inform your thinking on allocating capacity to it going forward versus your traditional take-or-pay deals? And I have a follow-up to that as well.

Yes. Thank you for the complement. It really was a great quarter for us. We're very excited about it. Look, when we think about our offering, we really think about it holistically and we have made enormous strides through the last several years to focus on building scale through these long-term take-or-pay contracts. As we've hit hyperscale, we understand that we are going to need to broaden our offering to provide the products that our clients need that to deliver products that have higher margins to provide the software solutions to provide access to CPUs, all of the things that are necessary for our clients to be successful. And when we think about the lessons that we've learned, as we've gone through this unbelievable and unique scaling of our managed inference product, which went from $1 million to $100 million inside of a single quarter -- we really think about the fact that, that is an incredible opportunity for us to offer the most bleeding edge compute that we have, but also a wonderful way for us to access and use contract that are coming off contract in a way to extract maximum value for the company over time.

So look, the market is very deep. We think that we have an embedded advantage because of our control over the silicon, and we think that we're going to be very successful in that market over time.

And Brad, 1 thing to note here is we announced yesterday around our DDTL 5.5 closing and that shows that the capital markets are extremely interested in supportive of Core's product in terms of underwriting shorter duration contracts, which is definitely a tailwind as we look at these markets to support our customer needs.

That actually leads to my next question. So on the 5% to 10% better margin that you're seeing on recent deals that you're signing, can you help unpack the drivers how much is a function of shorter duration deals, versus strong competitive differentiation or other factors? And what are you seeing more broadly just out there in the market as it relates to pricing?

So it's a combination of a lot of things, and it's difficult to deconstruct it. The infrastructure that we deliver through the core weed cloud is more valuable to our customers than any other solution that they can encounter. The quality of the platform, the reliability of the infrastructure, the security the TCO, all of those things contribute to customers coming back to us again and again and expanding their footprint within our cloud and infrastructure. And so there is a piece of it, which is they understand how much more valuable a given piece of infrastructure is delivered through us. The second piece of it is many of our clients are monetizing their products. And so they are more aggressive about coming in and willing to pay us higher margins because they need access to the compute that will allow them to be successful.

This is a phenomenon that's occurring across the infrastructure space but it's particularly occurring within our ecosystem. And it's very exciting to see as the premium product that we deliver is priced in a premium fashion by the consumers of this compute.

Your next question comes from the line of Amit Daryanani with Evercore ISI.

This is Irvin Liu on for Amit. I had 1 and a follow-up. So my first question is, it sounds like there's upward pressure to pricing across multiple vectors. -- including the higher value you provide to your customers, the pass-through of higher component costs and the recontracting opportunity coming up. So -- with that in mind, should we still think of kind of the $18 billion to $19 billion in ARR as kind of the exit target for 2027?

Yes. So we increased the exit ARR number that we provided in guidance to you folks right now at $18.5 billion to $19.5 billion for 2026, -- so that is baked in our guidance that we provided to you.

Okay. Got it. Got it. And then for my follow-up, I think the regulatory backdrop for data centers appears to be increasingly difficult. There have been reports of local opposition to data centers. With this in mind, can you talk about your confidence level in deploying more than 3 gigawatts of active power by the end of next year and kind of your road map to 8 gigawatts by the end of the decade?

Sure. And your question is very timely and very important for the entire AI space in the entire data center space. I guess I'll start with -- we believe that the certain communities have moved forward with moratoriums. And we feel like moratoriums are they're not going to impact the demand for this infrastructure. They are going to impact where this infrastructure gets built. And so our approach to how you engage with the stakeholders is that you have to be extremely collaborative with the communities that ultimately host the infrastructure. And that's based on transparency. You have to work with the local governments. You've got to work with the utilities, you've got to work with the policymakers in order to allow yourself to ensure that what you're building fits into the communities that you're entering.

Ultimately, at the end of the day, it is in our interest, and it is in their interest for us to be good neighbors of their community. A lot of that comes down to making sure that you're paying for grid upgrades so that it doesn't fall or impact the rate base. You create an enormous number of construction jobs. You are -- there are long-term jobs that are left that survive within the data centers, the data centers that are being built contributed to the tax base. All of these things are incredibly important to how you enter into a community and how you engage that community as you're building this infrastructure that is so necessary in order to be able to provide America's AI leadership. And so when we talk through the numbers with you guys, we -- basing our progress on where we are today and what we have guided here, none of those numbers will be impacted by the regulatory pushback as of today.

We are comfortable with it. We continue to expand. We continue to engage our data centers are best-in-class, and we expect to be held to that as we continue to build our infrastructure across the globe.

And Irwin, just to give you some numbers in perspective here, if you look at our gigawatts contracted today, they're already at 4.2 gigawatts contracted. In addition, we have about 1.5 of powered land options to execute LOIs that we have executed. That puts you close to about 6 gigawatts already in terms of how we think about power and it's middle of 2026. So we remain well on track to execute against our stated goal of greater than 8 gigawatts of active power by end of 2030.

Your next question comes from the line of Raimo Lenschow with Barclays.

I just -- I wanted to talk a little bit about the growing importance of inference for you guys. How does your fleet need to evolve? Because inference needs to do a lot more CPU, a lot more storage -- can you do that in the existing data centers? Do they need to involve? Can you speak to that as well to make sure we have the capacity there going forward?

Yes. It's a great question. It's a question we've been talking about now for several quarters. We believe that when you're building infrastructure, you don't build infrastructure for training, and you don't build infrastructure for inference. You build AI infrastructure. And when you build AI infrastructure, you need to ensure that you have all of the components to be able to serve the full AI loop, everything from training through inference as it cycles back and forth as it moves through the iterations that are required in order to serve your clients and those companies that are consuming this. And so really, the infrastructure that we built will move seamlessly into the ability to serve inference over time.

Okay. Perfect. And then 1 follow-up is like, obviously, with the news from Meta yesterday, a lot of questions that we face today was around doing AI in the edge, et cetera, and then all these concerns came up again. Like can you talk about like how you see the market evolving between edge, smaller clouds, neo cloud and hyperscalers.

Yes. One of the things about CoreWeave that should never be underestimated is we sit at the epicenter of an incredible amount of information flow from across the entire industry, right? The hyperscalers use us, the labs use us. You've seen enterprise begin to scale within our platform. The information flow that's coming back and feeding us the clues to how the world is going to look in the future has been incredibly powerful for us in terms of how we position ourselves and our compute to serve our clients. Look, at the end of the day, we believe that there are workloads that are going to be served from the edge and there are workloads that do not require the same level of latency protection. And we have built our cloud to be able to serve both of those constituents effectively. And we will continue to build in that fashion. We will be informed by our clients continuously whether they need a little bit more of edge, they need a little bit more of scale that is not as latency sensitive. All of those things are being fed to us on a continual basis.

And so yes, we do see workloads on the edge. And yes, we do see workloads that don't require to be on the edge. -- and we are very, very comfortable that the scale of our infrastructure and the ability to move it back and forth is going to provide a competitive advantage for CoreWeave over time.

Raimo, to your point in terms of increased competition. Even with this increased competition, we are seeing demand, pricing and margin all expanding which is a signal for the growth in the CoreWeave kind of product as well as our growth overall in an already massive TAM that exists.

Your next question comes from the line of Michael Turrin with Wells Fargo Securities.

I realize there's likely some rounding here, but you added an impressive 500 megawatts of active power in the quarter. The revenue, if we're looking at the sequential adds is fairly consistent with last quarter. We can hear all the commentary around the uplift that's coming. So maybe help us think through the linearity of capacity added. And when that 300 megawatts added in June started to hit more of a steady state in terms of model contribution and also would be useful as a second part to hear any early market signals you're gathering on Vera Rubin monetization and what the uplift there could look like versus prior generations?

Yes. So as you mentioned, right, you saw Q3, we added about -- sorry, Q2, we added about 500 megawatts of power. 300 of that alone was in the month of June, which is higher than any amount of power that we've added in any histo prior quarter for CoreWeave. So definitely, that power was back-end loaded in terms of Q2, which you would start seeing kind of come through in Q3 and Q4 in our business.

Yes. So let me -- maybe I'll take a moment to speak to Vera Rubin. Vera Rubin is a generation that is seeing the margin expansion right from the start. And so it's really exciting for us. The demand for the Vera Rubin platform is enormous and the pricing power that CoreWeave has been able to garner with its CoreWeave cloud, delivering that infrastructure really bodes well. And when we were talking about that 5% to 10% margin step function that we're seeing, a lot of that is coming in, in the Vera Rubin SKU. We're excited about where that's going to lead. We think that it's going to be a very, very successful SKU for CoreWeave and CoreWeave's clients.

Your next question comes from the line of Brett Knoblauch with Cantor Fitzgerald.

Congrats on the very strong quarter. Mike, I guess, just kind of based on the prepared remarks, it looks like the price environment has never really been better for older generation and obviously, newer generation GPUs here. As you look at the GPU fleet that's maybe rolling off contract, can you talk about the cadence of how you guys look to either recontract that or kind of put it on spot or in your inference products? And how far in advance of the roll off of those contracts, would you look to kind of make that decision?

Yes. So look, it's a good question. It's 1 we're working through. Keep in mind that the environment for inference is incredibly dynamic, and it is scaling so fast, as we kind of struggle to keep up with the build-out of new infrastructure, the flexibility that we are giving because we have infrastructure coming off-line allows us to continue to scale the inference product as we're continuing to explore exactly how big, how extensive is the managed inference opportunity for us. Some of the infrastructure that comes off-line, we go ahead and we place back into a term contract. If we think the economics warrant putting it in.

And the economics include both the term that we're able to garner as we think about the long-term stability of the company and the long-term obligations that we need to support as we continue to build and scale the company. But we also do recognize that in the short term, there is an opportunity to sell on shorter-term contracts and extract additional margin on this infrastructure as the world tries to catch up with what is a systemic disequilibrium that has really existed for several years now and will continue to exist for the foreseeable future.

I think that kind of leads to my next question. I guess, off the back of DTL where you were able to kind of get funding for shorter duration contracts combined with kind of this data center ibis political at that's kind of taking off. It feels like you guys should be quite well positioned given you are the most scaled to realize the most price benefits. How does the success of the TL 5.5 change the way you view on kind of target durations on a go-forward basis? Does -- is that an avenue you want to use more to maybe extract more margin in shorter duration contracts, given you know the useful life is there?

Yes. I mean, you're exactly right. The execution of the DDTL puts CoreWeave in a position where we get to populate the curve in terms of what we think is the most profitable configurations for term leasing. And so we want to sell our compute on long-term contracts. We also want to sell it on shorter-term contracts to extract additional margin. And we have been really, really aggressive about doing that. We were the first ones to bring the 5.5 to market in order to be able to really plug into those. There's one more really important part of the short-term contracts that I think it's important for everyone to understand. When you're thinking about enterprise, enterprise tends to want to enter into contracts that are not 5 years in length. They tend to think in shorter cycles than that. And so by enabling the financing market to support the contracts in 5.5, we're able to go ahead and diversify our terms so that we're able to support additional contracts. It opens up whole new markets for us.

These clients want to buy compute for 2 years or 3 years -- and that's not a market that was easily accessible to us until we were able to bring DDTL 5.5 to market. And now that market will accelerate meaningfully as we're able to offer compute to our customers on a time frame that they are able to consume it, buy it and entering the contracts to purchase it from us.

This concludes our question-and-answer session. I will now turn the call back to Mike Intrator for closing remarks.

Before we sign off, I want to thank our team customers and partners for their trust, hard work and commitment to Core weave. None of these achievements would have been possible without you. I'm incredibly proud of what we have accomplished together and for what comes next as we build the essential cloud for AI. Thank you all for joining today. We appreciate your support, and we look forward to updating you on our progress in the quarters to come.

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

I'll start. I was asked to present myself. My name is Tal Liani, and I'm the analyst that covers CoreWeave. Here we go. It's for the transcript.

So Nick, thank you so much for joining us. Your stock has been terrific, and we have a few questions about your position in the market, your longevity. I -- because this is the first time I interview you in this kind of setup, just I want to start with 30 seconds what we think about the space and your company, and then we'll go to our Q&A. That's what I'm doing in all these kind of sessions.

So we recently launched with a buy rating on all 4 data center companies. We call them the GPU companies, CoreWeave, Nebius, Oracle and Microsoft. And the reason why we had a buy is -- on all 4 names is because we have a terrific cycle. This cycle, I don't see -- at least at this point of time, I don't see the sign of slowdown. We're going to talk about it. We're going to ask about it. Hopefully, this thing is okay.

Here we go. And what I want to focus on in this discussion, I want to focus on the differentiation of CoreWeave. I want to focus on the value you bring to market. I want to focus on the longevity of -- and visibility of orders and things like that.

And the first question I have is what makes you different? What is the difference between the way you are structured and the way that hyperscaler is structured from a conceptual point of view?

Absolutely. And I think what makes us stand out, there are a few things that are absolutely true and then there are a few things that are relatively true, right? And it's, okay, are you comparing us to Microsoft? Or are you comparing us to a smaller neocloud, right?

In the context of comparison to hyperscalers, it's the technology stack. It's fundamentally what we did on a first-principles basis is rearchitect the way that cloud stack is built. And the reason we did so is AI cloud is based on a different type of workload. It's based on parallelized compute, which is quite a bit different than the way CPU workloads historically were based, which is serialized compute.

Serialized compute is -- the concept is built for redundancy. The workload is small enough, it's going to be run on a CPU over here. And if that CPU breaks, I'll run it over here. And so I'll be able to serve the problem no matter what. Parallelized compute is the workloads are really big. And so actually, what you're going to do is you're going to have a bunch of GPUs working together on a single problem. The challenge with that is when the GPU breaks over here, it can't be replaced over here. The entire system is down, right?

And we built the technology stack geared towards optimizing a few different things, which are, one, the efficiency of the overall ecosystem, right? We effectively ripped out the virtualization layer that lives in a hyperscaler's cloud stack. But on top of that, we built a proprietary orchestration layer that's gotten uniquely good at provisioning GPUs of understanding are they healthy or not, of predicting when they might not be healthy and ensuring the workload remains safe.

And you combine those 2 things and you end up with a cloud stack that is more performant. And what that allows you to do is, one, be kind of the cloud partner of choice for virtually every sophisticated user in this ecosystem. And I think we're the only independent cloud company that does service both OpenAI and Anthropic and Meta and Google and Microsoft and NVIDIA and the next layer down, whether you want to call that a Cursor and a Cognition and Perplexity or a Cohere and Mistral, right? Like we're pretty singular in that capacity because of the unique quality of what we can deliver. And what it also allows you to do economically is ultimately charge a higher price per GPU hour while still delivering a lower TCO to your customer because you are delivering a much more efficient and performant product.

When I first -- I interviewed all the companies because I had to launch coverage and I spoke with Microsoft, the first thing they told me when we spoke about the space was this is a temporary solution. When we build enough data center capacity, we'll bring all the capacity in-house. What's the risk of you being a temporary solution versus a permanent solution?

I think highly limited. And I think it's highly limited for a litany of different reasons. One, I think that the hyperscalers were the first large consumers of this infrastructure, but that is not the steady state of this. Already, you have OpenAI and Anthropic rivaling just how much compute they want to consume relative to a Microsoft. Already, you have a long tail of enterprise customers looking to consume directly, right?

Already, you are seeing like even our reliance, like when we went public, Microsoft was 85% of our revenue backlog, right? Today, they're not even our largest customer, right? They're not even our second largest customer, right? And so the natural diversification that has come into this industry has diversified away that risk. On top of that, the way I kind of think about the renewal problem because I've been getting this question, particularly with regard to Microsoft since 2023, right?

And I got the question as an adviser to the company. I got the question as an investor of the company. And now I get the question as an employee of the company, right? And the way I look at it as I try to study the history of the cloud in general. And 2 observations I've made that no one has really pushed back on yet are, one, I am not aware of any point in time in the history of the cloud when a hyperscaler has chosen to actively reduce their data center footprint of revenue-generating data centers. To not renew is to shrink.

That seems to defy the history of the cloud. The other part of it is using the CPU cloud as the analog. If you look at the Azure portfolio of data centers for CPU, a lot of it is owned and some of it is leased, right? I think the steady state of GPU cloud or AI cloud looks like a lot of it is owned and some of it is leased. The core difference being when you were building CPU cloud, about 2/3 of your cost was the shell itself and about 1/3 of the cost was what goes in the shell.

When you build AI cloud, it's the inverse, right? 2/3 of the CapEx goes towards what goes inside and about 1/3 is the shell itself. So I think it stands to reason that for the portion of a hyperscaler book that is leased that they probably are going to want to not only avoid the CapEx of the shell, but avoid the CapEx of what goes inside given it's twice as expensive. So you put that all together, and frankly, that is just not -- that's not the thing that keeps us up at night.

Got it. Let's talk about visibility. Spending CapEx growth has been phenomenal for the last 2 years, including this year. How confident you are that this spending cycle continues? And I know we don't have the answer. I'm trying to get to your thinking process, meaning what are the drivers?

sure. So I do come back to what are the drivers and how we think about it or maybe even slightly different things. What are the drivers? It's adoption and productivity expansion, right? And you're seeing so much of it happen this year right now, right? If you look at the growth of the Anthropics of the world, right, and the OpenAIs of the world, like you are seeing enterprise adopt, you are seeing AI diffuse. We are seeing it show up in our pipeline with a long tail of enterprise customers who want to consume this technology directly as opposed to indirectly.

The how we think about it is directly informed by the pipeline, right? In so many ways, our business benefits from a flywheel that starts with the fact that we uniquely serve and are the trusted engineering partner to virtually all of the sophisticated users and consumers of this technology in the world. We understand where they want this to go, and we build towards it. The anecdotes I would give you are back in 2023 and early '24, we were building with InfiniBand.

And a lot of people were saying, why are you spending all this money on InfiniBand? All the hyperscalers just build with Ethernet and they say it's cheaper and they say it's just as good, right? And models are going to get quantized down to a single GPU or a fraction of a GPU. And what do you need all this dense network fabric for? And then what happens in September of 2024? O1 comes out. And all of a sudden, you have inference being run not on a single GPU or even a single node, but across nodes. And you need a denser network fabric to run that efficiently, right?

And all of a sudden, building so much with InfiniBand almost seemed clairvoyant, right? Like I would say a similar thing about focusing our kind of efforts on procuring liquid cool data center capacity back in 2024, right, building ahead of GB coming out in 2025. We've made very similar investments and bets in the portfolio across things like storage and CPU to position ourselves to take advantage of those tailwinds over this course of this year and next.

And I would say a similar -- like we're doing that because we know what the customers are going to need. We are working closely with them, and they are telling us because they want more from us because of the quality of what we deliver. And what they are telling us is this is not going to slow down. And frankly, they're telling the world that, right? Like in 2023 and 2024, even for a lot of last year, it seemed almost contrarian to believe that this cycle was going to continue and in orders of magnitude larger, even though Jensen and Elon Musk and Sam Altman and Dario Amodei and Satya Nadella and Sundar Pichai and all these people were telling you it would, right? Like just believe them, right? Like they're telling the truth.

And like what is a better data point of that than Google raising $40 billion yesterday, right? And like who had on their bingo card that Google is going to be raising equity securities this year in tens of billions of dollars of scale. I think that there's no one. Why are they doing that? It's not because they intend on slowing down anytime soon.

Got it. Another question we're getting a lot is about understanding the business model, meaning people ask me about unit economics. And so -- but I'll start with a kind of high-level question. What's your business model? How -- when you sign up a customer, how do you think about the first period, renewal, other customers? What value can you extract from a GPU?

Yes. And so I'll speak to the core business model and how it's evolving over time. The core business model, and you got to think about where we got started is you sign longer-dated contracts with customers where they're contractually bound to pay you a fixed price per GPU hour regardless if they use it or not, it's called the take-or-pay contract for the next 4 to 6 years.

Why did we start there? We started there because to build a cloud business is equal parts technology and infrastructure. Scaling technology often comes without capital intensity. Scaling infrastructure never comes without capital intensity. And so to build a hyperscale business, which is our aspiration, and arguably, we're there already. We're more than 1 gigawatt of active power at this point in time, right?

But to do that, part of the name of the game is you want to be able to have as much access to capital as possible at the cheapest price imaginable, right? And signing these longer-dated take-or-pay contracts, taking them to the asset-level financing market has been a way where we could borrow capital at costs that are way closer to our customers than our own, right? And that is how you build scale as quickly as we have.

I would say we are the only company in the world in this ecosystem that has built the scale that we have as quickly as we have. No one is even close, right, who didn't have an investment-grade balance sheet to begin with, right? Like we are singular in that world. And so you got to start, right, with that core foundation with the beauty of it being, hey, a 5-year contract is going to pay for all the financing costs and all the CapEx associated with standing up that cluster. It's going to cover all of the OpEx during the life of that cluster, and it's going to pay for another 5 years of data center expense on top of that.

And so what you're buying yourselves is -- or what you're positioning yourselves for is 4 years from now, 5 years from that initial contract, you're going to own infrastructure that is your own to monetize. And every dollar you get out of it is just cash flow accretive to what you paid initially. And being in control of that massive infrastructure that is of critical importance is incredibly valuable.

So that's been the foundation for a while. It will continue to be, right? But what we've been able to do as we've gotten bigger, right, and access to capital has gone up and cost of capital has gone down is we've been able to position the portfolio, right, to include some shorter-dated contracts, too. And we like shorter-dated contracts, right, in that what it exposes you to is a higher margin, a higher ASP, right?

If you're committed for fewer years, you're willing to pay a higher price. And what it allows you to do is to take advantage of the increasing demand for this technology such that, hey, a contract 2 or 3 years or a piece of [indiscernible] 2 or 3 years from now might be more valuable and you might get a higher price than what you're charging today. That's been our experience with Hoppers, right? We're selling Hoppers today at higher prices than we were 3 years ago. And the reality is no non-investment-grade business can build a hyperscaler of on-demand product without diluting their shareholders by like 80% or 90%, right?

Like you just can't do that, right? But as you get bigger, you get to twist the dial a little bit and that positions us to better take advantage of repricing existing infrastructure of selling more on spot over time. And that is why in Q1, we announced our spot product, right? Like we are getting to the point where we are hitting escape velocity and we're able to take advantage of those market dynamics better at scale.

Yes. What happens to the value -- what happens to the GP -- so there is an initial -- I'm talking about the big contracts. There's an initial contract, let's say, 5 years. What happened -- what is your margin during that time? And what happens after that with the residual value?

Sure. So the margins in that time for that base contract are mid-20s contribution margin. And we consistently underwrite that for new infrastructure for that 5-ish year deal, right? If it's a shorter deal, that has higher contribution margin. What's the opportunity after that? The opportunity after that is you go sell it in the on-demand market. You go sell it at spot, which when you're signing a 5-year commitment, right, you are signing at a price that is lower than on-demand, right, because you are paying for -- or you're selling 100% utilization for every second of every hour of every day of every month of every year for 5 years, right? But what you can start to do is take advantage, right, of, okay, this is paid for. I don't need to go finance it. And what that allows me to do is go sell it at spot where pricing might be higher.

The fact that the technology at that point is going to be 5 years old, does it mean that you have to find new types of customers or...?

I think it's more likely that you find new types of workloads than new types of customers, right? Like the -- what we can observe, right, is Ampere and Hopper pricing has gone up pretty consistently, right, over the past few months and maybe even a bit longer than that. And I think it's highly unlikely that Ampere, which for us went up in Q4, ASPs went up and in Q1, they went up again, right?

I think it is unlikely that a bunch of people are contracting Amperes, which is at this point now rapidly approaching 6 years old, right? I think it's a late 2020 SKU for training, right? That is we have workloads that run very well for this -- from an inference perspective, and we are making really attractive returns by buying this. So we are willing to pay more for it. And so I think the customer might evolve. The customer might be the exact same. They might just match a different workload to it.

Like I almost think of OpenAI's router model, right? Like people hopefully haven't already forgotten that like 6 or 9 months ago, you got to pick your OpenAI model, right? You can be like I want o1 or I want o3 or I want GPT-4. And I was the person who was like, okay, I want like the most performing model for everything. It doesn't matter what the query was. And that was probably an irresponsible use of compute, right?

But what they did is they introduced a router model where they said, okay, based on the query, I'm going to map this to GPT-3 that might be running on Ampere or o4, which might be running on a B200 or GB200. I think you will also see more of that where customers get more sophisticated about, okay, this workload goes here, that workload goes there.

Got it. I always tell my fiancee, don't thank ChatGPT, you're just burning tokens. There's no need...

I remember when Sam Altman [indiscernible] but when the machines take over, you want to be nice to them.

You're accounting -- you're depreciating your assets for 6 years, the GPUs for 6 years. Will they survive 6 years?

I think every data point that we can observe in the market suggests that the answer to that is, if anything, we're being conservative, right? Like the analogs we can point to are Voltas and Teslas, right, like older GPU SKUs are still running in clouds. Those are late 2010 SKUs that are still being monetized today, right? I imagine a similar thing could be said of TPUs that are 6-plus years old, right?

And again, we're looking at Amperes, like we're getting to 6-year-old SKUs and those things are humming, right? And so I think it feels to me like this debate part of it has maybe waned a little bit in the last few months as people have seen Hopper pricing be higher today than 3 years ago, right? They're like, oh, I guess it wasn't a 3-year useful life. I think it will continue to wane. But like a fear of the unknown is definitionally unknowable until you get there.

And Hopper is a late '22 SKU. So we'll see 2 years from now. But everything we see suggests that they will be monetizable. And what I would say if there is true risk from a hardware perspective of will it perform, I feel way, way, way better about Hoppers running in CoreWeave cloud than any other cloud in the world, right? Because what our orchestration layer does, what Mission Control does in large part, is it keeps GPUs healthy. And the healthier that you keep the thing, it's more likely that it's going to run for longer, right?

Got it. Yesterday, I hosted for a keynote, the founder of a company called TECfusions, and they build data centers. They have -- it's a real estate with power company. And he said, half of the companies that tell you they're going to build data centers, they're not going to make it on the time. And the question I'm asking you is your backlog had grown up tremendously, your revenues, the outlook is great. Talk about the operational risk, talk about the operational challenges in bringing capacity online to meet your liabilities or your commitments.

So CoreWeave exists because we are excellent at 3 things. Like -- and you need all 3 to build the business we have in the time period we have. We deliver excellent technology, most performant cloud out there, right? And that's -- ask our customers, ask experts. I think that's the consistent feedback. We are excellent at scaling this infrastructure and delivering cloud, right? And we are excellent at navigating the capital markets to permit us to do so, right?

You can't exist and build a business from zero to hyperscale without being excellent at those 3 things. We feel exceptionally good about our ability to deliver on the time lines that we've agreed to with customers. I think our track record is, I would argue, wildly underappreciated in the market. And what I mean by that is we have close to 50 data centers online. The overwhelming majority of them have been on time, some have been early, a few have been late.

We did get on our Q3 earnings call, right, and say, look, this is an industry-wide thing. We have one data center development partner who's struggling more than others, and they're delayed, and that's impacting our Q4. It was not an overwhelming impact to Q4, right? But it was an overwhelming reaction from the market because I don't know that they appreciated that things are going to be delayed.

I think that what is unique and my guess about this is in terms of what's going to happen. There are 6 companies in the Western world who have delivered AI cloud at scale, right? It's Microsoft, it's Meta, it's Google, it's Amazon, it's Oracle and it's us. right? I'm not saying there aren't other people who are signed up to do it or who may do it in the future, but the reality is those are the 6 companies that do it. And 5 of those companies have gigantic other beautiful businesses that obscure away the economics and what's actually going on quarter-to-quarter of AI cloud, right?

And then there's us who we don't have those businesses to obscure it away. There are going to be more companies, neoclouds that are scaling real size of infrastructure in the coming quarters. And they're going to try to do it multiple times, which is something we've done. Like I said, we have close to 50 data centers as of the end of Q1. I think 49 was the number. I think the world is going to see more delays and more struggles and the world is going to think that things have gotten worse, not better, whereas I don't think that is true.

I think you are just going to see more people attempt to do it who can't obscure things away. But do I think working through operational challenges is something that is part of this job? I think it absolutely is. And why do I feel that we're well positioned to do it? Well, one, we've done it a bunch of times. We have an excellent track record in this regard. And two, one of our real superpowers is we are able to take a power shell and turn it into a supercomputer that's part of AI cloud in something like 6 weeks.

It takes most guys 3 to 6 months to do that. And so when you're able to do things as efficiently as we have and by the way, we think we're going to keep getting better there, we are able to offset some of those challenges that other people face in operational delays, oh, something is -- we were planning it would take 3 months to deliver this. Well, oh, we only need 6 weeks. So if you're 2 weeks late, we can still be early, right? I think that is the biggest part of it.

Got it. And what about supply constraints? How do you manage the fact that cost supply -- component cost is going up constantly?

We pass it through to customers, right? The reality is you do that in 2 ways. The overwhelming majority of our CapEx is spent on servers, where we're signing purchase orders with our partners. At the same time, we're signing order forms from customers. And so we're able to say, Oh, the pricing just went up. Great. We have to charge more for it, and then we lock it in.

That's before you sign the contract, at the time of signing contract...

At the time. Think of those things as concurrent. And then for smaller parts of the business where you have a bit of exposure like storage, I would think of some of our pricing mechanisms as a bit more like cost-plus oriented where there can be an escalator in price if there's an escalator in cost.

Got it. Got it. Enterprise customer. So enterprise and other -- you spoke about other types of customers. Talk about your efforts to go after the inferencing opportunity of enterprises. Some of the other -- another neocloud company, much smaller cloud company, they make it their vision. They only focus on the enterprise. What about you?

I think there's difference between making it your vision and only focusing on it, right? I actually think those are 2 disparate things, right? I think to -- I don't know that there's actually any neocloud out there that actually only focuses on the enterprise, right? Because if you actually look at what -- who's paying them revenue, I think it's not even research labs, it's hyperscalers.

I'll stop you for a second because I'll define it better. They use hyperscalers in order to fund the build-out, but they say we take all -- once we build it out for Oracle, for whatever, once we build it out, we're going to shift it and only address the enterprise. They really only want to focus on enterprise.

So what I think I'm hearing from you is you're saying they're going to take a longer-dated contract, use it to pay off the infrastructure.

Exactly.

Then take that infrastructure and sell it to other people because it will be entirely theirs.

That's their strategy.

I would say that sounds like the business model that we brought to market in 2023 and defined, and it seems like a lot of people have adopted it.

What about go-to-market? The other part?

So we've invested pretty heavily in building out that muscle, right? We hired Jon Jones last year, who is a senior go-to-market leader at AWS, who is now our first CRO, and he's been building out that part of the sales force. It is of extreme focus to us. And I think the way we think about how we allocate capacity because it really is an allocation conversation is one of you want to -- like hyperscalers are really nice in that they also give you a prepayment, right, a lot of the time, and that gives you a more effective way of financing.

We want exposure to research labs, right? Like we want to be a trusted partner to OpenAI and Anthropic and Meta, right, and the next tier of the list. But we also want to be partnering with the enterprises. And we've had real success there. In Q4, we announced folks like Mercado Libre as customers. In Q1, we announced that 10% of our $100 billion revenue backlog was financial services companies, right? That's close to $10 billion of enterprise within a single vertical, right? That's not accounting our success in industrials and health care, et cetera. And I think you will see us continue to add new logos and continue to allocate capacity to those enterprise customers.

Great. I used up most of the time. Is there any question from the audience? Raise your hand. Yes. Instead of waiting for the mic, just shout it out.

[indiscernible]

So when you talk about performance, I guess, let's take a step back and maybe I'll hit on 2 things. One, you're absolutely right. I think what you will see in the agentic era is more use -- the attach rate or the ratio of the CPU to GPU will go up. The necessity for storage, right, and keeping data close to the GPUs will go up, right? And I think what you've heard from us, right, in Q3, we started talking about our storage business, how it eclipsed $100 million of ARR, and it was growing like a weed, and it was a business we're really excited about.

And then similarly, we started talking about our CPU business in January. And obviously, I think that today, the world has a better appreciation of why we started talking about it in January. When it comes from a performance perspective, what you are doing is making a complex system more complex. You're making it bigger, you're adding more directional data flow and communication. The bigger clusters get, the more complex workloads get, the more differentiated our software stack proves to be. That was true in training. That's been proven true in inference as well. And I think that feeds into our competitive advantage and differentiation over time.

Great. Is there a risk -- last question, but is there a risk of capacity commoditization over time?

Yes. But it's a question of when. Is there a risk of that this decade? I do not think so, right? But to be clear, right, this platform -- like we've been the guys since 2023 when I first met Brian and Brannin and Mike and Peter, they were saying this was a rest of the decade problem. And a lot of hyperscalers are saying this was a 6-month problem, right? In the same way they said, "Oh, neocloud short-term thing" and then they go out and sign $80 billion of more neocloud stuff, right?

But our view has always been what is your right to survive and thrive as a cloud participant, right, in a world in which you do hit supply-demand equilibrium. And our simple thesis has been this is what you need to achieve. Hyperscale, I think we're well on our way. Competitive cost of capital, I think we're well on our way, right? We've cut our cost of debt by close to 700 basis points since the beginning of 2024. And you need to deliver interesting technology, I think we're already there, right? And if you can do all those 3 things, then you will have a right to compete and an ability to compete from a cost of capital perspective in a more balanced world.

Got it. Great. Thank you, Nick. We ran out of time. We could have continued another hour. Thank you so much.

Thanks so much, everyone.

Welcome, everybody. Thank you for joining. Brannin, you know how to draw a crowd. For those who can't see, there's a lot of people standing here. So I'll give all that credit to you.

Quite an extraordinary 14 months for the team, IPO-ed last year. Full year FY '25 revenue, $5.1 billion, 2 weeks ago, nearly $100 billion in backlog, $40 billion new bookings in a single quarter. So for the 1 or 2 people who don't already know you, could you introduce yourself, what you do as Co-Founder and Chief Development Officer, a little bit about CoreWeave.

Yes, absolutely. Thank you. Brannin McBee, CDO, one of the -- CDO, one of the co-founders. I lead all things capital markets for the business. So that's capital formation, M&A, venture investing, basically, whenever you see one of these GDPL deals convert high yield, that's my team leading those processes. We're -- our responsibility is keeping the business funded and being able to grow at the pace of AI.

Yes. Yes. We'll hit on one of the items in the news for you guys, which is that Blackstone Google deal. So anything you want to rip on there, let investors kind of noodle on?

Yes. Yes. Look, I think first and foremost, it's just yet another demand signal that's out there. I don't think that this room really needs more signals and more conviction that there is overwhelming and insatiable demand for AI, but that's our takeaway that's there. I think as everyone knows as well, this is a TPU cloud that's being built. I mean, to me, it makes sense that Google is going to want to empower others to go build TPU cloud, much as NVIDIA has empowered others to go build GPU clouds. Our business is not TPUs. Our clients come to us asking us to build GPUs and explicitly NVIDIA GPUs.

I think the last thing I'd say is Blackstone and Google are both massive and fantastic partners of ours. Blackstone did our first DDTLs. I think they've participated in every single transaction that we've done. I was speaking with one of the partners this morning, and they're going to be participating on our upcoming transaction, right? Like that relationship doesn't change. And Google is a large GPU client of ours as well. They're a multibillion-dollar client consumer of our GPU platform. So look, it just comes back to, this is another demand signal, and we wish those guys the best of luck in the TPU space.

Yes. Yes. Yes. I think you guys have kind of emphasized this and we've tried to as well, which is this is obviously very rapidly growing pie. So there's a lot to go around for everybody, it seems like?

There is. And that pie for us, kind of, sits across 3 clients, or 3 client types, I would say it's hyperscale demand like traditional cloud coming to us because we offer a differentiated product to them, and they keep coming back over and over and over again. It's AI Labs, which I think everyone knows well what that cohort looks like. We were very excited to announce Anthropic as a customer last quarter that got added. I think 9 of the top 10 global non-China AI Labs are on our platform today, which is incredibly exciting from a diversification perspective, incredibly exciting from a, like -- just verification of we have the best performing product in the space for all these entities, and it's just become kind of de facto mandatory to work with CoreWeave.

But the last customer segment is one we don't talk about very often, that's enterprise, right? The great indicator for enterprise is inference demand. A great indicator for enterprise, I think, is famously just following what Anthropic has done in the space, right? Their client base is predominantly enterprise, and they've grown so much. I mean, I believe the questions I was getting from this room 6 months ago is where is enterprise demand? Is it going to show up? Is enterprise adopting AI? And I think that answer today is overwhelmingly yes.

And if the question is, are they working with CoreWeave? That's also overwhelmingly, yes. It's just we don't talk about those deals as much because they're not -- those are 8- and 9-figure deals. They're not 10- and 11-figure deals like our huge banner contracts have culminated in $100 billion in backlog. But make no mistake, in Q4, we added double -- more than double the number of logos that we had ever added in any preceding quarter. Those logos are coming from enterprise. Enterprise is rapidly growing on our platform. And across those 3 segments, we're incredibly excited to have such a diverse set of clients out there.

Yes. A lot of threads we're definitely going to pull on. I think to start off with the one area we can't miss is, you guys are very clear about the demand environment, more than anybody else out there. So just a couple of quotes here.

Demand is insatiable. We're turning customers away at the door. I think Nick said, I'll give my onboarding shot for 100 megs of contiguous power in 2026.

And he's [indiscernible] now, too.

He's got to follow through.

He's got to follow through, we find that.

So I think especially right now, it's a little bit rare to hear about that level of demand. If we just go one level below that, and I'm sure it's all of these, but maybe hit on them, is it a new cycle of chips, agentic workloads, enterprises kind of crossing that threshold for AI adoption. What is kind of coming together here all at once?

Yes. I think it's inference, right? That is really kind of pushing this next lurch in demand. Training was obviously a massive part of establishing the AI product. Well, now it's monetizing the AI product that's out there. And if you don't have the infrastructure, you do not have an ability to monetize AI as a revenue stream. And I think one of the clearest indications of that maturation of the monetization of AI is moving into these other components, right? It's moving into CPU. It's moving into storage. It's using agentic workloads.

Like this is all an evolution of an already existing demand base where it was all just like, frankly, kind of simplistic LLMs. Well, now we're moving into truly empowering workforces to work with this Agentic platform. And to do that, it requires different components to work with now. And like these things that like start resembling the cloud that everyone in this room is familiar with from the 2010s that relied upon lots of peripheral components to serve different elements of the workload stack, or the demand stack. And that's something that we are purpose-built for. It's something that our clients have been telling us that they need for quarters at this point.

Like it's sort of an underappreciated fact, like having such a diverse client base, we are directly supporting the leaders in the AI space every day. And that direct support is a dual way dialogue, right? They're telling us what they need. And we are then able to be proactive in market. We're able to be proactive in supply chain with these components and ensuring that we are moving our technology, our engineering teams, our procurement teams in the right direction to support where these guys are going over the next 12 to 24 months.

Yes. And maybe just one aspect of that. I mean, how much of that pull forward, and what specification customers are wanting also figures into that software layer of it? Is that kind of increasingly becoming a big factor as more of the inference is coming in?

Yes. I think it's -- that software layer for us is mission control, it's sunk. It's how do you provision and operate these clusters at scale. At the end of the day, there's a handful of companies that have delivered over 1 gigawatt of billable compute. It's a massive number, right? Like people just throw around gigawatts today, like it's nothing. But I think a lot of the time when people are throwing around gigawatts, it's they have secured a gigawatt of power, or they bought land that has a gigawatt of power associated with it.

That's -- there's a chasm of execution risk between contracting for power and delivering billable GPU hours, and that's where our software stack sits. That's where our supply chain and procurement sits, like that is the magic of CoreWeave that we can build supercomputers at scale. And maintain them and keep them online and operate them. And it's a little bit of a tangent, but like doing this in space is going to be an entirely different thing. And I struggle to see how that's a near-term reality given just how hard it is to do on the ground.

But CoreWeave is simply best-in-class at doing it on the ground. And we're recognized by not only the most demanding AI clients in the world, we're recognized by the most demanding AI suppliers in the world with our relationships across the supply chain. I mean these GPUs aren't just going to anyone. They're going to the people that can bring it online and have a demonstrated track record of execution.

Yes. Yes. I think one very important topic to touch on is inference. I remember at the beginning of this year, I think you guys were at a conference saying, hey, keep an eye on this. It's coming. There's observability players who have been calling it out as well. So a lot of subtle signals. I think it might be an underappreciated part of what the downstream impacts of that are.

But if we just kind of dig into it, you guys had said on the Q1 call that it's materially in excess of 50% of your power drop based on what you guys can see. And you've been saying that it's been picking up. You referenced it earlier. So are these workloads -- like how should we imagine they're percolating across your installed base? Is this customers who are transitioning from training to inferencing? Is it people coming net new just for inferencing? What does that look like from your perspective?

Great question. I would say it's both. Let me start on the architecture side and moving from training to inference.

So for the most part in an infrastructure -- in an architecture's life cycle, the first 12-ish months are focused on training, right? That's your opportunity to advance your models beyond the rest of the competition, right? After that, it's largely fine-tuning and inference thereafter, right?

What we build is AI infrastructure, meaning it's infrastructure that we use for both training and inference. We don't build explicitly for training or inference. Our clients seamlessly use their infrastructure with us for everything, right? It crosses both training and inference. So when we qualify 50%, I would say that that's reflective of Hopper maturing. I would say that's reflective of also new clients coming in who want only inference, like they're not training foundation models. That's like financial services, for example, which I think -- financial services, in my notes, they're over $10 billion of our backlog today, right?

Like that's a massive number that I don't think many people in this room expect that our backlog has that much financial services associated with it. And they are heavily relying on Hopper, Blackwell, Ampere even for inference. That does bring up another question that is just coming up less and less frequently now, which is what is the appropriate depreciation, or life cycle of GPUs. We use 6 years, our peer set uses 6 years. I think at the end of the day, that might end up being conservative, frankly, I wouldn't expect any change anytime soon. But from what we're seeing, I mean, Hopper prices are going up. Ampere prices are going up. Blackwell prices are going up across the board.

This isn't a people only want the latest generation of technology. They want the technology that's the best fit for their workload. And surprise, surprise, that's not just the latest generation GPU. It reaches all the way back to Ampere, and that's on our platform, right? You can look at the broader cloud and you're going to see Tesla and Volta GPUs, that are late 2010 SKUs, and those are still running online. They're not running for free and for fun, like they're running because they're profitable.

And interestingly, those are well past their 6-year depreciation curve, and that is the single largest input cost of running our infrastructure is depreciation. So for us, our focus has been signing these 5- and 6-year long-term take-or-pay agreements to fully derisk the depreciation period to fully handle the CapEx, interest, operational expenses and kick off a 25% contribution margin up to the parent bill. But after that, man, we have a pretty interesting asset class to work with after we paid the GPUs after the depreciation has run off and demand looks like it's just going to continue to be insatiable.

Yes. And maybe one framing that might be helpful is when you have these customers that are transitioning from training to inferencing, what does that life cycle look like? I mean, is that within the 6-year contract like do 3 years training, 3 years inferencing? Because I've heard from contacts in the field sometimes it's 4 years training, 4 years inferencing, all of a sudden, you have the data center kind of being used with the same CapEx you put into it for quite a while.

I would say it varies. But for our clients, it seems like literally, it can be hours later. They can be using the exact same infrastructure for training of next-gen foundation model across hundreds of thousands of GPUs. And then the next hour, they're running inference on it, right? That's how seamless it is. And that's because of not only how we build the infrastructure, but how we operate it from an infrastructure management infrastructure orchestration perspective.

Yes. Is there any future where there are inferencing dedicated facilities or you have -- I mean it's a bit of a myth that inferencing needs to be at the edge, right? But obviously, some of that will be. Is that in the cards or it kind of doesn't make sense at this point for you guys?

We are client-led in what we build, what type of data center capacity we procure. Clients aren't asking for that, right? They're not too latency sensitive, right? Like I don't know about you guys when I use ChatGPT or Claude, I can't tell the difference if it's a 10 mil or 20 mil response time. And accordingly, our clients really aren't emphasizing it too much. Like financial services, probably emphasize that a little bit more in the location of the site. But overall, we're not getting client demand to build something that is edge, or client demand that is inference only. And we're not getting client demand for, frankly, chips outside of NVIDIA's infrastructure either. It's just been consistent of -- and it might be a little bit self-selecting just because we're known as the best operator of NVIDIA's infrastructure on the planet.

But at the end of the day, that's not like a question in the pipeline, right? The pipeline isn't full of clients saying like, well, here's my TPU pricing. Can you match that on GPU? Like, no one asks that. That doesn't come up. I think the client has already made a decision what they want to do and they're not -- there isn't fungibility between those 2 architectures, right? That's a very fundamental decision between the 2 of them.

And I know I've heard you say this, but just to put a finer point, if you did have clients coming to you saying, hey, we want to use XYZ semiconductor technology, you think you would do it?

Yes. Look, I think we would absolutely consider it. It would need to be at a pretty large scale. But I think the main point there is our operational stack, like how we provision and operate compute is not dependent on the underlying hardware, meaning we can operate really anything we want.

I think we've demonstrated that even within NVIDIA of moving from Hopper to Grace Blackwell. Those are entirely different architectures, right? It's a completely different compute platform between those things. And we were first to market with that infrastructure because of our provisioning and operational solution that we have. It's not just us plugging things in quickly. That's us having such a malleable operation stack to run and incorporate new pieces of infrastructure, CPU, storage, like all these other peripheral components that are starting to come into market right now. The CoreWeave solution is the best solution in the market to run workloads for artificial intelligence.

You hinted a bit earlier, but this kind of large tail of enterprise customers coming in. I'd love to hear your perspective. Is that -- does that change how you're going to market with those? Are those customers like a lot of others just coming to your door knocking on your door and asking you for your solutions? Or are you guys kind of doing that outbound as well there?

We do that outbound. We certainly get a lot of inbound as well. We recently brought in Jon Jones to lead our revenue organization. He's building that and has delivered a very enterprise forward sales mechanism that's out there. And those clients, I mean, they've been flooding our platform since last year. I think the enterprise demand is robust. I think that that's really indicative of just how differentiated our platform is as well.

I mean it's a heavy lift to exit, or split your workloads from a hyperscaler platform to moving somewhere else. There has to be a lot of reason for you to do that. And that reason isn't just supply in the market. I think that reason is product differentiation. And we're really proud of the client base that we've been able to pick up on the enterprise side.

And beyond the fact that I would assume they do a little bit more inferencing versus training, right, these enterprise customers. But is there any difference in how they kind of contract, utilize your platform? I mean like you said, they're not as big commitments in terms of pure dollars individually?

Correct.

So do they have similar kind of dynamics of 6-year contracts a little bit shorter? Is there any on demand there?

Yes. I'd say our contracts are 4 to 6 years in duration. That's inclusive of enterprise as well. Like they're wanting to sign longer-term commits on those contracts. It's all similar margin profiles as well. As I mentioned, we have materials on our website. We target a 25% contribution margin of operating these clusters at the SPV level to go back up to the parent. And we target that regardless of what the underlying SKU is, right?

We have margin targets at the launch of each next -- each generation of architecture, and then we bring that into the market in that kind of sets pricing, et cetera. But I wouldn't say material contract differences between enterprise, AI Labs, hyperscalers, not that much of a duration difference either. And the use is inference. We see enterprise pretty heavily leaning into inference relative to training.

Do you think there's any chance some of these enterprise customers kind of upscale their contracts before they run out? I mean it feels like they're very early innings. I assume they're not kind of projecting 10x, 20x growth in that. So do you think there's any chance they're undershooting how much demand they'll have just because they're still experimenting with it or in early innings, however you think about it?

I think so. And I think that's what you're seeing out there as well. I mean we've all seen the reports of engineers using 10x the credits that they had been budgeted, but it's driving productivity. So management teams are allowing for it. And that -- and I would say that, that's the exact same cadence we saw with our AI Lab clients, our hyperscale clients when we were really growing within those sectors over the past few years, as you get on to the CoreWeave platform with your first contract, you get some -- you get used to operating with us, you say, wow, this is fantastic. And then you go sign your expansion contract and expansion contract. And to us, it's less of like renewal cadence, right?

Renewals sound like kind of stable market, meaning that it's not in hyper growth anymore. Everyone kind of knows what they need and what their demand profile is, et cetera. It's just not the cadence that we're in right now. When renewals do come up on our platform, we are seeing clients take advantage of that, right? Like -- and we typically work with that as well. As they get renewed, like our Hoppers, Amperes are being renewed into new 1- to 3-year contracts right now.

I would say our business plan was for that infrastructure to kind of roll off into on-demand pools instead after its initial contracting period, but it's really hard to say no to 100% utilization rates and firm economics for a multiyear period, right? Like sure, margins are more attractive in an on-demand environment today, but we don't know what that looks like 3 months, 6 months, 9 months from now. What we're trying to play is the longer game of how do you derisk a business with so much capital consumption and deployment in such a high velocity technology market, you do that with long-term take-or-pay contracts, in a way that fully derisks the infrastructure.

Yes. I think that's a great segue talking about the contract structure, contribution margins earlier. So let's just talk about margins as a whole.

Q1 is supposed to be a trough for you guys. You reaffirmed that you're going to exit the year at like a low double-digit pro forma operating margin. Long-term target, I think, still is 25% to 30%. So we're in mid-May. You got a lot of deployments coming up in the back half. What's giving you that confidence in that you'll be able to exit the way you want to exit?

Yes. So I think last year, when we were presented with kind of a similar scenario, we had so many deployments that were coming online in December. Like that's really tough to kind of cram it in, in December and Q4 at the end of the year.

This year, those deployments are coming online right now, right? And like this is all verifiable information. If you go look at our data center providers, you can look at their capacity ramp schedules for us and just directly translate to our platform, but the bulk of our capacity is ramping today in Q2 and in Q3. We've seen that show up in our numbers as well from an OpEx, CapEx perspective. And so I think that just kind of like a small learning curve for the market was that investment precedes revenue within the infrastructure space.

And so what you've seen with our margin compression and agree that Q1 was the trough and we're expanding out of there and everything that we gave in the guide, I think is absolutely correct. You're watching that investment period with lots of infrastructure that's coming online. And that infrastructure is coming online right now. And I think that's what really offers my confidence in the H2 numbers that we've given into the market is like we're watching all this come online. And these are -- all these deployments are coming online for contractual commitments, right? We know the economics of them. We have all the infrastructure for these deployments. We know the cost of the infrastructure for all these deployments, and thus, we know what the margin profile is for them accordingly.

Yes. So for you guys, it's a mechanical math problem, kind of you guys looking through it.

And execution is -- we've delivered a gigawatt of billable GPUs. I think there's maybe 4 companies on the planet who have done that. It's an unbelievable amount of infrastructure that we've brought online. I could not be more proud of our team to have done so, but execution is what matters. And execution, by the way, I think what's allowed for us to drop our cost of capital so aggressively, right?

Like 2, 3 years ago when we were doing our first DDTLs, [ LTVs done ] at plus 850 for Microsoft offtake, right? I think that same contract today, all variables being held equal, we're getting done at S plus 200, S plus 225. But that gap, that massive decrease in cost is all execution, right, that we have a proven track record of being able to participate in the credit markets and deliver billable GPU hours. And I don't -- I think that we're kind of second to none in the market for doing so.

Yes. Again, great segue. Let's talk about financing, a big topic for you guys to say the least. I think there's 2 parts for it, right, which is the cost of it and the access to it. And both of those are pretty important to you guys.

You talked about how that cost has come down pretty substantially. You've got a lot of investments coming in debt, equity, all sorts of mechanisms. And you closed, I think, the DDTL 5.0 today officially, right? So maybe just talk about where you see that cost side going? I mean, is there a floor you kind of reaches the execution than whatever your customers are? And how do you see your path towards investment grade?

Yes. So our thesis on financing has been this is a debt finance business, right? This was an equity finance business, we would just be raising tens of billions of dollars of equity, many tens of billions of dollars of equity every year, and that would be a pretty tough case to the equity market. And look, at the end of the day, like it makes sense to use credit to fund this infrastructure, right?

You have a physical asset, you have take-or-pay agreements. There -- it's a concept that's not unfamiliar across the credit market, right? And so we kind of think about the financing flywheel in 2 ways, right? There's ParentCo financing and then there's AssetCo financing. AssetCo is where all of our large contracts sit. It's where all the infrastructure associated with those large contracts sit. That's the DDTL cell facilities that we've been doing. Those facilities are being done at like 90% to 100% LTCs for investment-grade offtake and 70, 75-ish percent LTCs for non-investment-grade offtake. That's the leverage profile that we're able to bring to market there.

And then ParentCo is responsible for funding any of that like kind of delta in G&A, right? That's where ParentCo financing sends back down to AssetCo. ParentCo will continue to be a mix of convert instruments, high-yield instruments, equity issuance, like whatever kind of sits there. But I think it will be a diminishing percentage on a relative basis because we will keep ratcheting up the LTCs down at AssetCo. And then also AssetCo is kicking off net proceeds to parent, right?

Like AssetCo is profitable from that sense. Like it is margin accretive up to the parent and parent will get a larger and larger stream of these like kind of clean net proceeds from AssetCo that it just turns around and back down into AssetCo, but that will reduce our reliance on issuing these parent-level securities over time.

And then related, but maybe we can hit it quickly, like there is sometimes a question like how are you guys going to raise all that debt, all that capital, but it seems like your access to capital is pretty substantial.

We've done a pretty good job. Yes. The DDTL 4.0 was our first investment-grade rated instrument, that's a massive milestone, right? Like being able to get there, it opens the world to be able to participate in our credit instruments. That was a -- that was just a phenomenal deal for us. And within that deal, we introduced some technology. We actually had a -- it was a 90% LTC transaction in like the construction phase. But the revenue phase, we have an additional -- I think it's a 14 percentage point unlock for an ABS style financing that takes it to 104% LTC. Like that's fantastic for us, right? Like that's just a further enablement of AssetCo to be able to self-finance and not require these parent level financings. That facility, again, was for investment grade.

I don't think we're quite there for non-investment-grade counterparties yet, right? The best example of non-investment-grade deal is what we just completed and announced today, which was DDTL 5.0. That was done at S plus 450, I think like kind of 70-ish percent LTC on that deal. But the advancement of that transaction, we really like, it was a publicly syndicated security. It's a traded asset now. That's the first time that's really been done, and it was met with overwhelming demand. I mean that was a $3.1 billion, $3.5 billion facility. It had $19 billion of demand. It was the largest ever TLB demand book, pretty wide, right? And I think that just goes to show you how much demand there is for exposure to AI within the credit world, even if it's non-investment grade, right?

Like that was OpenAI and Cohere. The demand is there for it. I think headlines, media will say one thing about the market's appetite of demand, but when you have $20 billion worth of demand for these credit instruments showing it in the market, I mean that's the reality.

I want to hit 2 things. I'm a little short on time, but let's talk about supply. It's not just energy. You have electricians, transformers, all sorts of stuff. But can you talk about where you're seeing the most tightness right now? And when you look forward to 2027, or any time in the future, when do you see that loosening up? Because you have a lot of push and pulls and it takes time to train electricians, right? So do you see that loosening up at your point?

I feel like we've been asked this question for the last 4 years. And every time we've said it's a ways out. I think it's still...

Always 5 years from now.

Yes. And that's the reality, right? Like we are -- I struggle to see a kind of supply-demand balance before the end of the decade. It's truly. I don't know how that gets resolved. I think today, it's on powered shell. It's not electricity, right? Like the electricity is accessible. It is there. It's the ability to consume electricity at the rack level that's not there. And we refer to that and the industry refers as powered shell capacity. What is powered shell bottlenecked by?

You're absolutely correct. Electricians, massive bottleneck within powered shell delivery time lines. You have transformers, you have backup batteries, like you have all of these components where these are global supply chains that were not built to scale and react at the pace of AI, and it's going to remain constrained. For us, how do we navigate that?

We have over 43 sites in operation today. Like we've been navigating this for years. We know how to deal with supply chain disruptions. We know which partners to work with. We know how to solve problems when they pop up. And I don't believe that, that is going to change in the near term.

One question we'd love to ask, and I think for you, it's probably more relevant than other companies is, when we're sitting here a year from now, what do you think the audience is kind of going to appreciate that they maybe don't appreciate now, right? What do you see on the horizon that others probably aren't giving enough way to?

It's a point I hammer on a lot, and we touched on it briefly, but it's this concept that like signed power like -- or signed leases does not translate to revenue, right? I think, again, that there is an oversimplification in the market of, well, this company has 500 megawatts of signed power. That must mean that they're going to be able to easily translate that to 500 megawatts of GPU associated revenue. It's just not the case, right?

We really have not seen execution across the rest of the sector enough to confidently say that it's easy to deploy GPUs. I think you might see it, I can confidently say it's intensely difficult to build and deliver this infrastructure. I think you have to do it at scale, right? You do it at gigawatt scale. It's just unbelievable physical feats of engineering that are being accomplished. And I think that CoreWeave is simply best-in-class at doing that.

Yes. That's why we're here, too. Listen, Brannin, thank you very much. It's been a pleasure having you here, and I think everybody has enjoyed that. So thank you.

Thank you.

Hello, everyone. Thank you for joining us, and welcome to CoreWeave's First Quarter 2026 Earnings Call. [Operator Instructions] I will now hand the conference over to CoreWeave. Please go ahead.

Thank you. Good afternoon, and welcome to CoreWeave's First Quarter 2026 Earnings Conference Call. Joining the call today to discuss our results are Mike Intrator, CEO; and Nitin Agrawal, CFO. Before we get started, I would like to take this opportunity to remind you that our remarks today will include forward-looking statements. Actual results may differ materially from those contemplated by these forward-looking statements. Factors that could cause these results to differ materially are set forth in today's earnings press release and in our quarterly report on Form 10-Q to be filed with SEC. Any forward-looking statements that we make on this call are based on assumptions as of today, and we undertake no obligation to update these statements as a result of new information or future events.

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 an accompanying investor presentation are available on our website at investors.coreweave.com. A replay of this call will also be available on our Investor Relations website.

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

Good afternoon, everyone, and thank you for joining us. Q1 was a transformational quarter for CoreWeave. We delivered our strongest order for customer bookings, signing more than $40 billion of new commitments and growing contracted revenue backlog to nearly $100 billion. We generated approximately $2.1 billion of revenue, up 32% quarter-over-quarter and 112% year-over-year, and surpassed 1 gigawatt of active power. A milestone only a handful of cloud companies have ever achieved as we convert contracted capacity into revenue-generating cloud services. We concluded Q1 stronger than ever. AI diffusion is accelerating and our addressable market customer base and platform are all expanding rapidly. CoreWeave remains at the forefront of this generational shift.

The 4 themes I would like to highlight today are: one, the demand environment is intensifying, driven by the hyper growth of our existing customers and the rapid maturation of new enterprise verticals. Two, we have broadened the capabilities of our platform. to serve every customer use case from training to inference to agentic workloads, positioning CoreWeave to benefit from sustained margin-accretive growth. Three, we have reached hyperscale with more than 3.5 gigawatts of contracted power, up more than 400 megawatts this quarter alone, with the substantial majority expected to be online by the end of 2027. And four, our financing engine has taken a significant leap forward, unlocking new sources of capital across markets at a lower weighted average cost, enabling CoreWeave to secure more than $20 billion of debt and equity year-to-date.

Beginning with the demand environment. CoreWeave's addressable market is expanding and customers are choosing our platform for the long term. AI workloads are moving from training to inference, agents and enterprise production across industries, which is increasingly compute-intensive. As a result, our core customers, historically hyperscalers and foundation labs are deepening their commitment to us. While an entirely new wave of enterprises are arriving and demanding access to CoreWeave's platform at scale. This trend drove record revenue backlog additions in Q1 as we signed our initial virarubin deals while continuing to monetize our Blackwell, hopper and Ampere capacity. These new customer commitments mostly contributed towards our 2027 targets.

Importantly, we expect them to be highly contribution margin positive and consistent with the return profiles we have historically underwritten for new deployments. In Q1, we added Anthropic as a customer to support the development and deployment of the Claude family of AI models. We also signed multiple new orders with Meta, including the $21 billion agreement announced in early April. The world's 4 preeminent AI model developers now lie on CoreWeave Cloud as 9 of the 10 AI leaders outside of China. At the same time, new verticals are emerging that have already reached $1 billion-plus scale.

Within Financial Services, technology-driven firms are scaling their core machine learning workloads with CoreWeave. These are not AI labs, but rather enterprise customers who see the tangible financial impact and attractive returns that come from adopting CoreWeave Cloud. This vertical is already approaching $10 billion in our revenue backlog, driven by expanded commitments from existing partners like Jane Street, who added $6 billion of capacity in Q1 and new customers like Hudson River trading. Physical AI and spatial computing has also surpassed $1 billion in revenue backlog contributions.

Companies pushing the frontier of world models robotics, autonomous driving and scientific discovery are choosing CoreWeave because of our unique combination of performance, specialized infrastructure and developer tools that accelerate training and deployments. Recent new customers include World Labs, Physics X, and Sunday Robotics. Taken together with our focused execution, these dynamics are driving CoreWeave's customer diversification. Today, we have 10 customers committed to spending at least $1 billion with CoreWeave. Serving this breadth of customers and workloads requires a mix of new and prior generations of NVIDIA GPUs as inference demand skyrockets and agents enter the workforce.

As a result, demand is accelerating across the board. Average pricing for the A100, H100 and H200s and L40s all increased quarter-over-quarter, and we remain largely sold out for near-term capacity across our fleet. This validates what we have long believed. Demand for inference ready compute across generations of GPUs is compounding. We expect this will be durable and accretive to our long-term margins and earnings power. The scale and quality of demand deserves a moment of context. Inference is the monetization of AI and its acceleration is driving real-world productivity gains that are justifying increased investment and broader enterprise adoption. As a result, we added more backlog in a single quarter than most AI cloud platforms have in their history.

Moving on to CoreWeave's cloud platform. We are deliberately strengthening our integrated stack to deliver the most capable AI cloud. CoreWeave even powers researchers, developers and platform engineers to rapidly iterate across training inference and agentic workloads, accelerating the journey from experimentation to production at scale. As we do, we are introducing new capabilities to ensure enterprise customers can build on CoreWeave. For example, in Q1, we introduced CoreWeave Trust Center to allow enterprises to productize AI quickly and efficiently without compromising their security or compliance standards. Customers rely on CoreWeave for our fully integrated set of AI cloud capabilities, not just GPUs. To unlock the full potential of accelerated computing, they need CPUs, storage, networking, software solutions and developer tools working together across every layer.

Today, more than 90% of our reserved instance customers use at least 2 of our products, while more than 75% use 3 or more. Our storage business continues to multiply quickly. We are also seeing similar trajectories in our software, CPU and networking businesses, each of which we expect to exceed $100 million of ARR by the end of the year. It is inspiring to see how customers are leveraging these products to innovate. For example, Perplexity will power its next generation of inference workloads on CoreWeave's platform, while also leveraging weights and biases to help train fine tune and manage their models. Meanwhile, Advaita Bio, is using CoreWeave to accelerate pathways and single cell analysis on large-scale biological data sets, compressing workflows that previously took weeks into minutes.

As we scale, we are offering customers greater flexibility in how they consume compute. In Q1, we introduced Flex reservation and spot pricing, revolutionizing how customers manage peak demand and unpredictable bursts, allowing them to budget more effectively while taking advantage of larger preemption windows CoreWeave provides. Both offerings were immediately oversubscribed. We are also making cross-cloud AI easier for customers, while ensuring CoreWeave becomes the critical cloud partner. Building upon the momentum of CoreWeave's AI object storage and our 0 egress offerings. We recently announced CoreWeave Interconnect in collaboration with Google Cloud. In addition to Sunk Anywhere and [indiscernible] Crosscloud, these aim to remove the friction of managing a multi-cloud footprint, making it simpler and faster for organizations to run workloads anywhere and are already proving to be highly effective at capturing increased wallet share.

And to meet customers where they are, we are beginning to offer CoreWeave Omni, enabling us to deploy and operate our full cloud stack in customers' own data centers with their GPUs. Early interest is strong across potential cloud, enterprise and sovereign customers, and we look forward to sharing updates as it comes to market. Taken together, these capabilities reflect a simple but powerful shift. Customers come to CoreWeave to build and deploy with us.

Turning to execution. We convert scarce AI infrastructure into revenue-generating AI cloud capacity quickly, reliably and profitably. CoreWeave surpassed 1 gigawatt of active power this quarter. We remain firmly on track to reach or exceed our target of more than 1.7 gigawatts by the end of 2026. We are 1 of only a handful of cloud platforms in history to reach this scale, and we are the only 1 that is purpose-built for AI. Each new build-out is complex with 5 phases: power, cooling, networking, servers, and the software orchestration layer. We have navigated this complexity across close to 50 data centers, consistently delivering best-in-class dock to live times for customers.

CoreWeave views these build-outs as investments in the communities that host them, and we work to earn our place in the regions we operate. Our facilities serve as anchors for regional economic activity that compounds over time, grid upgrades, workforce development, sustained local investment. Each project comes with its own set of unique opportunities and challenges. We approach every market we enter with that in mind, helping our team and partners broaden their expertise and execution capabilities for the next project. We added more than 400 megawatts of contracted power in Q1 and bringing our total contracted power to more than 3.5 gigawatts. We expect the substantial majority of this contracted power to come online by the end of 2027.

We added this capacity entirely via long-term leases as data center partners and the financing markets recognized a unique combination of execution, technology and contracted demand that defines our business. Looking ahead, we plan to continue to expand our contracted power footprint through leases while also accelerating our development of self-build sites, which will provide us with greater operational control and long-term financial upside. We expect our first self-build site to come online later this year.

Our lease and self-build strategies are further complemented by our strategic relationship with NVIDIA, where we continue to evaluate opportunities to accelerate the expansion of our footprint together. Our multifaceted approach uniquely positions CoreWeave to grow in a highly competitive market. Finally, touching on our financing approach. In Q1, we reached a transformational milestone that will drive CoreWeave's weighted average cost of capital lower, closing our $8.5 billion delayed draw term loan 4.0 facility. While Nitin will speak to our broader success in the capital markets, I wanted to highlight a few elements of this specific facility.

This is the first ever investment-grade delayed draw term loan backed by HPC infrastructure, achieving an A- equivalent rating from Moody's, Fitch, and DBRS. The facility was nonrecourse to the parent. The transaction, which was well oversubscribed, was priced at a level implying a cost of less than 6%. With this financing, we have taken an important step towards accomplishing our stated goal of driving our cost of debt to investment grade. This is not an incremental improvement. This is a structural shift in how we expect to finance investment-grade customer contracts going forward in what is among the deepest parts of the capital markets.

As a reminder, we have already reduced our weighted average cost of debt by approximately 600 basis points from 2023 to 2025. As of today, we have further compressed our weighted average cost of debt by approximately 80 basis points year-to-date, while securing more than $20 billion of debt and equity capital. De-risking our execution plan and positioning CoreWeave for continued hyper growth.

Before I turn it to Nitin, I wanted to reiterate that we have been building this company against a clear set of defined objectives: one, continue to deliver the most technically advanced cloud platform for AI workloads, empowering our customers to innovate build and deploy; two, diversify and grow our customer base; three, deliver best-in-class execution at hyperscale and finally, position our capital structure to scale with the opportunity. Our backlog is now approaching $100 billion, all tied to contracts that are either online today are expected to begin to come online through 2026 and 2027.

We have built a diversified customer base that includes each of the world's leading model platforms and extends to large enterprises across industries that represent tens of billions of market opportunity and growth. We have moved beyond just GPUs to deliver an integrated AI cloud platform for our customers, who are rapidly adopting our CPU, storage, networking and software solutions. Active power now exceeds 1 gigawatt and our contracted power is more than 3.5 gigawatts, leaving us strongly positioned to meet our goal of reaching more than 8 gigawatts of active power by 2030. And we are continuing to innovate in the capital markets, executing the first investment-grade-rated financing ever secured by HPC infrastructure and defining a new asset class along the way.

Each of these makes the business more durable, each of them compounds, and each of them is a building block for our next stage of growth. The constraint in AI is no longer whether enterprises and AI labs want to deploy. It is how quickly high-performance, reliable AI cloud capacity can be delivered. That is what CoreWeave does best.

With that, I'll turn it over to Nitin.

Thanks, Mike, and good afternoon, everyone. Q1 marks another historic quarter for CoreWeave. Record customer commitments, bringing backlog to nearly $100 billion, more than $2 billion of early revenue and more than 1 gigawatt of active power while unlocking deeper, more efficient sources of financing. We are delivering precisely in line with the road map we laid out on our last earnings call, diversifying and growing with customers, signing customer contracts with attractive and consistent margins, bringing new capacity online rapidly and expanding our purpose-built AI cloud platform.

Demand for CoreWeave cloud is accelerating and we remain largely sold out of our 2026 capacity with prices increasing across the board from Ampere to Hopper to Blackwell. We are seeing this extend into 2027 as well as we've begun allocating capacity, we expect to come online next year.

Turning to Q1 results. Revenue was $2.1 billion, in Q1, up 112% year-over-year and 32% sequentially, driven by continued strong execution in deploying our capacity. Demand for CoreWeave cloud continues to intensify. Revenue backlog for the quarter ended at $99.4 billion, up nearly 50% sequentially and close to 4x year-over-year. This revenue backlog is near-term weighted with 36% expected to be recognized in the next 24 months and 75% in the next 4 years.

With enterprise adoption intensifying and our customer base diversifying, commitments from noninvestment-grade AI-native companies and foundation labs now represents less than 30% of our overall backlog. Customers continue to commit their foundational AI workloads to CoreWeave, resulting in weighted average contract length for new capacity remaining at approximately 5 years. Operating expenses in the first quarter were $2.2 billion, including stock-based compensation expense of $153 million. The increase in our operating expenses was a direct result of continuing to scale our active power capacity, converting backlog into revenue. This drove the corresponding increases in our cost of revenue and technology and infrastructure spend.

In addition, the increase in sales and marketing was driven by increased investment in our go-to-market organization as we further diversify our customer base and expand into new products and markets. G&A increased driven by personnel costs to support our growth while moderating relative to revenue growth on a quarter-on-quarter basis. Adjusted EBITDA for Q1 was $1.2 billion compared to $606 million in Q1 of 2025, growing 91% year-over-year. Our adjusted EBITDA margin was 56%. Adjusted operating income for Q1 was $21 million compared to $163 million in Q1 of 2025, just above the midpoint of our guidance.

Our Q1 adjusted operating margin was 1%, which we continue to expect to be its low point as cloud capacity further ramps in coming quarters. Net loss for Q1 was $740 million compared to a net loss of $315 million in Q1 of 2025. Interest expense for Q1 was $536 million compared to $264 million in Q1 of 2025, driven by increased debt to support the continued scaling of our infrastructure and delivery of our contracted customer commitments. We recorded an income tax provision despite a net loss due to valuation allowance on net deferred tax assets. Absent significant discrete items or a change in circumstances, our tax rate should remain broadly consistent over 2026. Adjusted net loss for Q1 was $589 million compared to $150 million in Q1 of 2025.

Turning to capital expenditures. CapEx in Q1 totaled $6.8 billion as we continue to execute on schedule. Construction in progress, CIP remained roughly unchanged sequentially. As a reminder, construction in progress represents infrastructure, not yet in service and not yet being depreciated. When these assets come into service, they drive incremental revenue and corresponding depreciation. Our financing structure is designed to match this deployment model. Large majority of our term debt is structured as delayed draw facilities, meaning capital is only drawn as the data centers are operationalized. While global supply chain remains complex, we continue to navigate these challenges with operational discipline, leveraging our partner relationships to strategically source required inputs.

Turning to our balance sheet and strong liquidity position, as of March 31, we had more than $3.3 billion in cash, cash equivalents, restricted cash and marketable securities. Since the start of the year, we have made significant progress in strengthening our balance sheet and expanding the depth and breadth of our access to capital. Our success is grounded in the principle that the capital we raise is tied to customer demand. That mindset has informed the business since its inception and has proven critical in our ability to efficiently scale our access to capital.

In Q1, we raised $2 billion of equity in connection with the expansion of our relationship with NVIDIA. We also secured approximately $8.5 billion of investment-grade debt via our fourth delayed broad term loan, which was a seminal transaction for CoreWeave. As Mike mentioned, DDTL 4.0 marked the first-ever investment-grade rated HPC infrastructure backed debt facility, receiving an A- equivalent rating from 3 independent rating agencies. Not only did we price the facility at an implied rate of less than 6%, a meaningful decrease from our previous facilities, we also introduced an ABS style draw feature unlocking an additional $1 billion of drawable capital upon stabilization of the underlying contract. We can use this incremental capital to help future investments for the delivery of subsequent capacity at a highly attractive price.

Further, DDTL 4.0 was structured as nonrecourse allowing us to create a facility that offered enhanced capacity and improved pricing without impacting CoreWeave Inc. and its lenders. Overall, we expect this approach to become the new norm for CoreWeave refinancing the build-out of capacity for investment-grade customers. This represents a substantial step forward in our ability to access capital at immense size and at rates competitive with the hyperscalers. As we entered Q2, we have built upon this momentum. Securing more than $10 billion of additional debt and equity across several different transactions. Each of these was significantly oversubscribed, highlighting the overwhelming investor demand to participate in CoreWeave's hypergrowth. In fact, both of our recent convertible and high-yield offerings were upsized meaningfully due to investor interest.

While the $1 billion strategic investment we received from Jane Street highlights the value and differentiation our customers see in our platform. In conjunction with these raises, S&P also moved our corporate rating outlook from stable to positive. Yesterday, we priced our fifth DDTL facility, the first to be syndicated in the public loan markets to finance contracts with Open AI and Cohere. Again, investor appetite proved to be significant allowing us to price the facility 50 bps inside our initial marketing range. Following this transaction, we have secured all financing required to deliver the entirety of our existing commitments with Open AI. This transaction further underscores investors' support for CoreWeave, when financing investment grade and AI lab customers alike. The combination of these transactions brings us to more than $20 billion of debt and equity capital secured year-to-date.

Our broadening access to capital at lower blended cost will continue to be an important lever for CoreWeave as we convert backlog to revenue and operating cash flow and proactively manage our capital stack. Accordingly, we have no debt maturities until 2029 other than self-amortizing contract back debt and OEM vendor financing.

Turning to guidance. The strength of our Q1 results, disciplined execution and continued momentum we are seeing across our customer base gives us confidence in reaffirming our full year guidance of $12 billion to $13 billion of revenue and $900 million to $1.1 billion of adjusted operating income. For Q2, we expect revenue in the range of $2.45 billion to $2.6 billion, we expect Q2 adjusted operating income of $30 million to $90 million as margins start to expand from their Q1 lows, consistent with our previously discussed expectation. This margin dynamic is timing based, not economic.

To provide some further detail here. Upon receipt of a powered shell, we incur lease and power costs while depreciating server and other data center equipment during the fit-out process, which, on average, takes us about 1 to 2 months. During that period, we recognized costs but no revenue, causing these new deployments to run at negative contribution margins. By month 3, however, we are typically generating revenue with contribution margins normalizing in the mid cities. Since the beginning of 2025, we have almost tripled active power at more than 1 gigawatt, we are rapidly approaching escape velocity and continue to expect adjusted operating margin to expand sequentially for the remainder of the year, returning to low double digits by Q4.

Our Q2 interest expense is expected to be in the range of $650 million to $730 million, reflecting the growth in our debt balance to finance our accelerating deployments. We expect CapEx to be $7 billion to $9 billion as we continue to bring significant capacity online in service of our contracted revenue backlog. For the full year, we now expect CapEx of $31 billion to $35 billion. The increase on the low end from our previous guidance is related to increases in component pricing. The long-term nature of our contracted revenue backlog continues to provide us with clear visibility into 2026 and beyond, and we remain confident in the revenue and margin targets we have put forward.

We now expect to end 2026 with $18 billion to $19 billion of annualized run rate revenue, increasing the low end of our expectations by $1 billion. Continue to expect to grow annualized run rate revenue to more than $30 billion as we exit 2027. More than 75% of which is already contracted, excluding any benefit from not yet exercised customer renewals. We have already secured sufficient power capacity to deliver on our 2027 target and expect to continue to add new capacity and customer commitments to further build on our incredibly strong foundation.

With each quarter of execution, each megawatt delivered each contract signed and each new customer added, we are methodically executing against our long-term plan and building further conviction in our ability to meet or exceed our long-term revenue and margin targets. Q1 was a quarter of measurable progress. Revenue and backlog grew materially as we continue to add new blue-chip customers while growing with our existing partners. Our deployments continue to be profitable at the contract level. While Q1 represented the trough of our margin trajectory, we remain on track for sequential margin expansion through the balance of the year. with revenue and margin growth expected to inflect as we cross from Q2 to Q3.

As important, we have made significant progress in developing our capital structure, more than $20 billion of debt and equity secured, all via meaningfully or subscribe transactions. Each of these is independently significant. Together, they compound because the ability to finance the scale of the opportunity ahead at declining cost is one of the most important levers we have. We have a contracted revenue backlog that provides multiyear visibility and a capital structure that is deeper and more cost efficient than at any point in our history. We look forward to updating you on our progress through the balance of the year. Thank you.

With that, we'll open up for questions.

[Operator Instructions] Your first question comes from the line of Keith Weiss from Morgan Stanley.

Excellent. And congratulations on a really great quarter across like all of the key kind of strategies that you guys are trying to push out into this marketplace and the velocity of the business is just outstanding. A couple of questions that I just kind of clarifying questions within this because there's a lot of numbers, a lot of new stuff being thrown at us. And maybe starting on the CapEx side of the equation and the higher component pricing. I think that's something that's weighing on the stock a little bit after hours. Can you explain to us how the higher component pricing works through the contract and how that ultimately affects your profitability?

And then -- and maybe if we could talk about the NVIDIA relationship and the expansion of that relationship, what fundamentally changed in that relationship this quarter? And how should we think about that 5 gigawatts within the like 8 gigawatt active power target for 2030. I'll leave the rest of my questions for the call back.

Thank you, Keith. That was a mouthful question there. Let me try and kind of deconstruct them and kind of work through them one at a time. So the first thing is a question regarding the CapEx budget and the higher component pricing. And this has been a thematic reality for the cloud and for artificial intelligence infrastructure across the space. Look, we've kind of built this company in an environment that has always been challenged on the supply chain side. So we're really good at it. We think about it a lot. We've built the company from an efficiency perspective. And so we're used to operating in a challenging environment around components or inputs into our product. Having said that, in the last 6, 9 months, there has been an acute shortage of certain components that have moved up. And like I said, you've heard about that across the space.

When you're thinking about that for us, the way to think about it is that we are a success-based company, which means that we build our contracts to incorporate the cost of all of the components that are necessary to deliver infrastructure. And so by and large, we are insulated from the price inflation on some of the components because we include that in our pricing that we ultimately bring to clients in order to target the margins that Nitin spoke to, right, is up in the mid-20s is how we think about it. on a unit basis. And so it's an issue, it's a problem, but we have an incredible capacity to navigate the supply chain. We have great partners, and we include the pricing that is required in order to end up delivering the infrastructure that's required, but also ensuring that we're able to secure the economics that we're targeting.

The next question you asked was about NVIDIA. And it's been a very exciting period for us with NVIDIA. They came out and they really did a series of different things, but the 2 most important were the qualification of our software solution as a reference architecture for them. And that was an incredible validation of the quality of the software solution that we deliver to market. We've talked about this before is that we are dedicated to delivering the best solution to artificial intelligence infrastructure consumers in the world. And we believe we do that and we believe that NVIDIA supports that position because of their position around our software as a reference architecture, which is fantastic.

With regards to the 5 gigawatts worth of infrastructure, I kind of want to start that from the position of in the last 12 months, CoreWeave has secured 2 gigawatts worth of infrastructure. Within the last quarter, we have secured 400 megawatts worth of infrastructure. We are capable of securing infrastructure at gargantuan scale, independent of any support from NVIDIA or anyone else. What the 5 gigawatts does for us is it gives us the ability on an opportunistic basis to accelerate our ability to go out and secure additional infrastructure at a truly amazing scale as our clients are trying to secure our solution for delivery of computing infrastructure. And I think those really are the highlights around our relationship in the last quarter with NVIDIA, beyond just the standard us integrating with them on an engineering first basis, which has just been fantastic.

And Keith, to round up your answer on the CapEx fees. CapEx shows us for us before revenue and cash flow. So you see showing up first. The piece on the P&L impact of that is already incorporated in the guidance that we've issued you today. So with that I know this is your last quarter earnings covering us. We would like you to thank you for your partnership and the support that you've shown for us since our IPO today take today. Thank you so much, Keith. You will be missed.

What you guys have built or has really been incredible and has been awesome to watch firsthand, how you guys have scaled this out so quickly.

Great team.

Your next question comes from the line of Brent Thill from Jefferies.

I'll throw in an excellent for Keith, congrats. Nitin, $81 million in EBITDA in the front half, but you're guiding to $919 million in the second half on the bottom line. What's giving you conviction in the bottom line build in the back half of the year?

Absolutely. So from a Q2 perspective, it is coming in exactly where we expected it to be. Remember, everything in our business is defined by active power and capacity ramp schedules, which are not necessarily linear. What is important in this context is that our revenue growth and our margin trajectory is coming across exactly as we articulated in our Q4 earnings. And I think your question was probably around EBIT, not EBITDA. So I just want to clarify that for people. For EBITDA, we generated $1.2 billion in EBITDA this quarter. When you think about from a Q1 perspective, Q1 was the trough of our margin start. What we remain on track for sequential on expansion through the balance of the year, as we had described. With revenue and margin growth expected to inflect from as we cross from Q2 to Q3. We are reaffirming our full year guidance, including full year revenue full year adjusted operating income and that we will exit the year with low double-digit adjusted operating income margin.

We are reaffirming our active power, and we are raising the floor of our 2026 exit ARR guidance. All of these are indicators in the conviction that we have on our execution plan for the remainder of the year. What you will see us is executing through this plan and expect to see our adjusted operating income accelerate faster than revenue growth in the second half of the year.

Brent, just a couple of other things. You asked about what gives us confidence. And a couple of things I wanted to say as you're watching a business kind of achieving escape velocity is that we have gone ahead and work through our supply chains to ensure that we are able to hit those numbers. We are in approximately 50 data centers. We have no single data center provider that delivers more than 17% of our active infrastructure. We have multiple OEMs, ODMs we are really built now to lean into a resilient supply chain that crosses all of the components that we are required to be able to deliver in order to drive that revenue.

And we are reaffirming it because of the level of confidence that we have that the cost of that resiliency, we will hit those numbers.

Your next question comes from the line of Mark Murphy from JPMorgan.

I'll add my congrats to Keith. And congrats on the strong Q1 performance. To the extent that you have a business that you booked previously, and it's not fully up and running and now the cost of components has risen. I would think the cost of energy has risen, although maybe you have some of that locked in. Is the margin profile of this previously signed contracts slightly different than originally contemplated? Or do you have some recourse somehow to flow through increased pricing or maybe restructure some of those contracts? And then I have a quick follow-up.

Yes, it's a good question and thanks. It really -- like I really do want to say that this is an extraordinary quarter for the company, and we are incredibly excited. Look, in terms of the components, you see us making a slight adjustment in the CapEx to capture small components of the CapEx that might go through some price inflation. But when we are pricing our deals, we are pricing them with purchase orders in hand for the infrastructure that is required to be able to deliver on it. We understand what the power cost is going to be today, next year and out through the term of the contract because that is contractually delivered to us.

And so we've done a really good job of understanding what the components and electricity costs are going to be, we have it structured so that it is effectively passed through when we enter into the contract once again, to ensure that we're able to hit our targeted margins on a unit basis.

Okay. Understood. And just given the success you've had with the bookings in Q1, the data center build execution is great, you have the diversification across enterprises. The successful capital raise. And then you do have noticeable revenue upside in Q1. What holds you back from just passing through that revenue upside into the full year revenue guidance because, I mean, in some sense, technically will reduce our rest of year revenue forecast a little. Is there some other effect? Sometimes just weather or just the infrastructure shortages or maybe labor shortages in there that maybe is creating a mild headwind.

Yes. From our perspective, we mentioned this in our last earnings call as well. We pretty much remain sold out for our 2026 capacity. So that is continuing to be true for us at this point of time, where you would see this kind of index for us is around 2 vectors. We've raised the guidance for the floor of exit ARR. So we are going to exit the year in a much stronger position than what we expected a few months ago when we did last quarter. At the same point of time, where it's also inflicting is showing up in 75% of our 2027 ARR guidance that we provided, which is $30-plus billion already booked at this point of time, excluding any potential renewals better than 75% of that. So that is where you see that reflecting. From a 2026 perspective, we pretty much remain sold out of our capacity.

Your next question comes from the line of Tal Liani from Bank of America.

You have some people here that that's their last quarter of covering the stock, and some people here this is their first quarter of covering the stock. So generation, those on generation...

It's great to have you. you're welcome.

Well, I want to ask about the operations, 2 things. Number 1 is your backlog of revenues, what determines the recognition of revenues from it? Is it the completion of data center build-out and live traffic going into it? And that means that we're going to have kind of step-up in recognition of revenues, really some really good quarters when you finish building out certain capacity? That's the first question.

And the second question about gross margin. So gross margin has been going down throughout. I look at the last 5 quarters. We started at 78%, now 68%. And I'm trying to understand the operating structure, how it changes over the next 2 years, let's say, meaning what are the drivers for gross margin improvement I see also technology and infrastructure expenses went up more than revenue growth. So what are the drivers for that to improve? Like what takes you from the current margin structure to a better margin structure down the road when revenues go up?

Yes. So before we enter into the revenue question that you asked, let me kind of first answer your second question, which is around how we think about the margin dynamic. The margin dynamic in our business is predominantly timing based, not economic. We receive -- get receipts of our shell and we start incurring lease expenses as well as power expenses at that time, and we start depreciating server and other data center equipment during the pit out process. That process takes us about 1 to 2 months. During that period, we are recognizing cost but no revenue. That is what costs us new deployments to be running contribution margin during the negative, during the deployment phase.

We've tripled our active power capacity over the last year or so. And what you see in our business is a reflection of that rapid ramp because we deploy capacity ahead of revenue generation by a few weeks. By month 3 in these deployments, we are typically generating revenue and the contribution margins stabilize normalizing to a ramped contract in about mid-20s. Operating margin inflection also means a gross margin will inflict on time as we inflict into Q2 exiting Q3. That's what will happen through the remainder of the year, and you'll see sequential expansion.

To your question on revenue generation. When we deploy capacity and we have deployed tested our GPUs and when we handed them over to the end customer is when we start recognizing revenue. So as data halls as data centers come online and we deploy capacity for our data centers, that is when we start delivering those capacity components over to our end customers on a contractual basis, and then we start recognizing revenue on a straight-line basis through the life of the customer contract.

Tal, just one more quick point. When you're thinking about the gross margin, right, Remember, we're installing such massive amounts of infrastructure relative to our installed capacity, right? So if you think of it as we're running 50 megawatts, and we add 300 megawatts in a quarter, the impact on gross margin is going to be enormous. On the other hand, when you're running 2,000 megawatts and you add 50 megawatts, it's not going to have as material an impact on your gross margin. And when you're watching a company like CoreWeave go through a scaling exercise like we have been owned, this journey is unique. And so you're going to have impacts that are short term in the scaling exercise. But as soon as you become large enough that the ad that you have a relative basis is more normalized, the margins will immediately reinflate.

And when I say that they're achieving escape velocity, that is exactly what I'm talking about. You're talking about a company whose installed base is getting large enough to handle the next incremental unit of compute, the next data hall, the next data center as they are brought online.

Your next question comes from the line of Amit Daryanani from Evercore.

I have 2 as well. Maybe Mike, to start it, you sort of always talked about inferencing as a monetization of AI, and that's why you've framed that I'd love to understand how fast do you think inferencing as a share of your consumer power is growing right now in your installed base. And as that keeps going, what does that really imply for the utilization and contracting economics for the H100, A100 over the next 12 to 18 months to you.

Yes. I mean this is something that we spend a good bit of time thinking about over here. I'm going to answer the question in a couple of different ways, right? First of all, when we build infrastructure, we build what we call AI infrastructure. And so it can be used fungibly back and forth across both training and inference. And it is the buyers who have purchased this infrastructure over time that move their workloads back and forth to optimize the use of the compute that we deliver to them. And so we don't know necessarily exactly what compute is being used for inference or what compute is being used for training because it changes all the time. However, we can look at the power draw and extrapolate what we believe to be, how much of the compute is being used for inference versus how much is being used for training. And when we do that, we think we are now materially in excess of 50% of our compute is being used for inference.

And that is a wonderful thing for a company like CoreWeave because it gives us tremendous confidence that the consumers of our compute are making -- are driving revenue at their entities because they are able to monetize their investment as they sell their compute and their models on to their clients. With regard for the demand for Hopper and Ampere, I have been pretty steadfast around what I believe is going to be the useful life of this computer. Whether it is the Amperes or the Hoppers or the Blackwells or ultimately as we move through additional iterations of architecture. And what I have said and what I continue to believe is that the use cases within the AI labs within the companies that are productizing artificial intelligence are broad and they require different types of infrastructure to run some of their training loads, and they require different types of infrastructure for some of their inference loads.

And so when we talk about contracts, when we do multiple contracts with a counterparty, they come in and they buy the most bleeding edge infrastructure. And then they use that infrastructure to train and then they take that infrastructure and move it down to the inference load, which is probably less compute-intensive, and they bring in new infrastructure of a new generation with which they use to train the next generation of their models. And so we have seen incredible amounts of demand for our H100, for our H200s, for our A100s, and that is all driven by the fact that there are many different use cases that require different power and scale and pricing of compute.

We are sold out in our H100s. We are sold out in our A100s we are seeing price appreciation as more inference is coming in and making demands upon that compute work loads in order to be able to deliver to their clients. And we think that is an incredibly bullish signal to the space at large.

Perfect. And I have a really quick one. On the component side, One of the other issues I think folks are started to have is just component availability to build out these data centers. And so as you think about the $12 million to $13 million in revenues for calendar '26, how much of component procurement is locked in already, be that GPUs or memory or something else or you folks versus you still have to work your way and get those things locked in.

Yes. The overwhelming majority of it is locked in already, right? So for 2026, as we said, we are virtually sold out. But likewise, we have placed RPOs, we have secured the infrastructure. We have secured the power and everything else that is necessary for us to execute on delivery on our road map. We are very, very comfortable with the guidance that we've given because of having secured that infrastructure and those components already.

Your next question comes from the line of Nehal Chokshi from Northland Capital Markets. Nehal, please go ahead.

First, so you brought in an incremental 400 megawatts of contracted power that's up from 200 megawatts and 4Q '25, but below the 600 megawatts per quarter in 3Q '25. Just as we think about on a longer-term basis, I know you've already provided the calendar '26 and calendar '27 ARR, but even beyond that, what's kind of like the right way to think about how much incremental power you can bring in each quarter?

So I want to be very clear that the 400 megawatts that we added is incremental and distinct from the 200 that we added prior to that. And so we enter into contracts for power with data center providers. Our pipeline for additional capacity is extremely large. We're reviewing lots of different sites where we could build infrastructure, we're reviewing lots of different deals that would allow us to continue to our ramp. Another thing that I want to highlight here is, in addition to the transactions that we're doing with third-party data center providers. We are also doing a series of self-build data centers. And that's going to give us additional operational control over our pipeline of data centers. And that's very important to us as we move out through time.

So like I said, we are building our pipeline of infrastructure in coordination with the signals that we are getting from our clients so that we are able to lease the infrastructure or schedule the self-build to be able to deliver to it. There's no exact number that I can give you and say, "Hey, we can get you x number of lots per quarter for the next 3 years, but we have a very, very robust pipeline of opportunities that we're looking at. And like I said, 400 megawatts this quarter, last quarter was 200 megawatts, but it's been 2 gigawatts in the last 12 months.

Okay. Great. So if I may some rate, basically, do you expect to be able to match supply to your demand and demand is off the charts.

That is correct.

Perfect. All right. The second question is that with about $100 billion of revenue backlog, I would say that should translate to about 2 gigawatts that you've now contracted out. And you've contracted in 3.4 gigawatts as of the end of Q2. So roughly, I would say that means that you have about 1.4 gigawatts to go out and allocate. And I use the word allocate purposely, because I think you guys are trying to be fair about who gets what capacity. So the question is a, is the word allocate rather than sell accurate; and b, is the amount to allocate or sell, if you prefer that word, increasing or decreasing relative to a quarter ago.

I think the word allocate is probably factoring. It's unusual to be in a business where the demand for your product is so high that you get to really be thoughtful about which clients you want to bring on to your infrastructure, where do you want to support parts of the infrastructure that are being built, whether it's in life sciences and biology or in foundation labs or inference products, all of those things. And that is a privileged position, and we are trying to build an ecosystem of clients that we allocate power to that are going to be the leaders in the space as it continues to grow.

You made a comment before that, hey, are you able to coordinate your securing of data center capacity with demand and demand is off the charts. Like I don't want to be flip about this, right? Like the truth of the matter is the limiting factor isn't just power, it's labor, it's memory, it's storage, it's our ability to bring up infrastructure. And so we're really orchestrating the coordination of all of those things so that we're able to deliver infrastructure to our clients. So they depend upon the quality and scale of the infrastructure that they require to be able to drive their business.

That concludes the Q&A session. I will now turn the call back to Mike for closing remarks. Mike, go ahead.

Thank you. So as we conclude, I want to thank the CoreWeave team and our customers and partners None of these accomplishments would have been possible without you. I'm incredibly proud of the growth and the execution that CoreWeave has delivered across every part of our company. Our focus, discipline and commitment to innovation and operational excellence will keep us at the forefront of this revolution. It was true an amazing quarter, and we look forward to speaking to you guys as we continue to move through the year to update you on our progress. Thank you.

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

Excellent. Thank you, everyone, for joining us this afternoon. My name is Keith Weiss. I run the U.S. software equity research franchise here at Morgan Stanley. And really thrilled to have an opportunity to talk to Brannin McBee, Co-Founder and CEO of CoreWeave.

So Brannin, maybe just to open up the conversation. Already, we've seen a growth algorithm from CoreWeave that I haven't seen in my career, right? And I mean, one of the things I love about research is getting to understand new companies and new opportunities and new businesses. And this is all new, right? And over the last couple of years and the last couple of quarters, you guys have described that demand that's already showing up in a huge backlog already showing up in growth rates that are really eye-popping as insatiable and relentless, which to me means that we're not going to see this come to a trickle out anytime soon.

So can you talk to us about where you've seen this demand come from? How foundational is it? And how certain are you in sort of the durability of this demand, not just through 2026, but you guys have started talking about targets into 2030.

Yes. Yes. So it -- look, you're hearing this across the space, right? You're hearing it from our peers, you're hearing it from our clients, you're hearing it from our suppliers. The demand profile is truly overwhelming, insatiable. I would characterize 2026 is probably broadly sold out in terms of billable compute capacity that's available into the market. And it's robust, and we'll get into this throughout the conversation, but it's robust across several sectors, whereas I would say previously, like it was isolated to AI labs, right? That was really the starting cohort of where demand grew from like 2022 plus.

And then for us, it hit our hyperscaler cloud clients, right? And they were coming us to support their product build-out. And now I would say we've seen this rapid advancement of enterprise demand.

The enterprise cohort is absolutely there, whether you look at our numbers for that, you can look at some of the numbers floating around with Anthropic and market for just a true understanding of how quickly enterprise adoption is scaling right now, like it's truly fascinating. And within our guidance, we offered color of exiting 2026 at $17 billion to $19 billion, exiting 2027 at over $30 billion in ARR.

To contextualize that a little bit further, we exited 2025 at $6.7 billion in ARR of demand, right? So it's just this massive step-ups in change in revenue. And within all that, the customer behavior is changing as well. There's 2 main points I'll hit on there. One is the customer is looking for longer duration contracts, right? As most people in here know, we signed multiyear take-or-pay agreements, right? And 24 months ago, those were, call it, 3-year contracts. 12 months ago, there were 4-year contracts. I would say now in our $66.8 billion of backlog that we have, that is 5-year weighted contracts, right, with some contracts in there extending up to 6 years.

I struggle to see visibility like materially beyond that, but the customer profile is saying, we want this infrastructure for longer. And as a reminder, that's for like kind of single SKU exposure. They're coming in saying, we want Hopper for 5 years. We want Blackwell for 5 years. That's one aspect of customer demand.

The other aspect that we're seeing that's so strong is on older generation infrastructure, right? So clients are coming in asking specifically for A100s. They're asking specifically for H100s, 200s. And of course, for Blackwell. Also, it's not a cadence of they come in asking for Blackwell, and they can't get it. So they're like, okay, I guess I'll take Hopper instead of it. And the driver of that is they've engineered their workloads. They have specific use cases for the specific pieces of infrastructure, right? And that to us, the main driver of that is inference, obviously. We'll chat about that more as well.

But this all speaks to this like highly sustainable demand profile, not only for latest-generation compute, but the broad set of infrastructure that's being delivered in the market today.

Got it. When you talk about, it's sensational level of demand and there's other vendors talking about it as well. And you're right, it's not just CoreWeave talking about it, I fear that might overlook the advantages that CoreWeave has in the marketplace, right. And the question that I get from investors, a lot is, is there a differentiation, right? Is this just they're able to provide the capacity and therefore, they get the demand? Or is there some reason that the customers are coming specifically to CoreWeave? And from our work, there is. There is differentiation in terms of your ability to build out faster than anyone else to get the most recent technology out there faster than anyone else, but probably most importantly, to keep it up and running, more durably than anyone else. So can you talk to us about how you guys have built out those competitive advantages? How durable you feel they are? And how you're going to extend those even further with the software layer.

Yes, I think very intentionally, right? And this all goes back to 2019 when we were really standing up our cloud business originally. And that cloud business was built around this concept of parallelizable workloads. And the fundamental idea that parallelizable workloads are different than serializable workloads has different infrastructure, different operational infrastructure requirements around it. So you have to build the thing for the thing, so to say. And if you don't, you're asking your clients to take compromises, right, compromises and stability, scale, ultimately performance, and so the product that we have in market today, I think, is widely recognized as the most performant solution for operating parallelizable compute at supercompute scale, right?

These aren't supercomputers that are being delivered in the market, and they're widely complex to not only bring online, but also to stabilize, and it's the CoreWeave, like operational infrastructure suite that sits on top of it that allows for that to exist.

And so who recognizes that? Third-party consultants recognize that right? I think, analysis does a phenomenal job, really benchmarking the different solutions that are out there in market for running this infrastructure end. We've been singular through two of their first two reporting cycles.

Now it's our clients who keep coming back to us for the products that we have. They are choosing to work with CoreWeave over and over again. And that is diversifying within not only the AI lab cohort and the hyperscaler cloud cohort, but also within the enterprise cohort, right?

These are the blue-chip enterprises that are coming into CoreWeave saying, these guys are running this infrastructure correctly. And it's a long-winded way of saying this infrastructure is not fungible, right? H100 at one of our -- at one cloud is not the same as the other. And within that set, we are the best operationally at this infrastructure.

How do we keep that pace? I mean this is our business. This is what we invest in on a daily basis. We have incredibly tight engineering relationships up and down the supply chain. We're working incredibly closely with our suppliers, with our clients, with our data center operator partners to understand and deploy what is the most effective engineering solution to delivering this supercompute infrastructure at scale. At the end of the day, I was like a lot of proprietary information that comes into our business, right? Like we are solving the problems of the most intensive AI users in the world on a daily basis. And it allows for us to kind of skate to where the puck is going, right?

An example that I really love, and I'll stop talking is when chain of thought models were introduced into the market. This was at a time where I think the broader buy-side, sell-side thesis was models going to be quantize. Everything is going get smaller. We're going to run lots of models on one GPU. Perhaps you don't even need data centers to run models in. But you give these engineers the most performant infrastructure out there, and then they started looking at it saying like, well, what have made the model bigger? And inference left the GPU and left the node, and you could use hundreds of GPUs to run inference on instead, leveraging a high-performance bandwidth capability between those GPUs.

And you got chain of thought reasoning that was introduced to models. And it was like a complete paradigm shift in the way that the infrastructure was consumed. And that immediately led us to understanding memory is going to be a path that really matters, right? Like how much context can you hold in memory on the nodes and within the clusters.

Where is a similar analogy today? I think, it's agentic workloads, right? Agentic workloads, we're seeing increasing pressure on peripheral demand. I would say like CPU demand is absolutely going to increase as Agentic workloads are scaling. We're seeing that pressure from clients. Storage is another component that's been really exciting for us across our client base.

We disclosed in our Q4 earnings that we have a -- I believe it's a greater than 80% attach rate with our clients who we generate more than $1 million in revenue from for our storage product. And our storage product today is well into the -- well north of $100 million in ARR.

That's a product that like didn't really exist not that long ago, right? Our ability to attach peripherals and for them to scale quickly and be quite attractive for us operationally and quite attractive for keeping customers on the CoreWeave platform, I think, is a really exciting opportunity for us.

Got it. And you talked about the tight relationship with your suppliers and one of those big suppliers and an investor is NVIDIA, you talked about expanded relationship with NVIDIA this quarter. Talking about a couple of ways that you're going to be working more closely together. There was a $2 billion incremental investment, which is interesting. But even more interesting is what you guys are doing together with software. So can you dig into that? What's the nature of that relationship? And how is software becoming a bigger part of the story at CoreWeave?

So that announcement in January, that more comprehensive relationship was all about accelerating growth, right? Accelerating our ability to grow at the pace of AI adoption. I think the software side of it is you're absolutely correct to highlight and it's this acknowledgment that the CoreWeave software stack is the best way to run this infrastructure. And for us, when we discussed it in our earnings was, this can lead to an opportunity for us to sell that software solution to other entities. So for entities who may want to have GPUs on their balance sheet, right? Or they might have data sovereignty priorities. And thus, they need to have ownership of all the infrastructure. That ability to take that into market, I think, is a very margin-accretive path for us to be able to go down.

Outstanding. All right. So insatiable demand, market-leading product and solution that you're bringing to the market. And it's created this tremendous backlog, $67 billion of revenue backlog as of the last quarter. Along with it comes a lot of CapEx. You guys have -- you got to build out the infrastructure to be able to support all this demand. You guided to $30 billion to $35 billion. I think one of the investor concerns is the financing of that on a go-forward basis. Can you talk to us about how you're planning on financing that level of investment in 2026 and then beyond?

Yes. So -- we break down -- I think that we've been kind of market leaders and thought leaders in how we finance this infrastructure. And starting with our original DDTLs I think, 3 years ago at this point. We have ParentCo and AssetCo is kind of how we break the business down to, right? And all of our assets sit at AssetCo. That $30 billion to $35 billion guide on CapEx, let's call it $32.5 million midpoint. That all sits at AssetCo. And AssetCo, we're able to bring these financing facilities into where, I would say, we have extreme levels of demand to participate in the paper. One aspect to highlight, and like you've probably seen headlines about an asset level raise that we were working on. And while I'm not going to talk explicitly about those headlines, that's something that we're very excited to announce in the market.

This advancement of the structuring at the AssetCo, I think, it's all representative of not only our execution track record, but also the durability of our contracts and our data center agreements. It all says, yes, we love CoreWeave paper. Yes, we love the way that we're bringing into market. We want to underwrite more and more of it. So that backlog will get financed down at AssetCo with some participation from parent down into AssetCo.

I think the other aspect of your question was a little bit driven by margin of the business, near-term margin. And look, like I spent the last 2 days at LevFin in Florida. And at the end of the day, CapEx requires an investment, right? CapEx requires investment period as it comes online, and we brought some slides onto our IR deck that I encourage everyone to go take a look at that breaks it down in more detail, but the net of it is we have a sort of 3-month investment period on bringing CapEx online.

And once that CapEx is online, and stable, take it like month 3 to 60 in a contract, each contract each deployment really has a 25% contribution margin up to the parent, right? So for every dollar of revenue that's coming into these deployments, you have $0.25 going straight up to the parent or it's straight up to the holdco afterwards, right? So you take that one deployment and you now layer it with lots of deployments that are coming online over time, and you have this very robust sort of revenue stream going up to the parent. Now where are we today?

Today, we're in this extreme growth period, right, where quarter-over-quarter growth. I think we grew our active power by nearly 30% quarter-over-quarter, Q3 to Q4. And when you're incurring the expenses of growing active power that quickly, it's, of course, going to weigh on ParentCo, right? Because you're in this like 3-month period where you have revenue starting to generate, but you're paying for the data center costs. You're paying for the beginning part of depreciation on the infrastructure during that ramp period. And so -- we have large blocks coming online and like it's a smaller amount of infrastructure or in other words, like such large percentage growth, it will weigh on near-term margins.

And so all this is to say, it's an extremely intentional path that we have to growing our business and moving at the pace of AI demand, right? Like this is a phenomenal opportunity for growth. We're doing so in an incredibly risk controlled and risk-adjusted manner with highly accretive contracts that underpin the entire business.

Got it. So there are 2 real veins in there. On the financing side of the equation, your -- the people are looking to finance you are getting more convicted, not less convicted sort of in the underlying business financing costs, if we look at it from what we've seen over the past couple of years have come down from 12% to 9%. And there's an expectation that's going to continue to come down in terms of what it's going to cost to finance it. The other side of the equation, I think what scared investors on the most recent conference call was a forward operating margin forecast that was below what consensus had. But the other part of the equation of what was wrong with what consensus has was the CapEx number. We were well below what you were expecting. And there's just a natural like absorption period, right, that compresses margins in the near term as CapEx is ramping that quickly. That wasn't question.

I completely agree.

CapEx requires investment. And we're at the kind of the trough or Q1 is the trough of that margin profile for us. From here, it's expanding. Okay. So I think part of the reason why that touched the nerve, like with operating margins, is what we're hearing in the marketplace in terms of there's the demand that you're seeing, but you guys are also creating a tremendous amount of demand and not just you. It's all the hyperscalers are creating demand. And component costs are coming up. And memory costs are skyrocketing, and it's hard to get people actually build out these data center facilities. So can you talk to us about that side of the equation, like the degree of difficulty in executing to these build-outs and keeping the project like on time and under budget, if you will, right? How do you maintain that margin profile with these potential costs.

Yes. Supply chain. Supply chain is immensely difficult. I think, it's something we've been quite vocal about private public version of us like it is, these are utterly enormous engineering projects that are being brought online. And I'd encourage, if you ever have an opportunity to go to one of these sites, like please go visit them. And I think you'll begin to get an understanding of just how hard it is to bring these things online.

But they should be invited because there's a lot of security at these data centers.

Yes. We thought -- there is -- and I think that, that's been underappreciated by the market, right, like we're throwing on terms like 1 gigawatt, 5 gigawatts, 10 gigawatts, like it's just a number on a spreadsheet, but the reality is like it is thousands and tens of thousands of people to deliver infrastructure at that scale. And I think if we were to ask like, where is the bottleneck in the market today.

I would differentiate between power and active power, I mean, power and data centers, right? It's less about electrons, right? We observed that the electron availability on U.S. grid is there, but it's how do you deliver those electrons into the racks, into the servers, whether it's the physical infrastructure that sit on the site, like transformers, backup gen, back or battery, everything along those lines, or it's just the people, right?

Like electrical engineers are an incredibly critical part of delivering these sites. You can't really make more electrical engineers very quickly, right? That is a skilled trade that takes years to bring a workforce online for it. So I think that, that is where the bottleneck of growth is in the near term.

Where do we sit within those profiles, we predominantly lease our data center capacity. We're doing a little bit of self-development ourselves, but predominantly, we're, I think at 43 active sites in operation, right? It sounds that we just have like 1 or 2 sites that we're looking at that like dictate the success or failure of our business. We have 43 that we've already delivered. We know how to do this. We've done it for years. We know how to navigate the complexities in supply chain.

As I'm sure everyone recalls in Q3, we disclosed that we got surprised, right, by 1 site. And we bring a lot of conservatism into our forecasting. We know how our operators work, but everyone gets hit by supply chain problems, right? Like it is just incredibly hard. So that one site, we worked closely with that operator to get it back on track. We're happy to say, as we disclosed in our earnings, that site is firmly back on track. And I think we actually delivered relative to our expectations a little bit early on that site. But we're only able to do that because we have all this experience of executing on sites.

So will it remain challenging? Absolutely. Do we build a lot of conservatism already into that supply chain? Yes. But it is a -- I think going to see these sites in person is -- brings a lot of context to them.

Yes, definitely. Maybe just double-click on that. I'm not a hardware guy, but I know some hardware guys. And when memory prices are up 4x, 5x, it seems like a pretty bad day for Dell, right? It seems like a pretty bad day for HP. Is it a bad day for CoreWeave in that same way, -- like -- or are you able to -- is it too small of a part of your bill of materials? Or are you able to pass on those higher costs to your end customers?

It's a very small component of the actual node costs, right? Overwhelmingly, it's on the GPU side relative to memory. For us, I would say, we're far more focused on supply chain, right? Like ensuring that we can get the components because you're absolutely correct, like as component costs increase, that just gets passed on to the end consumer, ultimately, right? Same with like electricity costs. If electricity costs were increasing on new sites that we're entering into contracts with, that gets passed on to the end consumer. And I think the end consumer is very comfortable with like regional pricing, for example, but for us, it's entirely supply chain and ensuring we get the components to deliver the infrastructure to our clients. So going back to my example earlier on chain of thought being introduced to the models, it was Q1 of last year where we really saw that there was going to be this increasing focus on memory, right, within the LLM space.

And that was driven by how close our relationship is with our clients on an engineering level, to understand what matters to them, where the technology is going. It gives us this really unique lens to understand how the demands of the market are going to evolve. And so memory was not that much of a surprise to us, right?

Moving in, we see peripheral demand in general increasing. I mean I think that's only exciting for our business because those are all products that we're able to offer in to our clients to bring them more robust like full cloud experience.

Got it. And then on the other side of the equation on power, it's -- there's been a lot of concern about access to power and availability. On the most recent conference call, you guys talked about a goal of getting an incremental like 5 gigawatts of power by 2030, how comfortable are you on the ability to find that, the ability to source that is there regional difficulties? Is there global difficulties? Or are you guys pretty comfortable that, now we've got a pretty clear line of sight to being able to bring on an additional.

Yes, I think that power is out there. It's again going to be more navigating the supply chain on the data center side. But I believe our data center partners are very focused on bringing online these larger and larger deployments. I think super important for us in there is, we are only procuring that capacity as demand is there for it, right? And the way that we approach demand in our general capacity procurement in general is maintaining conversations with our largest clients, and it's more of a cadence of us asking them what they're looking for on what time frames in what regions and that informs us to go back out into the market to find the sites, whether it's 250 megs, 500 megs gig of deployment. That conversation cadence probably sits 12 to 18 months in advance.

And while we won't sign an agreement with them as we're signing the contract for the data center, like we've soft-circled our clients of like, all right, this client wants 250 megs in Q3 of '27. Let's go procure that site, and then we enter negotiations for passing that site through them for billable GPU hours.

Okay. And then when you think about -- you started to build out some of your own powered shells and started to do some of that development yourself. But the majority of what you're doing is vis-a-vis your partners and having a lease against that. How are you thinking about that dynamic going forward and the balance? How important it is to own both parts of the equation? Or does the lease get most of it done or enough of it done?

So I would say the important part of it is getting access to the active power, getting -- we referred to active power is something that we can step into and start delivering infrastructure out of, right. And what we care about most is getting access to that active power on a time line that our clients want. Because from there, I think we have the best execution in the industry of delivering stabilized supercomputers once it's been given to us, right? It's measured in weeks. I think, it would -- we would probably say like 4 to 6 weeks, somewhere in there on a deployment level perspective. What will that mix be? We're already so engaged in the engineering side of these sites, right? It's not like there's just blueprints out there like, oh, here's how you go build a next-gen AI data center, right? Like we're heavily involved on the engineering side and the deployment side that we're kind of already there. And I think we really benefit from continuing to build out that internal competency of build ourselves.

So we will remain opportunistic in the build versus lease approach. I think we'll have a healthy mix. I think we'll likely be bringing online some more self-development in there, especially in the context of 5 gigawatts, I think we benefit from being a strong component of that. But we're not going to offer like an explicit guidance because it will ultimately be all informed by customer demand, what the customer is looking for, and then we'll prioritize around it accordingly.

Got it. You mentioned the investor deck that you guys updated on Monday, and I would definitely tell everyone to look at it. There's a lot of really good data in that investor deck. You talked about the 25% contribution margins in a 5-year contract in months 4 through 60. I can't do that math in my head. It's probably more than that, it's more than 60. I really messed up that math. You know exactly what that 3-year through year 5 -- through 5 years.

Months 3 through 60, 25% contribution margin.

The entirety of the contract, I think you guys talked about like a 15% free cash flow margin. And that in and itself against a $67 billion backlog, super interesting. What's even more interesting is if you could continue to monetize that asset in year 6, 7 and 8. And I think that brings us back to one of the big debates around GPUs in the GPU economy is the useful life. What is the useful life of that GPU? Should we still be thinking 6 years? Should it be longer? Should it be shorter?

Six years. I think we've been very consistent about that. We're aligned with our peer set on 6 years. I think that is the absolute right number to be using for depreciation today.

It's kind of funny in our analyst bring down the call after earnings. We went through 45 minutes a group call of questions about model business, et cetera. It was the first time that no one asked about useful life. And this has been a persistent question for years. So like what is useful life, and I took a little bit of a sign that I think people are really coming around understanding that 6 years useful life is the correct number to be using. And there could be a little reality that useful lives may extend beyond 6 years.

And we -- the oldest infrastructure that we really have at scale from like an empirical data perspective is A100s, right? A100s are 2020 SKU. So that's putting it like right at 6 years right now. And we disclosed in earnings that in 2025, A100 pricing for us actually increased, right? It held its pricing power. H100s were within 10% of where it was at the start of the year, right? And I don't look at those as like a relative metric between each other. I look at those both as -- these are immensely strong demand signals that this older gen infrastructure, I think some portions of the market thought just went to 0 value after 2 years or something, is really not. And it's driven by this fact that there are use cases specific to the infrastructure, right?

And I think, it's driven by inference, so heavily inference mean inference is just exploding and the monetization of inference, I think that very real revenue and return on revenue is being driven off of that. we're signing 5-year contracts for committed take-or-pay use on the compute. 1 year of life that would need to be extended on the end of that, I think overwhelmingly, we are seeing the empirical support.

At 6 years, we would see no reason to change that in the near term, but we're quite excited about the prospect, especially from a margin perspective of useful life beyond 6 years. And I think that, that is a reality that we're about to enter for our business.

Got it. So NVIDIA has been dominating the GPU market, and the accelerator market for some time. And you guys have been very much an NVIDIA fleet there's more silicon coming into the marketplace. And there's a lot of buzz around GPUs right now. Is there any chance that we're going to see CoreWeave with other silicone anytime soon? And why or why not?

Yes. It's a question we've had for a while. And our answer has very consistently been we are client-led in what we build, right? We're not purchasing compute, building it and hoping people come and use it right? The path we take instead, which I think is a much more risk-adjusted path is -- we wait for the client to come in and say, "This is what we are looking for you to build." and this is how we enter CapEx and why our CapEx is so derisked when we enter it that way. The client, and it could be a little self-selecting just because we are so well known as an NVIDIA shop, but the client is asking only for NVIDIA, like we aren't getting requests for other types of silicon and make a mistake like we can run anything, right?

We're hardware agnostic as what we operate. And I think a really good example of that is this transition we went from Hopper into Grace Blackwell. For all intents and purposes, that is different infrastructure, right? It is an entirely different way to build and deliver compute once you get into an NBL 72 configuration versus just 42U air-cooled rack. Our software just adopted right along with that, and we were first to market, bringing H100 available to clients, J200. We're first to market with GB-200s,GB-300s that's all because of this software solution that's able to operate really any type of infrastructure underneath it, but again, we just don't get demand for anything but that NVIDIA infrastructure right now. We think it's the most performing platform that's out there.

Got it. One last topic I want to hit with you. We talked a little bit about the software, the enablement software that enables you guys to run these GPUs so effectively. But you've also built out a whole another layer of software on top to expand the capabilities of what you could do. And a lot of it through acquisitions, there's Weights & Biases, OpenPipe, Monolith, Marino, how should we think about the software strategy for that layer within CoreWeave? And is it something that becomes material to the story? It's hard because you guys have built up such a backlog of the GPUs.

Yes. It's something we have a playbook that we've all seen, right? The playbook of how do you build the cloud to run the Internet and host websites store data lakes. And the first step in that playbook was get the infrastructure right, get the foundation correct, allow for people to run their workloads on you efficiently and build a purpose-built platform for the highest performance possible. And then what do they do next. They started building apps on top of it. They started adding peripherals on top of it. And I think that's the exact same cadence that you see us in right now. Like we are the best operator of GPU supercomputer on the planet. I think that is an overwhelming agreement that's with our suppliers, our clients, the broad market.

Next step in that process is build app layers on top of it, add peripheral infrastructure around it. And you're seeing us go through that in real time. Absolutely correct. It's hard to say how all that scales. But I look back to our storage product, which is well north of $100 million in ARR that has that 80% attach rate with over 1 million clients, that product like came into existence out of nowhere. That scaled so quickly for us. So I think that speaks to how the velocity at which our clients are able to adopt these peripheral components, these software components that we're bringing around our core products, that's a really exciting thing for us.

Outstanding. It's been a rocket ship of the story. Congratulations on the success, and thank you for coming and sharing with us at the Morgan Stanley TMT conference.

Thank you so much. Appreciate it.

Hello, and thank you for standing by. My name is Tiffany, and I will be your conference operator today. At this time, I would like to welcome everyone to the CoreWeave Fourth Quarter and Fiscal Year 2025 Earnings Call. [Operator Instructions] I would now like to turn the call over to CoreWeave. Please go ahead.

Thank you. Good afternoon, and welcome to CoreWeave's Fourth Quarter and Fiscal Year 2025 Earnings Conference Call. Joining the call today to discuss our results are Mike Intrator, CEO; and Nitin Agrawal, CFO. Before we get started, I would like to take this opportunity to remind you that our remarks today will include forward-looking statements. Actual results may differ materially from those contemplated by these forward-looking statements. Factors that could cause these results to differ materially are set forth in today's earnings press release and in our annual report on Form 10-K to be filed with the SEC. Any forward-looking statements that we make on this call are based on assumptions as of today, and we undertake no obligation to update these statements as a result of new information or future events. 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 an accompanying investor presentation are available on our website at investors.coreweave.com. A replay of this call will also be available on our Investor Relations website. And now I'd like to turn the call over to Mike.

Good afternoon, everyone, and thank you for joining us. 2025 was a defining year for CoreWeave. We generated more than $5.1 billion of revenue, up 168% year-over-year, grew our contracted revenue backlog to $66.8 billion, an increase of $11.2 billion sequentially and more than $50 billion year-over-year. We reached more than 850 megawatts of active power as of December 31 and added approximately twice as many new reserved instance customers in Q4 versus any quarter in our history. We delivered these results while quickly resolving the data center delays we discussed last quarter, delivering the impacted deployments ahead of our Q3 earnings call expectations. CoreWeave is the fastest cloud in history to reach $5 billion in annual revenue. We remain in the early stages of the most transformative infrastructure build-out in history, and CoreWeave is at the forefront, building and operating some of the largest purpose-built AI clusters for the world's most demanding workloads.

Strip away the complexity and four fundamentals define where we stand: one, a demand environment that remains relentless, driving rapid adoption from an increasingly diversified set of hyperscalers, AI native and enterprise customers; two, expanding opportunities for new margin-accretive avenues to monetize CoreWeave Cloud, unlocked by the evolution of our platform beyond GPUs and the recent expansion of our partnership with NVIDIA; three, a rapidly growing data center footprint underpinned by unmatched execution and a strategic approach to capacity expansion; and four, a disciplined financial model deliberately designed to invest ahead of revenue to fulfill contracted demand backed by $66.8 billion in revenue backlog and providing strong visibility into durable cash flows, attractive returns, and the ability to drive down our cost of capital. We will speak to each of those today.

Regarding demand, the signals we are seeing across hyperscalers, AI natives, and enterprise customers are only intensifying as AI workloads get more complex, models scale faster, and adoption continues to proliferate. The breadth of this demand has translated to deepening engineering relationships with our largest customers and material progress on diversification. CoreWeave is supporting the next generation of AI pioneers. As I mentioned in Q4, we added approximately twice as many new reserved instance customers as any prior quarter in our history, including AI native and enterprise companies like Cognition, Cursor, MercadoLibre, Midjourney, and Runway. We also expanded our relationships with some of our largest partners, including both of our existing hyperscale cloud customers. Demand accelerated from each of these customer types, while pricing remained stable throughout 2025, trends that we have seen continue as we have started 2026.

In total, for the year, we grew the number of customers committed to spending at least $1 million on CoreWeave Cloud by nearly 150%—these are not one-time infrastructure deployments. They represent sophisticated multiproduct opportunities, the early chapters of enduring platform relationships, and a growth engine that compounds as AI becomes more deeply embedded in how these companies operate. We are also seeing a significant increase in demand for prior generations of GPU architectures, where supply also remains constrained. Average H100 pricing in Q4 was within 10% of where it started the year, while average A100 pricing increased in 2025. From our customers, we understand the demand for this infrastructure is largely for inference use cases, which are proliferating rapidly. We are signing this infrastructure into new reserved instance contracts ahead of when it becomes available, firmly putting to bed concerns about demand for older generation SKUs.

With largely all of our new 2026 capacity allocated, we continue to work diligently to expand our footprint to meet the overwhelming needs of existing and prospective customers for both near and long term. These trends reinforce our conviction in the durability of demand and the longevity of this technology while running on CoreWeave Cloud. In light of the insatiable demand environment and the persistent signals we are seeing from customers, we are accelerating our roadmap with the objective of adding more than 5 gigawatts of additional data center capacity beyond our already contracted footprint by 2030.

Moving on to new avenues to monetize CoreWeave Cloud. Our platform is evolving as we unlock margin-accretive avenues for growth through new products and services as well as offering our proprietary cloud stack outside of our data centers to the broader NVIDIA ecosystem. AI natives and enterprise customers are not just consuming our core GPU infrastructure. They're engaging with our unified platform at significantly higher rates across CPU, storage, software, and development tools. The opportunity to add additional value to these customers as their AI workloads mature is substantial and represents meaningful upside over time. This is already showing up across our platform. For example, approximately 80% of CoreWeave Cloud customers paying at least $1 million per year have adopted one or more of our storage products. Additionally, we are seeing strong cross-selling momentum with Weights & Biases as we added hundreds of millions of CoreWeave Cloud TCV from Weights & Biases customers in the second half of the year.

We have also accelerated the development of CoreWeave's proprietary cloud stack, reference architecture, and related software solutions, including Mission Control, which orchestrate every layer of our purpose-built cloud and increasingly define the CoreWeave customer experience. In January, we announced NVIDIA intends to test and validate our platform, including our software and reference architectures to work towards including those offerings within NVIDIA's reference architecture for cloud, enterprise, and sovereign customers. Already, we are seeing select customers license Mission Control as their default research cluster management platform across their multi-cloud footprint. We expect the broader distribution of our proprietary cloud stack to become a growing source of higher-margin revenue over time. The ability to monetize our platform, both inside and now beyond our own data centers through third-party licensing agreements substantially expands our addressable market. This represents tangible long-term upside potential that is not reflected in the 2026 guidance as we are providing it today.

On to execution. We ended the year with more than 850 megawatts of active power, adding approximately 260 megawatts in the fourth quarter alone across 43 active data centers, up from 32 at the start of the year. We contracted close to 2 gigawatts of additional power in 2025, ending the year with more than 3.1 gigawatts of contracted capacity, virtually all of which we expect to come online by the end of 2027. Our contracted but not yet active capacity represents latent revenue potential that we will monetize as built and delivered. We will continue to strategically source land, power, and data center shell infrastructure, particularly looking beyond 2026, we see robust opportunities in the current market for CoreWeave to grow our contracted power capacity and we'll also selectively leverage our expanded collaboration with NVIDIA to accelerate our roadmap further to better meet demand.

Operating at this scale and pace is inherently complex. When disruptions surface, we move decisively through disciplined coordination across teams and partners. We quickly cleared the delays discussed in our third quarter earnings call. And in total, we have now delivered more than 50,000 Grace Blackwells to the impacted customer, deploying servers on a rolling basis and delivering them within weeks of receiving access to the requisite data center infrastructure. We are delivering at this breakneck speed across several different sites, handing over tens of thousands of GPUs to different customers simultaneously. We believe CoreWeave is the only cloud platform that can move at this pace while providing the industry-leading performance and reliability that drives customers' trust and allows us to capture additional wallet share. The feedback and the results we are seeing from our closest customers is inspiring. Grace Blackwell running at scale on CoreWeave Cloud is revolutionary.

In Q4, we became the first cloud platform to reach NVIDIA's exemplar cloud status for GB200, while remaining SemiAnalysis' sole platinum-ranked AI cloud. We expect to remain at the forefront of execution and innovation across the AI cloud stack as we become one of the first to bring NVIDIA's new Ruben GPU platform to market in the second half of 2026, while expanding our product portfolio to include NVIDIA's Vira CPU and BlueField storage. The integration of these newer technologies into our proprietary cloud platform will help power new capabilities, including agentive workflows for our customers. The pace of our execution also explains why our capital expenditures for Q4 came in above guidance. Our teams were able to bring infrastructure into service ahead of our expectations, which we view as a high-quality acceleration of revenue capacity for 2026.

To put our current scale into perspective, according to third-party estimates, CoreWeave today is larger than the 15 largest Neo clouds across North America and Europe combined. Bringing more than 260 megawatts online in a single quarter requires simultaneously orchestrating hardware, networking, storage, and purpose-built software across more than 100,000 GPUs and millions of interconnected system components, all in near-perfect unison. This is among the most operationally complex undertaking in the technology industry. It is also what CoreWeave does better than anyone.

And finally, turning to our financial and business model. In 2026, we expect our CapEx will be at least $30 billion, more than 2x the CapEx in 2025. I want to frame that number in clear terms. This is a reflection of the extraordinary amount of contracted demand in front of us. Our revenue backlog has grown to $66.8 billion, and the vast majority of our intended capital deployment is to directly support this long-dated contracted demand, where we have direct visibility into our long-term margins, underpinned by durable cash flow. The dividends of these investments will compound, as you will hear from Nitin as he provides some commentary around our targets for 2027 and beyond in addition to our 2026 guidance.

This backlog will continue to be primarily financed with the asset-level delayed draw term loans that we introduced to this market. We expect to continue to reduce our weighted average cost of capital along the way while unlocking broader industry participation in the facilities. There is a diverse and growing demand to participate in CoreWeave's capital market journey. We have cultivated a phenomenal financing vehicle for our business that enables us to scale at the pace of artificial intelligence while staying on our targeted path to reach investment grade.

Before I turn it over to Nitin, let me leave you with a few final thoughts. We have $66.8 billion of contracted revenue backlog with every contract for new capacity expected to begin generating revenue by year-end 2026. We are delivering cloud infrastructure and converting it to revenue today. We are building and operating some of the largest purpose-built AI clusters for the world's most demanding workloads at a pace and quality second to none. The demand driving the build-out is relentless, diversified, and growing with customers engaging across our broadening product suite. Our contracted backlog gives us and you clear visibility into the trajectory ahead. The expansion of our collaboration with NVIDIA positions CoreWeave's platform as the natural destination for cloud, enterprise, and sovereign customers seeking optimal AI infrastructure performance inside and now beyond our own data centers. Our ability to see into the future of AI innovation and build towards its requirements is unmatched. We will continue to invest and grow this incredible market advantage with discipline and contracted cash flows as we deploy capital strategically to expand capacity, deepen our product suite, and develop the AI cloud that our customers demand. Our priority remains clear: deliver the most performant, reliable, and efficient AI platform for our customers at global scale. The market is accelerating, and CoreWeave is primed to be both the beneficiary and the enabler of the AI revolution. With that, I'll turn it over to Nitin.

Thanks, Mike, and good afternoon, everyone. Throughout 2025, we executed with discipline against the strategy we laid out for the year, beginning with our IPO. We significantly diversified our customer base, more than doubled our contracted and active power capacity, and strengthened our balance sheet by unlocking new funding sources while lowering our weighted average cost of capital. We also broadened our product portfolio, both organically and inorganically, successfully completing four strategic acquisitions to pull forward our roadmap. The current pace of the market and scale of demand for CoreWeave Cloud has created a clear opportunity. And in 2026, we are investing deliberately to meet it, accelerating our plans to further extend our leadership position.

Turning now to Q4 results. Revenue was $1.6 billion in Q4, up 110% year-over-year, driven by robust customer demand and exceptional execution. Full-year revenue was approximately $5.1 billion, up 168% year-over-year. Demand for CoreWeave Cloud continues to intensify with revenue backlog for the quarter ended at $66.8 billion, up more than 4x this year alone. As Mike noted, we made significant progress diversifying our customer base across hyperscalers, AI natives, and enterprises, a stated goal at our IPO last year. Moreover, the customers are committing their foundational AI workloads to CoreWeave for longer periods of time, resulting in the average weighted contract length increasing from roughly four years to roughly five years. We have $66.8 billion of revenue backlog with every contract for our new capacity expected to begin generating revenue by year-end. We are delivering on our commitments today. These commitments are being made to current and past GPU generations as a part of broader customer roadmaps with active conversations already underway for future SKUs.

Operating expenses in the fourth quarter were $1.7 billion, including a stock-based compensation expense of $157 million. We were able to deploy our data center and server infrastructure faster than expected while bringing online more capacity this quarter than any in our history. This drove the corresponding increase in our cost of revenue and technology and infrastructure spend. In addition, the increase in sales and marketing was driven by investments in scaling our go-to-market organization to capture the rapid growth of the AI opportunity. The increase in G&A was driven by professional services related to M&A and financing activities, public company costs, and additional headcount to support our growth.

Adjusted EBITDA for Q4 was $898 million compared to $486 million in Q4 of 2024, increasing nearly 2x year-over-year. Our adjusted EBITDA margin was 57%. Adjusted operating income for Q4 was $88 million compared to $121 million in Q4 of 2024. Our Q4 adjusted operating margin was 6%. Adjusted operating income was lower than expected as a result of deploying infrastructure ahead of our expectations. Net loss for the fourth quarter was $452 million compared to a $51 million net loss for Q4 of 2024. Interest expense for Q4 was $388 million compared to $149 million in Q4 of 2024 due to increased debt to support the scaling of our infrastructure. Adjusted net loss for Q4 was $284 million compared to $36 million in Q4 of 2024.

Turning to capital expenditures. CapEx in Q4 totaled $8.2 billion and $14.9 billion for the full year, higher than anticipated due to our team's ability to put infrastructure in service ahead of our expectations. As we previewed last quarter, the meaningful growth in construction in progress in Q4 to $9.4 billion, an increase of $2.5 billion quarter-over-quarter reflects the significant scale of infrastructure we are on track to deliver in the near term. As a reminder, construction in progress represents infrastructure not yet in service and not yet being depreciated. As these assets come into service, they will drive incremental revenue and corresponding depreciation. Our financing structure is designed to match this deployment model. The large majority of our term debt is structured as delayed draw facilities, meaning capital is only drawn as the data centers are operationalized.

While global supply chains remain complex amid persistent supply-demand imbalances, we have consistently navigated these challenges through operational discipline and strategic sourcing. Our track record of bringing infrastructure online at scale gives us confidence in our ability to adapt and continue accelerating capacity deployments in 2026 and beyond.

Now let's turn to our balance sheet and strong liquidity position. As of December 31, we had $4.2 billion in cash, cash equivalents, restricted cash, and marketable securities. We continue to make significant progress in strengthening our capital structure and lowering our weighted average cost of capital. In Q4, we raised approximately $2.6 billion via our inaugural convertible senior notes offering, where investor demand dramatically exceeded the offering size, leading to its upsize. We also expanded our revolving credit facility in the quarter to $2.5 billion to manage liquidity and support our various growth initiatives. In total, in 2025, we secured more than $18 billion of debt and equity, working with more than 200 investment partners and financial institutions, reflecting the depth and diversity of capital committed to CoreWeave's growth.

As Mike discussed, in January, we announced the expansion of our commercial relationship with NVIDIA, which was accompanied by a $2 billion investment in CoreWeave in support of our platform, team, and shared vision for the AI infrastructure at scale. Our efforts in Q4 and over the past year to optimize our financial structure and lower our weighted average cost of capital is evidenced by the 300 basis points decline in our weighted average interest rate during the year and represents a total reduction of nearly 600 basis points since 2023. To put that in perspective, the 300 basis points improvement represents nearly $700 million in annualized interest savings based on our Q4 debt balance.

Going forward, we expect to continue to be able to reduce our weighted average cost of capital as capital providers and rating agencies increasingly appreciate our best-in-class execution as well as the durability of and visibility into the cash flows that underpin our take-or-pay customer contracts. We have no debt maturities until 2029 other than self-amortizing contract-backed debt and OEM vendor financing.

Turning to tax. We recorded a noncash tax benefit in Q4, driven primarily by the impact of one big Beautiful Bill Act. Our tax rate might fluctuate significantly in the future due to similar factors. Demand continues to intensify and diversify across all customer categories. We are accelerating investments deliberately to capture the contracted demand and the long duration of those commitments give us clear cash flow visibility to deliver best-in-class cloud margins as the deployed capacity matures.

We expect 2026 CapEx of $30 billion to $35 billion, which is more than double our 2025 investment. Substantially, all of it is tied to our already signed customer contracts that we intend to bring online this year as we expect to double our active power capacity to more than 1.7 gigawatts by year-end. As I have described previously, when new capacity comes into service, data center lease costs, including power and depreciation expense commence, while customer revenue ramps over subsequent months. In 2026, this effect is amplified by the scale of our deployment program. We will be bringing online roughly double the capacity of 2025, which means a corresponding increase in depreciation running ahead of associated revenue recognition.

For full year 2026, we expect revenue of $12 billion to $13 billion, representing approximately 140% growth year-over-year at the midpoint. We expect adjusted operating income of $900 million to $1.1 billion. We anticipate margins will ramp sequentially from low single digits in Q1, expanding in each of Q2 and Q3 and returning to low double-digit levels by Q4 as deployed capacity matures and revenue scales against the existing cost base. Our 2026 margin progression is a result of deliberate investments we are making to meet the insatiable demand for our platform. As our business and growth normalize, we remain confident in our ability to achieve 25% to 30% margins over the long term. Our mature revenue contracts generate contribution margins in the mid-20s which combined with the ramp-up of margin-accretive products and services we continue to unlock gives us conviction in our ability to achieve this target.

Our 2026 guidance excludes any potential meaningful revenue or margin benefits from the further monetization of CoreWeave's proprietary cloud stack to other NVIDIA Cloud, enterprise, or sovereign customers, which we do expect to begin in 2026 and to become more meaningful in the coming years. This represents tangible long-term potential upside.

For Q1 specifically, we expect revenue in the range of $1.9 billion to $2 billion. We expect Q1 adjusted operating income between $0 and $40 million. Q1 represents the trough in our annual margin trajectory as we expect our CapEx deployments to be $6 billion to $7 billion of infrastructure in the quarter as we continue to bring online significant further capacity beyond the approximately 260 megawatts we added in Q4. Our Q1 interest expense is expected to be in the range of $510 million to $590 million.

The long-term nature of our contracted revenue backlog provides us with visibility well beyond 2026. As we continue on our hyper-growth trajectory, we expect to exit 2026 with annualized run rate revenue of $17 billion to $19 billion, which we expect to grow to more than $30 billion of annualized run rate revenue as we exit 2027. We are not building towards this trajectory speculatively. Contracted customer demand, deep strategic partnerships, active infrastructure deployment, industry-leading capabilities, and a thoughtful approach to capital markets give us the confidence to put these numbers forward.

We delivered a strong fourth quarter and full year, capping a transformative 2025. We grew our contracted revenue backlog to $66.8 billion while meaningfully diversifying our customer base, secured more than $18 billion in debt and equity capital at progressively lower costs, and strengthened our platform through new products, services, and strategic acquisitions. We entered 2026 with 850 megawatts of active power across 43 data centers, on track to exceed 1.7 gigawatts by the year-end with every contract for our new capacity expected to begin generating revenue this year. Our 2026 investment program is fully supported by contracted demand. And as we noted, our guidance excludes the potential upside from licensing CoreWeave's proprietary cloud stack, which we expect to begin contributing in 2026 and scale in the years ahead. Thank you. We look forward to your questions.

[Operator Instructions] Your first question comes from the line of Keith Weiss with Morgan Stanley.

This is Josh Baer on for Keith. Congrats on a good quarter. You came in nicely ahead on CapEx, and it's great to hear the delivery delays resolved quicker than expected. Trying to align that with seeing the active power, which is more in line and revenue guidance in the range, which is well below like the typical level of upside. So I was hoping you could unpack some of those dynamics. If you're moving faster, why didn't that show up in the Active Power and the revenue? Maybe it did.

Thanks, Josh, for your question. As we deploy capacity, a lot of that capacity came online towards the end of the quarter, and you're going to start seeing the monetization of it in 2026. We continue to build our capacity at a rapid pace. As we talked in our prepared remarks, we will continue to deploy that capacity for 2026 throughout the year as well, including Q1, and that's the impact that you're seeing. Relative to the Q1 number, like we are basically providing the guidance for the first time for 2026 and Q1 at this moment.

Okay. And really great to see the list of enterprise customers. I was hoping you could unpack what that type of contract and deal looks like from those enterprise customers. We have a great sense for what a hyperscaler mega contract looks like. But any chance you could run through size, duration, prepayment, pricing associated with those enterprise customers?

So we don't speak to individual contracts. But what you're seeing is in an environment where there is so much intense competition for the product we deliver, which is the most cutting-edge computing infrastructure delivered through CoreWeave Cloud. The contracts largely look very similar to the hyperscale contracts in tenure. And we work with each of the individual enterprise clients as we're putting together an appropriate structure for their business model and for their clients.

Your next question comes from the line of Amit Daryanani with Evercore.

I guess my question is really around the cost of financing, especially given the $30 billion plus kind of CapEx number we have for the year. I'm just wondering, as you continue scaling capacity, can you sort of quantify where you estimate your blended cost of capital is? How has that really evolved over the last 12 months? And when you negotiate with these data center operators, how do they assess your credit profile? Is it really tied to your customer contracts and who they are? Or is it something else? And then just on the financing side, does the NVIDIA credit support, the guarantor framework help translate into a measurable step down in your borrowing cost, you think, in '26?

Yes. Thank you for the question or questions. So look, we've made incredible progress at the company as the company matures as a business, as we have more extensive track record of operating this infrastructure, working with the client, delivering infrastructure. You've seen our cost of capital drop 300 basis points in the last 12 months. You've seen it drop 600 basis points over the last 2 years. We expect that, that will continue. It is a trend that's being driven by our business increasingly performing well with these ETL structures. When you're talking about the data centers, we added close to 2 gigawatts worth of infrastructure in 2025. And just to give you scale and perspective, at the end of 2024, we had 1.3 gigawatts in our portfolio. So you've seen a material increase in our capacity to access, build, and drive data center contracts. And we're excited about that. It's an important stepping stone for us as we continue to kind of drive the business. And once again, the ability to enter into contracts with that scale of data center capacity is once again a reflection of the business maturing, the creditworthiness and scale of the business increasing.

As far as the data center operators go, I can tell you what I think, right? And what I think is that data center operators are very interested in working with CoreWeave. They're looking for a diversified portfolio of tenants in their data centers. They're looking to go ahead and get exposure to a company like CoreWeave that represents so much of the AI infrastructure that's going to be ultimately delivered. And so they kind of look at us as a pure-play way of really getting access to the scaling of artificial intelligence, and they want that exposure. As far as our relationship with NVIDIA in terms of accelerating our ability to get access to data centers, I think the perspective that you should take here is that, obviously, working with an investment-grade counterparty as the offtake will have an impact on the cost of capital or the cost that is associated with the data center. Obviously, working with NVIDIA, which we do selectively, but certainly not exclusively when we're building out our data center portfolio will have a positive impact on the costs associated with our data center footprint.

Your next question comes from the line of Mark Murphy with JPMorgan.

Congratulations on just very, very strong bookings. Mike, some of the AI models have demonstrated a pretty gigantic leap forward in the last couple of months. And the one that's in the headlines is Claud Code. But I don't think we have seen models yet that were fully deeply trained on some of these gigantic Blackwell or GB200 or NVL data centers, really the stuff that CoreWeave has pioneered and mastered. I'm curious what you're hearing in the marketplace just in terms of how those Blackwell-based models are coming along. If we end up seeing GBT6 or any of the other ones in the next 3 to 6 months, do you think it's going to feel like a huge step forward in their capabilities? Or is it looking more like a steady evolution on the Blackwell systems?

Look, the Blackwell systems are amazing, right? They represent the next step function in computing power that allows these data scientists, these companies that are driving the models to be able to build and scale infrastructure in a way that they just haven't been able to historically. And my expectation is, and certainly every indication from the model companies is that the rate of increasing performance from these models, we're just getting going. Now it is early in the deployment of Grace Blackwell, right? Like there are not that many clusters that exist at the size and scale that we talked about, we have already delivered. As those clusters come online within our portfolio, within the global portfolio, I think it stands to reason, and you will see step functions in performance that are associated with this new technology. We're really excited about it. And I think our customers are extraordinarily excited about it because they understand what they're going to be able to do with this technology that is so incredibly performing, both from a training perspective, but also from an inference perspective when it becomes available for them.

And Mike, just by extension because you just said they were inference. How are you weighing the merits of focusing on the NVIDIA reference architecture? It's obviously very powerful for the massive training runs and some work beyond that. Just the other side would be any inclination to work with custom ASICs that they do legitimately seem to offer better inferencing price performance. And then I'm -- then obviously, NVIDIA's acquisition of Grok maybe kind of -- I don't know if you think that, that sort of resets the playing field in a way that seeing NVIDIA reference architecture might kind of reign supreme even for inferencing. I'm just wondering how you sort of project that forward.

Yes. So look, whenever we have these calls and whatever I'm asked about this, I kind of speak to the way that we've gone about building our business, which is we are client-led. Our clients are coming to us and they are telling us that the infrastructure that they need in order to drive their business. And I want to be clear that when they say infrastructure, it's not a training infrastructure, it's not inference infrastructure. It's AI infrastructure, right? And they're coming to us specifically because we're able to deliver such an incredibly performing of the NVIDIA technology. They know we're great at it. That's why they come to us. Are they looking for other technologies from other providers? That stands to reason. But what I believe is that or what I know is that we are unable to catch up with the demand signals that are coming in for the product that we deliver. And so we are going to focus on continuing to drive the solution that we have that is so performant and that has overwhelmed our ability and the market's ability to deliver infrastructure for the past 3 years.

Your next question comes from the line of Brent Thill with Jefferies.

Nitin, I had a quick question just on the guide. And just from a perspective, I know when you look at the revenue guide, you were in line, op income a little lower and your CapEx was way higher. I guess it just -- it kind of illustrates even the guide you gave us all that some of the metrics can really vary. I'm just curious just in terms of how you're thinking about the guide going forward. Are some of the variables out that you've taken out from maybe what you saw in Q4? Are those variables taken out? Or has your guidance changed a little bit? Again, I know this is incredibly difficult to make an estimation, but some of the numbers were effectively kind of outside the range of what you initially gave us.

Thanks, Brent, for your question. So from a guide perspective, let me break it down by a few variables here. We talk about the CapEx numbers that is fundamentally in service of our contracted customer backlog, which we disclosed this quarter to be at $66.8 billion. And that's what is driving the investment in our platform. And when you think about the revenue ramp, we talked about that as well that almost all of our -- most of our -- all of our contracts that we are -- our backlog would start generating revenue in this year. So that's the ramp that you are seeing. We delivered 850 megawatts of power in this fiscal year. And for the year, for the 2026, we expect to be at 1.7 gigawatts of power. When you think about margin, when all of this comes together in margins, our margin progression effectively is a result of these deliberate investments that we are making to meet the insatiable demand that we have in our platform. We talked about Q4. Q4 alone, we brought 30% of our total active power base, which naturally creates some near-term margin compression as capacity costs ramp ahead of full revenue maturity and recognition. As I mentioned in my remarks, Q1 represents the trough of what we would see in margins. And then from there on, as we scale into the capacity deployed, we will expand margins quarterly from there, returning to low double digits by Q4. We also talked a little bit about long-term trajectory of this business. Our strategy and our management philosophy has continued to be to invest in terms of customer demand with contracted backlog, which is what still continues to be the case. Over the long term, how it manifests itself in our business as it growth normalizes, we remain confident in our ability to achieve 25% to 30% margins. The factors that give us confidence in that domain, if we look at our mature fully ramped contracts and that portfolio, that generates contribution margins in the mid-20s. We continue to ramp up our margin-accretive products and services in our product portfolio. For instance, we had announced in Q3 that our storage revenue on our platform eclipsed $100 million in ARR. Today, we also discussed how attach rates for storage are now at 80% in our large customer base. While not included in our 2026 guidance, we see tangible long-term upside potential from further monetization of CoreWeave's proprietary cloud stack to other NVIDIA Cloud, enterprise, and sovereign customers. So what you're seeing in 2026 is a reflection of the acceleration of the growth in the back of our existing backlog, which continues to grow with customer demand.

Yes. So I just -- I wanted to add a couple of things here, right? And so our margins reflect the cost of building tomorrow's revenues, right? That's what we're doing, right? As Nitin said, the fundamental margins at a stabilized facility are in the mid-20s, right? And as we continue to build our infrastructure, as there is more infrastructure online, that will come to bear. The variance that you're seeing is a function of how much infrastructure we are bringing on versus the installed capacity. We bought on 260 megawatts worth of power in Q4. It's fully 1/3 of our installed capacity. So the variance that you're going to see there is higher. As we continue to grow and scale our company, as the incremental data center capacity that we bring on becomes relatively smaller, you will see less variance from us, right? This is what the acceleration looks like, making the decision to go ahead and invest to pull in tomorrow's revenue to be able to serve our clients is a fundamental strategic decision that the company made. And we're doing this extremely responsibly because we're not doing this, we're going to build it and they're going to come. We're doing this leaning into the backlog of contracts that we have already sold.

Your next question comes from the line of Gabriela Borges with Goldman Sachs.

Nitin, I wanted to ask you about the diversity of customers that you have on your platform. I'm curious if you can share with us your observations on how the unit economics or how the attractiveness of how these customers are using the CoreWeave platform is different between types of customers. So a little bit of a broad question. I know that your pricing model is based on dollars per GPU and then the length of the committed contract. But curious if you could share your observations on customer behavior across the different cohorts.

Yes. So I think a lot of this depends upon the variables that we've talked about in terms of how we structure our contracts. The term length, the amount of upfront payment, the generation and the demand for that capacity at that moment all dictate into it. Fundamentally, as Mike described, like our contracts look mostly similar across our customer base with the exception, of course, the volume element that we look at things when we are talking about larger customers versus smaller customers. But across the board in our customer profile, we look to generate similar economics for the infrastructure that we are generating as with the market dynamics go on. Mike talked about how for hoppers, we are continuing to see incremental demand and demand of recontracting those hoppers at about 10% of the original ASPs when they were first contracted to A100s where the ASPs are actually increasing as we write newer contracts. Those dynamics are broader market dynamics, but across our customer base, the economics kind of look relatively similar.

The only thing I would add is that one of the things that I do think is very exciting is that many of the enterprise customers, many of the smaller AI native customers, they're coming on to our infrastructure, and they have the ability to use the H100s and the A100s. Their ability to build product and to be able to serve inference with that infrastructure is a really wonderful sign of the depth and resiliency of the bid that is looking for compute. It's new use cases, people doing things that we've never seen before. It's really exciting.

Your next question comes from the line of Ben Reitzes with Melius Research.

I wanted to ask the other side of -- I believe it was the Brent Thill question. One of the things going on in the market is there's all this CapEx and not enough margin or cash flow necessarily in the near term. And so I think that's what he was getting at is that you guys were talking about your margins being 20% to 25% over the long term. But -- and then your confidence to get there. What is your confidence in the rate of the CapEx growth? Like we understand you're spending now to get the revenue -- but what is -- does that growth rate moderate more than we're thinking as those margins go up? Do you need to keep spending at this kind of upside versus the Street on the CapEx side in order to hit those numbers? Just a little bit more color on the CapEx side that balances Brent's question on the margin would be really helpful in the rate of that trajectory.

Yes. Thanks, Ben. That's a good question, right? And I think it's really important that we deconstruct that question into at least two pieces, right? The first piece is that our business is built on a success-based model where clients come to us and buy long-term contracts to get access to the infrastructure to build their business. right? And so when a hyperscaler comes to us and buys infrastructure for five years, they are going to be purchasing that infrastructure from us at a fixed price for five years. And that is a very stable way to go about building our revenue and our margins. And that's where Nitin gets the confidence around the margins from is because we know not only what we're going to make now, but we know what we're going to make as those contracts move through the five-year cycle that they have been put under a take-or-pay contract. And so that is a very stable way to go about building a business. It's a very stable way to go about getting access to the capital markets so that we're able to finance the builds. right? And that's a fundamental building block of how CoreWeave has gone about building its business really since the beginning.

The second question kind of, I think, embedded in that is what is our confidence interval that there will be a continuation of demand for computing infrastructure. And this is a great opportunity for me to talk a little bit about what we're seeing in the demand stack because the demand stack is actually fascinating, right? Not only are we seeing the proliferation of demand across the economy, going from where it was initially really housed within the hyperscaler clouds and the foundation models. You're now seeing it kind of explode into the enterprise, you're seeing it move into sovereign. You're seeing all these new participants beginning to come in and securing the infrastructure that they need. You're also seeing a really fascinating component where it's moving from just the GPU, which was the initial wave of demand that we used to launch our company, but really now starting to move out into storage, into CPUs into -- and that is really a function of the portfolio of clients that are using our infrastructure really extending into the application layer.

And so as far as the demand goes, we've got our fingers on the pulse in a way that very few companies in the world have. We are getting information fed in from across the entire economy as people are trying to get access to the infrastructure that they require. So we're very confident in our contractual position relative to the portfolio of clients that we have. And we're very excited about the portfolio of different types of clients, and we're very excited about the different types of compute that they're using. And so it's really threefold of information, threefold of confidence around the drive towards accessing compute via CoreWeave.

And a couple of data points that I'll just add is we discussed this in our prepared remarks. For our 2026 CapEx, substantially all of it is tied to already signed customer contracts that we intend to bring online this year, and we expect to double up deployed power capacity. So that's one data point. The second data point that I want to make sure that you get is from an EBITDA margin perspective, I know you talked about where is the cash associated. The EBITDA margin perspective, we generated 57% margin this quarter. When I think about the long-term contracted customer contracts that we have and those scale customer contracts, we talked about they are in the mid-20s from a contribution margin perspective. When you look at the EBITDA margins for those, they are in the 70 percentage zone. So that is something that gives us confidence in terms of when our contracts scale, they generate a lot of cash for us.

Your next question comes from the line of Brad Zelnick with Deutsche Bank.

Congrats on an absolutely amazing year. Just guys, as we look to Rub contracts ahead, are you seeing demand for similar five-year duration? And is there anything different about how these deals are priced, the amount of prepayment that you'd expect or anything else? And how should we think about the ROIC on these deals versus prior generation contracts and the economics that you've outlined in your original S-1?

Yes. So thanks. It hasn't even been a year yet, although it does feel like it. I think we've got 11 months under our belt. So look, in the written statements that we made, we talked a little bit about the fact that one of the trends that we're seeing is the extension of the contracts from an average of four years up to an average of five years. And obviously, that is great for us. It's stabilizing for our business. It gives us a lot of confidence around building and scaling the infrastructure. And so that's sort of the first thing I'd highlight to you is that the clients that are coming in and using this infrastructure, they're gaining more confidence in the longevity of the usefulness of the infrastructure, which once again gets back to many of the important components that are required for us to build our business. The pricing on the infrastructure, we kind of take a -- we think about that from a margin perspective. And as the infrastructure changes in price, we are altering our pricing of how we deliver it to clients in order to kind of target the type of margins that Nitin spoke to. And so like those are some of the trends that you're seeing. Now as far as prepay goes, prepayment is always a lever which we can use with clients to change the economics around a contract. One of the things that's very exciting for us is as we continue to drive down our cost of capital, our dependency on prepayments is reduced.

Your final question comes from the line of Michael Turrin with Wells Fargo.

This is for Nitin or Mike. Can you just speak to what gives you confidence in the $30 billion run rate by 2027? And how much of that is already booked versus business your team needs to go get? And maybe as a second part, just if you could speak to any change in demand you're seeing, Mike, you touched on some of the segments of the market, but just any change in demand you're seeing across those segments and how you prioritize across those as well?

I'm going to do it in reverse order, if that's okay. So one of the things that we as a business are very interested in is making sure that we have a diversified perspective on what the compute is being used for. And so we're really out there working with everyone that consumes compute in every way that we can because we feel like that gives us the best view on where the demand is going to come from. And I've said this before, one of the things that's going to happen here is as compute becomes more available, you're going to see businesses that don't even exist yet, ideas that don't even exist yet have an opportunity to come into existence. Those will be new clients of ours. And we want to be able to pick them up right at the beginning as they go ahead and build these new incredible businesses.

Moving on to your question around the $30 billion run rate, like what we are doing is we are taking the contracted power that we have, and we are projecting out when the existing contracts that have already been sold, and like I said, we are virtually sold out in 2026 of all of our capacity and then continuing to add contracts that will be allocated once they come online in 2027. And we have vast and sustained interest from our clients to get more capacity to bring on more compute. And these are some of the largest, most creditworthy companies in the world. These are some of the most important AI labs in the world. These are the people that are building the AI future, and they are trying to secure infrastructure through CoreWeave, and it's really exciting. And when you move through that exercise, we have a lot of confidence in the $30 billion number that we've put out there.

All right. So like I said, we've got 11 months in here, and I appreciate all of you working with us as we've built this company. So as we wrap up here, I want to thank the CoreWeave team and our partners. None of these accomplishments would have been possible without you. I'm incredibly proud and humbled of the execution across the organization from product velocity and innovation to operational excellence and financial rigor. The focus and intensity across our organization is what enables us to continue our hyper-growth trajectory and the size of the incredible opportunity that lies ahead. Thank you all for joining us today. We appreciate your support, and we look forward to updating you in the future. Thank you.

Ladies and gentlemen, this concludes today's call. Thank you all for joining. You may now disconnect.

Okay. Great. Thank you, everybody. I'm Karl Keirstead. We've got Tim Arcuri, who covers NVIDIA and honored to have Nick Robbins of CoreWeave here. Today has been a fun day for me because this has been the theme of today, partly by design, but we had Navios on stage earlier today. We just got off stage with Crusoe and Lanci Lancium who are building out the next star gates.

Now we've got Nick to talk through CoreWeave story shortly afterwards, I think in a couple of hours, we've got Blue Owl Capital and Magnetar to talk about how they're financing all of this build-out. So this has been a fun day actually to go deep into this whole GPU cloud build-out.

Did you want to start with -- Nick's got some profound words he'd like to share at the very start.

Yes, this is the CYA. All right. Before we get started, I would like to remind you that CoreWeave may make forward-looking statements during today's fireside chat. Actual results may vary materially from today's statements. Information concerning risks, uncertainties and the other factors that could cause these results to differ are included in CoreWeave's SEC filings. We did it.

Thank you, Nick. Let's chunk this up a little bit. Let's talk about some demand-related questions. We'll talk about some supply issues as you're scaling up the infrastructure. Tim is obviously going to want to talk to you about GPUs and CPUs and stuff like that. But let's start on the demand side.

So obviously, CoreWeave put up 134% revenue growth this past quarter. Your backlog is $55 billion. That's 10x your revenue run rate. Things appear to feel pretty good. You've begun diversifying more away from the earlier high concentration with Microsoft. Maybe, Nick, I can ask you to just describe to some extent, the demand you're seeing today. I mean it feels phenomenal, but you can go a little bit deeper.

Yes. I think the words we typically have used over the recent past to describe it have been -- they've ranged from insatiable to relentless to tremendous. It seems that we have, even within this year, kind of had a couple of step functions upward in demand, where I would say to start the year, demand was pretty relentless and then we found ourselves over the summer where seemingly all of our customers wanted -- or most of our customers wanted a whole lot more and a lot of potential new customers wanted a whole lot more.

And yet where we find ourselves today, as we kind of said on our earnings call a couple of weeks ago, is it seems like that's taken another step upward, right? It's -- I think as we see more use cases for the compute that are delivering ROI that are transforming business industry world, right, more people want more. And frankly, we continue to find ourselves in the position of how do we bring on capacity quicker to service this demand because that continues to be the constraint on growth for our business.

Okay. Let's talk a little bit about one layer deeper and your judgment on where that demand is coming from. Is it that the model providers woke up and realized the merits of reinforcement learning, they needed more compute. Is it that everybody underestimated, let's say, the compute intensity per prompt on consumer AI products like ChatGPT, therefore, you need a lot more inference compute. What was the trigger for these step-ups in compute demand this year as best you can estimate?

I don't think there's a single one. I think it's many things all in one. I think scaling laws are continuing to hold in a way, right, where even on a pretraining basis, you want more. I think the proliferation, right, of post-training and other types of training that are very compute-intensive and hungry, right, catalyze or if like further accelerate that. And then on top of that, yes, you've seen -- from an inference perspective, the world see innovation there that is also more compute-intensive, namely reasoning models or change of that, right?

And I suspect that as you see more AI labs and enterprises bring AI product to market, those products will continue to become more compute-intensive in hungry. And so you have all of these tailwinds pushing to the same bottleneck of -- they all point in the same direction of more compute.

Okay.

Maybe a year ago or 14 months ago, I think the world very simplistically thought of training is compute-intensive, you do it once, then you just ship a model, and it's just about latency and inference is very compute light. And it's just -- you ask a question, get an answer, right? Like 14 months ago or so, I think 1 was released and the world saw a new vector of inference. And I think the proliferation across those 3 things is the answer.

What about demand from a customer cohort? Obviously, the frontier model providers have been fueling a lot of this compute demand, much of it. But now we've got a number of smaller AI natives, the pool sides of the world, Cursor, et cetera, fueling demand. And at some point in time, the enterprise side will kick in. So where are we in that demand curve for -- maybe you think of it differently, but frontier model providers, AI natives and traditional enterprises like UBS?

Sure. I would say from a frontier AI lab perspective, it is more pleased, and it's a question of how do you access that greater compute and continue to grow your footprint, right? And I think that the runway there is very long based on where we are today. I don't -- maybe it won't always do this way, but it almost feels endless. I would say on the enterprise side, we are undoubtedly earlier in the journey, right?

I would say it's clear with the -- away from hyperscalers and away from how folks like Google and Meta are clearly using AI and GPU compute to reaccelerate growth and drive returns in their business. But when we think about the UBSs, I think clearly, there's going to be a role to play for AI labs who are helping productize AI and sell and the growth that has been unbelievable, right, of OpenAI and Anthropic and businesses like that, that are adding billions and billions of billions of ARR in a given year at an unprecedented pace are validating that.

But I would say enterprises are going to need some help to get there, right? We made an acquisition of a business called Monolith a couple of months ago. That was really centered around helping bring AI to the physical world. And what they do and what we'll be doing together, right, is focusing on more compute-intensive industries like industrials and in Monolith's case, autos that they have mature workloads and they have a clear use case in this -- one of them, for example, being battery optimization and experimentation, but they don't necessarily, you have the product to do it. And so -- we think about how we can deploy our own software and services to help accelerate that for some of the older world enterprises, while also serving the AI labs that are clearly driving that market ahead.

Okay. I'll ask you two more and then we'll turn to Tim. On the supply side, Nick, obviously, on this last call, you highlighted some pushouts. Now that's probably -- I'm sure standing up gigawatt scale AI campuses is one of the more complex go-live supply chain problems that anyone will ever see.

So the notion that there could be a quarter delay due to partner delays, I don't think it's shocking to anybody. But can you describe perhaps your confidence in hitting those supply targets? And if there's any new supply bottleneck that seems to be bubbling up that might be interesting to flag for the group?

Yes. And -- you're absolutely right. Like the scale at which and the pace of which these data center and AI campuses are being built is just simply unprecedented, right? And there's not a playbook, right? Like the joke -- I don't know if it's joke, that I like the analogy I like to make is like you're asking someone to build like the Death Star LEGO set without the instructions, right? Like it's just like it's -- you don't know what you're doing, you figure it out, and it makes you better for the next one.

I would say the supply chain issues that we're seeing are not new. We've been pretty artfully, I think, navigating those for the last couple of years, and we will continue to. And it's not a single thing, right? People ask, is it that there isn't enough labor? There isn't enough labor. People ask about long lead time equipment. There isn't enough long lead time equipment, and it's called long lead time in part because it takes a while to get it, right? And so we're seeing the confluence of those things. But again, we've scaled out clusters that now, I think, at 41 data centers across the North America and Europe in the last several years alone.

We've been doing that while working through this and we will continue to do so. When it comes to the fourth quarter, we got on our third quarter earnings call, we were pretty disappointed that we had to revise guidance because something was slipping bit versus kind of the expectations, we had underpinning our guidance. A matter of weeks, I can count on my hands, right? But when it slips from quarter-to-quarter, it appears to be pretty acute.

We talked about how the vast majority of that comes online in Q1. We are tracking very nicely versus what we said a few weeks ago on our earnings call. But I would say how do we keep working through it. I would say, self-build helps a bit. We've talked about our first couple self-build projects in Kenilworth in New Jersey and Lancaster, Pennsylvania. An externality of that has very much been that we've been investing this year and meaningfully growing our data center teams, data center technicians, project managers, et cetera.

And that enables us to not only have more of a self-build capability, which is a small minority of what we do today, but it also allows us to have more boots on the ground in more places and take our own view of the timing and development of those sites, such that we can more transparently and accurately communicate those to our customers and folks like the people in this room.

Okay. Let me ask you one more. We'll hit you with a hot subject. So that's the AI bubble concerns that people have on their minds because I'm listening to you and you're talking about insatiable demand, you're expressing confidence in standing up significant amounts of supply. Yet we can contrast that with some thoughtful investor concerns that you and the whole GPU cloud infrastructure business are overbuilding, that the demand is not going to be there, Nick, in 3 years' time. So take that head on and give us your rebuttal.

Absolutely. First and foremost, disagree with the assertion that there's a bubble, right? Like when we look about -- look at the pace at which AI is being rapidly adopted and monetized, right? And I talked a few minutes ago about like the tremendous and like overwhelming growth, it seems that -- of course, we're experiencing, but that's a direct output of leading AI labs and how quickly they're growing.

We're seeing what some -- how the hyperscalers are monetizing, and we're all saying there's just simply not enough capacity, right? I think people like to make a comparison to some other things like the dot-com bubble, where it's like -- and other people can yell at me if they disagree. But like from my perspective, it doesn't seem like you were really ever hearing in 1998 or 1999 you don't get it. There's just not in the fiber, right? Like it was kind of quite the opposite. We're building for something on the come.

We are just trying to build what our customers are demanding from us right now. We are signing longer dated take-or-pay contracts for capacity that comes online in the next 9 to 12 months that they are committing to use for the next 5 years, right? The cash flows from that are paying for that CapEx. They're naturally deleveraging to pay down our debt and delivering us free cash flow, right? And that's leaving us in a position where we will have effectively depreciated and paid for infrastructure some cash flow and the opportunity to continue to monetize what we think is the most performance solution in the market going forward.

But it's just a very little of what we do and nothing of what we do on the GPU CapEx side is speculative. We are building to demand. We are not trying to outpace it. In fact, we are struggling, and I think a lot of the hyperscalers have said, they are struggling just to keep up. I think that is a very different paradigm than some of the other kind of -- situations people claim to be analogous.

Yes. Got it. Okay. Helpful, Nick. Over to you, Tim.

Great. Thanks, Nick. So another debate you hear about is vendor financing. And you signed this deal with NVIDIA. And at the time, people thought, well, that's vendor financing, but it's not at all really. So can you actually talk about that?

Yes. It's -- it was -- you're absolutely right. And it came in -- we got some other questions about it, too, just based on headlines. And I think it's misunderstood, so thank you for asking.

We announced a $6.3 billion partnership or collaboration with NVIDIA, I think, in September. And that looks like a customer contract, right? Like virtually all of our other customer contracts, save for a key exception, which is this concept of interruptibility, which allows us to say, NVIDIA, we're pausing your access to compute. We know you have use cases for it and you would love to have it, but we're going to pause it and we are going to go sell it to somebody else.

Who that other -- somebody else is in this instance is the -- think of the small- and medium-sized companies or smaller AI labs that are not in a position today to commit to the 5-year contract that we like require to go buy the server for a frontier scale cluster, right? And that means that the barrier to entry is pretty high because if you can't -- if we don't think you can afford it, we're not going to go buy the servers and build the cluster on your behalf and give you the capacity, right?

And so what we're seeing here is like it's almost like a product, right, where it's -- it's a win-win-win for us, for NVIDIA for the end customer who we interrupt NVIDIA with where, hey, if no interruption happens, which is unlikely to be the case anytime soon, that NVIDIA gets compute and they have a use case for it. But more importantly, we are able to go service and acquire the smaller customer who wants to be on our platform that we want on our platform that NVIDIA would, I'm sure, love to have on our platform because we do deliver the most performing compute in market, but that you just can't afford it.

And so they get access and we don't sacrifice our discipline around CapEx, while being able to go out and acquire a customer that otherwise the barrier to entry would be too high. And hey, knock on wood, some of those customers grow and they graduate into their own direct contracts because they've scaled, but a big part of their scaling is likely because they have that access that we have unlocked through that partnership.

Right. So it's not like they do a deal just for the sake of moving GPUs. This is for the sake of expanding the TAM.

Absolutely, expanding the ecosystem and reducing that barrier to entry for those long tail of customers who aren't the world-leading AI labs who have seemingly limitless access to capital.

Great. And then another question that -- one thing that I like to look at to determine the supply-demand balance is I'd like to ask companies like CoreWeave, what the pricing is for the oldest instance that's being used for AI workloads, which in your case is Ampere?

Yes. It's certainly Ampere and then it's L40 and Hopper, right? And the pricing for each of those, Mike went on to you the day after earnings, in fact, we said we're virtually sold out of Hopper of Ampere of L40. And when you look at the pricing quarter-over-quarter-over-quarter, it's proven to be. We had our first, and I'll highlight it's our first. It's not an example. It's the only example we got of a large scale, so call it, 10,000-plus H100 cluster begin to approach contract expiry. So it's probably 2 quarters in advance.

And the end customer there said, we want to go recontract this out for an extended period of time and they like using that for an inference use case. They saw a really attractive ROI on that cluster and said, we can -- we are comfortable continuing to pay a very similar price per GPU hour because we understand the ROI, it is tangible, and it is attractive to us. So let's go keep doing that.

Got it. And maybe just one last for me. There's this notion that you're going to wait around and you're going to pick and choose generations from NVIDIA. You're going to skip over one generation, not just CoreWeave, but just customers generally. How do you -- this idea that customers will, "Oh, I'm not going to buy Ruben, and I'm going to wait for Ruben Ultra, or I'm not going to buy Blackwell, I'm going to wait for Blackwell Ultra." How do you kind of think about that?

Yes. I think our experience has been more of highest end of we're going to buy some of this, and we're going to buy some of that, right. And we are pretty deeply entrenched with our customers and understand like, hey, you want to buy some Blackwell, but you're not going to buy everything in just Blackwell right now because you know you're going to buy some VR later, right, and probably some fine after that. And so we take pride in being the first to market with seemingly almost every generation of NVIDIA GPU technology over the last couple of years.

I would expect that to continue based on the customer conversations we have. We think that the demand is pretty rampant across generations and people are being thoughtful around how much they buy this with that in the future. But we haven't seen someone say, "Oh, I'm going to skip over this one." It's more, I would say, of a timing perspective of like let's say we're beginning to talk about a data center that might come online in early 2027, then the customer has to make the decision of do I want to keep working with Blackwell because by then, I'll have scaled Blackwell clusters of configured software, by engineering, we'll devote time to it? Or do I want to get started and push that to VR. But haven't seen it as much as one or the other as much as how do we sequence the timing of everything.

Got it. Back to you, Karl.

Yes. I'll ask you 2 or 3 more, and then there might be time for 1 or 2 more for you, Tim.

So on the fungible infrastructure question, let's dig in that a little bit because all of us listen to Satyam's pods. And he's very fond of saying that Microsoft is building a super fungible infrastructure, not for any single customer, any one location any single workload type, but others are. And therefore, they're absorbing a lot more risk than we are. Can you comment on what CoreWeave is doing?

I would say were fungible fleets, I think, have been something we've been pounding on the table on for the last several years, right? And part of that is location, right, where we have some larger campuses, we're working on in certain locations, and we have some other place -- smaller campuses where location, we think matters. And we think kind of customers want a mix of all of the above. It's not necessarily one or the other.

But more importantly, right, it's the software, right? It's how we build the technology stack to fungibly serve across training and inference such that we don't -- our customer doesn't tell us what they're going to do. We enable them to do training one day and inference the next. And I do think that fungibility really matters. I do think the flexibility for us to be able to take a cluster and turn it into a bunch of small clusters or one massive cluster or to have one customer use it one day and a couple of weeks later, the next customer might be using it for something else. That's pretty critical, right, to the evolution of workloads and AI that will continue to develop in the coming years.

Okay. Let's talk about another aspect of the Street concern, and that's not -- it's not so much on will demand be there. It's more like will financing be there because Microsoft and Meta and Google don't need to lean on debt capital markets or vendor financing or GPU leasebacks to finance it. But CoreWeave does. And by the way, so does Oracle. And so there are several other emerging GPU vendors.

So how would you describe as best you can, the state of the financing demand right now for building out these AI infrastructures. As I mentioned, we're going to have Blue Owl, Magnetar up here this afternoon. Trust me, we'll ask them. But...

Our partners.

What's your perspective? How healthy is it today? Are you seeing any pullback?

So I think taking a step back, CoreWeave has been built on, I think, 2 vectors of excellence, technological and engineering excellence and then excellence at navigating the capital markets and designing our customer contracts in a way such that they are maximally financeable, such that it's no secret, right, like the -- there's been some volatility, right, in the equity market, right, even in the bond market, right, in -- over the past few months.

But I think you got to focus on how we primarily finance our business, right, which is these asset level delayed draw term market or maybe it's 3 years. Are the customers -- are the contracts written the right way. We know how to do that. We pioneered this market. It's can CoreWeave execute, right? I think we, over the last few years, have built a track record of excellence there.

And frankly, our first delayed draw term loan when we were pioneering this market was for investment-grade credit. And people have gotten more comfortable with us and understand this is what we are best at. We've driven down that cost hundreds and hundreds of basis points to the point where over the summer, we financed an unrated customer at SOFR plus 400, where we're talking about 500 to 1,000 -- or 900, I guess, basis points of savings right there.

And what we are continuing to see is the depth and breadth of that market is incredibly robust. If you understand how to, like I said, write those contracts, structure that debt, so it's self-amortizing and also build your backlog in a way such that you are being mindful of things like creditworthiness. We talked about how north of 60% of our revenue backlog at the end of the third quarter was investment grade, right?

From a technology standpoint, what is CoreWeave's enduring advantage when, let's say, any of your large contracts come up for renewal, the world is no longer supply-demand constrained, they pull that into first-party data centers, and you need to sink, or swim based on how good you are executing from a technology standpoint. Anybody can acquire data center space, cable together NVIDIA GPUs in a server -- I'm massively oversimplify. But what is your special sauce, Nick?

So I got to argue for one thing before I answer the question is. I think building supercomputers is a whole lot more complicated than buying some racks, bolting them into the ground and pushing the on button.

I understand. I oversea it...

Yes. And -- by the way, I think we have to continue to kind of explain that to the world. And I do think we're at a point in time in which the world might -- it might be more difficult to differentiate, right?

Yes. That's what [indiscernible] relative to others.

Yes. And I think -- so now to dive into the meat of your question, right, we have purpose-built this cloud from the ground up to deliver maximal performance. That has been an advantage of ours that has allowed us to grow our footprint and acquire customers at a rapid is underselling at rate. We are continuing to innovate right? We are developing new products and services that fit within GPU or AI cloud that are not germane to general compute, but are purpose-built for these types of workloads and what customers will increasingly need in the future. Take things like AI object storage, right? Like that's a product that fits into our broader storage business that we announced in the third quarter has grown to north of $100 million of ARR, growing like a weed.

And that product was built in direct, I would say, response to the advantaged position we're in, which is that we are deeply entrenched with our customers on a technical level. We understand what their pain points are. And when we see something that is missing, we either go out and build it or buy it, right? And so take storage, right? I think we used to live in a world, right, where you were a single cloud customer. You're an AWS customer. I think you are now increasingly an AWS customer and an Azure customer and a CoreWeave customer and storage was not built for that world, right?

You had data lock in, you had high latency, if you want to move it to another cloud, you had high egress fees and transaction fees. So what do we do? We built a product that's low latency with no egress or transaction fees and customer adoption and attach has been very attractive as we've gotten started in bringing that product to market. We are going to keep building those types of products and services such that you look 5 years from now and you say, "Wow, there's a corollary to what the hyperscalers did with the CPU cloud, but they did it specifically for this technology and this world."

Tim, do you want to take a some?

Sure. So there's also this debate, I would say, that's come up lately about alternatives to GPUs. And you build whatever the customer wants you to build. So can you speak to that? Do you have any demand for, number one, AMD GPU? And most importantly, what I'm asking for is, do you have demand for any ASICs such as TPUs? Do you have customers coming to you and saying, "Hey, I want to do development on TPU, so you go out and add any capacity on TPU?"

So you pointed out, we're customer-led in everything we do, right, whether it's entering a new geography or scaling out a different type of accelerator. I would say demand continues for us to be overwhelmingly for NVIDIA technology. I think if there comes a point in time, right, in which we start to hear something different from our customers in a scaled way, not just like a phone call here or there of what do you think of this, then that may change our behavior. But for now, kind of all signals that we get, get to, we need more NVIDIA GPUs, please.

And do you think that there's any -- as the inbound calls to you, do you -- I get the sense that there are some more inbound calls who people want to talk about, well, what if we wanted AMD or we wanted an ASIC.

No, we are...

[indiscernible] calls coming more commonly on that?

Not in any notable way. But I would say like we built this technology stack from the ground up to be fungible across silicon, right? Like we were prepared for a world in which our customers want different things, and we want to give our customers what they want, right? And that embodies what we do. That doesn't mean we don't explore, right, and make sure that we are prepared to work with other types of accelerators. But again, now like the pattern and the trend has been the trend, which is NVIDIA GPUs, please.

Great. I think that's all the time we've got down to 1 second. We squeezed every bit out of it we could. Nick, thank you for coming. Having CoreWeave here, I think, makes this conference phenomenal in terms of the GPU tracks that we've got here.

So I appreciate your attendance.

Thanks for having me, guys.

Welcome, everyone, and welcome to Bank of America's 2025 Leverage Finance Conference. I'm Ana Goshko on the credit side. I'm the research analyst covering technology and telecom. And we're thrilled to have CoreWeave with us today. And we have Paul Yim, the company's Vice President for Credit and Capital Markets, which is perfect for this audience.

So yes, Paul, thank you so much for being with us.

Great to be here today. Actually, before we get started, do you mind if I read the safe harbor?

No, go ahead. Go for it.

And it starts with before we get started. I would like to remind you that CoreWeave may make forward-looking statements during today's fireside chat. Actual results may vary materially from today's statements. Information concerning risks, uncertainties and other factors that could cause these results to differ are included in CoreWeave's SEC filings.

Okay. Great. Okay. So Paul, thank you so much. I know you've been in packed meetings all day long, and I'm sure you've been handling all kinds of very detailed questions. So the purpose of this session is really to take a little bit bigger picture approach. So we can kind of slow the speed down a little bit, talk about the company and its role in AI infrastructure. But since it is a leveraged finance conference, we're going to spend some time at the end to just talk about the debt structure and funding plans.

Worthwhile.

Yes. Okay. So in case there's anyone in the audience new to CoreWeave's story, which I doubt. But just in case, if you could just spend like a minute or two just explaining what you guys do and the role that CoreWeave plays like in the AI ecosystem. I know that's a loaded question but we'll go into more detail.

Yes. So I guess the best way to describe ourselves is we are the AI hyperscaler, right? We are focused on delivering high-performance compute, cloud-ready infrastructure at scale to the world's biggest companies as well as the world's leading AI labs. Our infrastructure is agnostic across inference and training workloads and we can seamlessly shift between both. Our customers consistently rely on us for delivering the most performing cloud in the market.

Okay. And then, obviously, you exclusively work with AI data centers and provide the high compute and the GPU infrastructure. Just as like a primer here, can you explain the key attribute of an AI data center as opposed to a general-purpose cloud provider?

Yes, definitely. I think the first thing that pops out to folks is utilization rates are a little different. The equipment that goes into an AI data center is obviously a lot more heavy duty. Liquid cooling to the rack is now paramount in all of our facilities, and it's just a requirement in order to power some of the high-power density servers versus what a general compute server looks like.

And I think you'll see that we've been very purposeful in designing our data center road map and selecting sites that either are ready for this development or projects that we're building or they can be retrofitted to suit our needs. And I think that's like something that we've been very tactical about and we'll continue to do so.

Okay. So CoreWeave is a very physical business. So there's obviously data centers, power chips, servers, racks. But there's also a software component of the business. And I think on the equity side, you guys are mostly covered by software analysts. So can you just talk about why you're a software company?

Yes. I think definitely from a credit lens, I'd prefer to refer to us as an infrastructure business, but we should definitely talk about the software side which, in many ways, is our secret sauce. I think there's a number of layers to it. CoreWeave Mission Control is our proprietary software orchestration layer and it really is kind of the differentiator between us and any other cloud provider, and this has been validated by third-party consultants like SemiAnalysis on our performance.

But basically, what it does is it provides active health management of the clusters and it's a tool that allows us to autonomously manage our AI cloud. And beyond active management of a cluster, it's observability on the site itself to ensure that we can deliver maximum performance against the longevity of the chips, which is equally important to us when we think about the long-term and intermediate term of our business.

Okay. So there is, I think, a well-known bull-bear debate on CoreWeave. So I think we're going to do a little bit of the bull-bear here. So first, I'm going to give you the opportunity to just list the key pillars of the bull investment thesis, and then I'm going to stand up some of the other side and kind of let you respond to those.

Yes. I think that's fair. How I would start is I think the bull case is actually fairly straightforward. You have to believe on some level what our customers already seemingly believe, which is that this infrastructure is, in fact, mission-critical. That to power an AI future, it requires these complex workloads that requires a fully developed tech stack in order to service this.

This is something we saw well in advance of ChatGPT, right? We've been doing this years, years, years before. If you look back to, I think, NVIDIA's website back in 2019, well before they were worth their first $1 trillion, they named us as a Tier 1 CSP for that reason. I think this is something we were going to actively work towards and maintain our industry-leading position.

Okay. So some of the key pillars, I'd say, of the sort of investor concerns, right? So one is really the GPU useful life and/or the residual value. So the risk that the GPUs cannot be re-leased or at an adequate rate at the end of their initial contracts either due to technological obsolescence, or even if the rate is too low, the diminishing kind of investor return to CoreWeave. So what can you say in response to those kinds of questions?

Yes. I think it's a fair question. The way I would first point folks to is, number one, we are customer-led, right? When we announce our CapEx, it's usually in conjunction with actually signing a customer. We only spend CapEx on a success base. It's not speculative when we spend capital expenditures. It's definitely earmarked against our contracts.

And similarly, I'm sure we're going to touch on this later, on the SPV debt, that's how we design funding plans for them, right? Our contracts must ensure repayment of not only the infrastructure itself but also the debt as well. So for our purposes of -- and certainly, our investors in the audience on the bondholder side, there's no residual value risk that they're actually undertaking when we think about it because we're already scraping levered free cash flow after amortization, after interest, during even the first contract period.

Renewal in many ways is a bet that we are taking as a company on the upside of what remains. And you've already heard this to be untrue actually on the re-leasing from Mike during Q3, right? We announced a big renewal for a 10,000 GPU cluster. It's not a one-off. For us, that's the only example we pointed out because it was a big material cluster. And in that particular case, we saw a customer renew with us at within 5% of the original sale price. I think that's something you're going to consistently keep seeing.

And part of that is a function of how high demand is for an infringed product. Typically, when we're at the renewal point for a cluster, our customers are probably looking at what it means to keep inference workloads live there. Now remember, inference for our customers is a revenue-generating workload, right? So for them, when they've already budgeted for what the expense is to continue to pay for CoreWeave and maintain access to these GPUs, it's just about making sure that they're still generating a profit on those sites, which is why we feel very strongly about our renewal curves.

But I think most importantly, our lenders don't take a risk on renewal because that's not what we -- I think this is something that's inherently different from other businesses that do leverage up in order to acquire their infrastructure.

Okay. And then you already touched on one of my next questions. So you cited which Mike had cited on the 3Q call, that there was a proactive early renewal of a 10,000-plus GPU H100 contract, which is one of the earlier generations. And that was done at a 5% discount to the original rate, right?

Yes.

And you're saying that wasn't just kind of anecdotally cherry picked, that, that is...

It's not a one-off. We are consistently close to, if not already sold out of any of our older-generation architecture. I think what I didn't touch on there is enormous enterprise demand behind all the IG guys or investment-grade guys as well as our big labs that want to keep active use of those clusters coming up from right behind.

And part of the reason for that is a H100s have been out longer than the GB200s or GB300s. And therefore, the software libraries that support them are bigger and wider and the engineers at these other companies have more familiarity with them, which is why it's driving a lot of demand towards these older generation GPUs despite the fact that they may be 3 or 4 years out in active use.

Okay, okay. So then I think the second pillar of sort of investor concerns is the potential for oversupply. So one of the issues, I think you already touched on. But does HPC demand wane when clients move from AI training to the inference stage?

And then the second part of that is really right now, it is a supply-constrained environment right now for capacity. But will the need for CoreWeave's capacity be diminished when hyperscales build up their own vertically integrated facilities? And by the way, we're kind of moving more towards the inference stage.

Yes. So let me start by addressing the inference point. And absolutely not. The short answer is absolutely not. Inference requires way more compute than training does. And the reason why is all of a sudden, it's not in a black box being used by the AI scientists that are running it through 100 billion parameters. Instead it's being demoed by the user base of said AI company and maybe a hyperscaler, where the inference side of the house, it's everybody that's accessing ChatGPT or whatever it may be. And as a result, it actually requires more GPUs.

Now it doesn't necessarily require as performing GPUs. What it requires is just more. And I think that's a big fundamental driver to, number one, our renewal pricings and why we feel so comfortable with what we're saying out here and what we're observing in our demand funnel, right? Just inference is requiring more GPUs full stop than versus training. And we haven't even scratched the surface of what is required to develop training at scale. That's what gives us so much conviction that, on a long-term basis, do we think supply chain will ease? Absolutely. Do we think it's near? Probably not.

And the one thing I would add on the second point of your question on this supply-constrained market, that this is something we've observed in the data center ecosystem for more than 15 years now. Every year we say, in 2 years, data center capacity is going to open up. We haven't quite hit a point where that's actually been true. And I think when that happens, does that require some recalibration of how costs and unit economics work? Absolutely. But I don't think we're close to it yet and we feel very comfortable with kind of our forward forecast.

Okay. And then on this idea that some of your customers may over time build up their own vertically integrated capacity, how fungible is the infrastructure? Can you repurpose your existing infrastructure for other customers?

Yes. So that's actually one of our secret sauces and it goes to the software orchestration layer. We are able to quickly rapidly repurpose clusters. It really just depends on how big the sites are. That introduces some level of complexity. But in terms of like speed and pace, we can absolutely do it. It's usually less than a month. But it ranges, right? Like I don't want to put that out there as like the only marker for it. So that's number one.

Number two, I think there's been a view long held in the data center world that eventually hyperscalers are going to start building their own and stop relying on the big public cloud data center builders. And some of them are private, some of them are public. We all know who the names are. And I think that's just something that hasn't come to pass. And part of it is incumbent upon that thesis or this bear thesis that hyperscalers are going to move towards first-party development for their own data centers.

And this is going to extend to my analogy on GPUs, is you have to believe Internet infrastructure has been fully developed and that we're done building more data centers, and compute is being serviced at a reasonable and acceptable scale to the world. And I just don't think we're close to it on the Internet side. I can promise you we're not close on the GPU side.

Okay. Any other arguments from naysayers that I haven't stood up yet that you'd like to address?

Yes. The only one I was going to point out is, and I think we're going to spend some time on this so maybe I'm jumping ahead too much, do cut me off, is I think there's been -- for us specifically, we've observed some criticism on our use of the capital markets, which I think is mostly linked with the misunderstanding of how we actually maximize our efficiency through the use of capital markets, right?

It's in many ways what allows us to equalize how we're able to compete with big hyperscalers that also do what we do, right, build high-performance compute clusters and service customers at scale, whether it's AI labs or enterprise customers alike. And I think it's been a great tool in our menu of options to service our needs. I'm not going to spend -- I know you're going to get to it later.

But I think the most important thing I'd call out here is in order to support our -- I think, as of Q3 end, we said $55 billion of backlog, even if you pro forma for all of the debt that is required to stand up that $55 billion of backlog, we will have sufficient cash flow streams to not only cover the SPV debt we would raise to fund our growth but also to repay the bonds that are outstanding on our balance sheet.

Okay. So shifting to the data center side of stuff. So with regard to powered shell capacity, is there a preference for leasing versus owning/self-build? And a couple of related questions. How do funding costs play into this? And then, two, after the Core Scientific deal cancellation, how do you think about data center acquisition potential?

Yes. So let me tackle this a couple of different ways. And I'll start by saying we are building one of our data centers right now in Kenilworth, New Jersey, which is 20 miles outside of Manhattan. It's going to be one of the biggest high-performance compute enabled sites within an acceptable zone in Manhattan. It's just going to make it very important. We are doing that. We have a joint venture partnership with Blue Owl and we're excited about it.

But at the same time, you'll see that we are primarily leasing most of our portfolio. A function of that is the timeline for data center ownership model and build-out is a little bit incongruent with our current unit economics, right? A dollar spent on an AI server is worth more to us than a dollar spent on a data center site.

Now as we start charting a pathway to investment grade and our credit improves, our cost of capital improves, are we going to reevaluate this? Absolutely. I think there are probably a set of trophy assets we care deeply about that we would like to build and own. But it's not something that impacts us today in a way that is forcing us to go build ourselves. So I think that's where I would start.

To answer your last question on how we think about data center acquisitions, and I don't know if this going to come in next, we think about acquisitions in two ways. One is strategic, one is opportunistic. Core Scientific is absolutely an opportunistic acquisition opportunity. It was us showing that we're willing to equitize some portion of our lease obligations, right? Like we were prepared to do that.

And we thought the price we put out was fair. The shareholders felt a little differently. And I don't think that changes anything. Ultimately, we've locked up the power that we want from their portfolio, and we still feel very strong about our partnership going forward.

Okay. So shifting to some more kind of pure financial type topic. So there is a ton of demand right now for what you guys offer. But you recently brought on a Chief Revenue Officer. So why?

Yes. It's a great point, fair enough. And we're very excited about having Jon Jones join us. Great name. He came from Amazon and he was a responsible for product for Amazon. And I think one thing that a little bit of a shame for our business is we're seeing just enormous growth, as you're highlighting, from the hyperscalers and it really just dwarfs what we see on the enterprise side.

And as we're growing up and maturing rapidly as a public company, right, we have historically serviced a fairly short list of big customers that represent the bulk of our customer portfolio. Now we've dramatically improved our diversification already from the start of the year. I think when we IPO-ed, we were showing somewhere around 80%, 82% first customer -- our biggest customer exposure. Today, it's less than 35%, right? So we've already dramatically improved that. Our IG exposure is also greater than 60% already.

However, there's not a big enough sales organization tackling the next layer of customers that exist out there. And it's enterprises like CrowdStrike, which we just announced a big partnership with, I think, a week ago or maybe 2 weeks ago. We announced one with IBM at the start of the year. And we expect to do a lot more of that, and that's going to serve as a launch pad for when we think about infrastructure in the outer years, a lot of these customers aren't necessarily needing to use the latest and greatest CPUs. And that helps support our business on a long-term basis. That's fundamentally one of the big reasons why we brought Jon in, and we're super excited.

Okay. Next topic is, speaking of big customers, you have a $6.3 billion backstop from NVIDIA. So could you explain how that works? And then I think it might be useful to recap NVIDIA's position as both a vendor, customer and then the third largest shareholder of the company.

Yes. So I think what you're going to see as we grow our business and build it going forward is we're going to have to show a little bit more of a bias towards investment-grade offtake. And a big reason for that is we are on a journey to hit investment grade. It's a little bit of a North Star for us. I know we're going to touch on it later so I'm not going to skip ahead. But part of that journey is we need to sign big investment-grade contracts in order to show a pathway of achieving that outcome.

Now that contract we signed with NVIDIA is an interruptible contract. And it basically allows us to use their credit to acquire and build an SPV to raise debt against their credit to buy GPUs, where we're then able to sell to another customer, AI lab or what have you, that perhaps isn't able to sign a 5-year contract today on the latest and greatest chips. That contract we announced was for GB300s.

Today, if you're coming to us for GB300s, you basically have to lock in for a 4- or 5-year contract. There really isn't an example of where we're going to build a big cluster of GB300s and sell it on a shorter than 3-year basis, right? But if you're an AI lab that hasn't raised enough money yet, you're not going to be able to commit to a contract greater than 6 months, 1 year or 2 years. But it's very important to us that we give them an opportunity to maximize the use case of highly performing infrastructure, number one; and number two, get them an opportunity to put a product out there that hopefully gets them a chance to raise capital and put them on the map in some way.

And I think that's something that NVIDIA saw, and this partnership is to further that journey, if you will.

Okay. So I'm going to touch on now sort of the guidance change that happened around the third quarter earnings report. So you had a strong third quarter but then you ended up lowering the revenue guidance for 2025, and so $100 million to $200 million lower but on a base of $5 billion.

Yes.

Okay. And that was all attributed, I believe, to a data center vendor construction delay, which I think now is -- you explained it slipping from fourth quarter to the first quarter. Is that fair?

Yes. That is probably fair, yes.

Okay, okay. And then secondly, in terms of the operating income guidance, that was also reduced by $110 million. So that's now $690 million to $720 million. So that was due...

Also to the same reason.

For the same reason, okay. But you're also bringing on -- despite that delay, you're bringing on 260 megawatts of active power still in the fourth quarter, which is like a big step up, right? And there is some element of having that come online and the cost of that coming online before the revenue fully ramps up. Is that fair?

Yes. That's fair. I think it's more of a quirk to the business model than anything else. Like oftentimes, we're paying for data center expenses before we bring the clusters online, right? Like there's a little bit of a time lag, it's not long, between data center readiness and customer readiness. And that's a little bit of the quirk that you're just observing.

Okay. And then any update on that data center construction delay? Do you feel confident that, that's going to come online now in the first quarter?

We feel comfortable with the guidance we provided in Q3, I think you heard from Nitin, and I think that's still consistent today.

Got it. Okay. Let me skip over a couple of things in the interest of time here. But so CapEx, so CapEx 2025 was reduced quite a bit because of that.

All related to the same thing.

All the same thing. So it's $8 billion to $9 billion now. It was $12 billion to $14 billion previously. Though I think from a cash flow perspective, a lot of that is still being spent because it's work in progress. So it's really just sort of a recognition of kind of what hits CapEx, right, on the accounting statement.

Exactly, yes. It's basically a timing of acceptance and delivery of the site creating this little time mishap, if you will.

Right. Okay. So now switching to the debt structure. So there's an evolution of the debt structure, right? And you have alluded to you have a contract-first financing approach. Can you talk about that?

Yes. I think it's been very much a guiding principle of ours, which is we're only going to spend CapEx if we sign the customer. If we sign the customer, we need to be able to raise financing to support it. And the historical way we've gone about this is we reach out to private credit. We've got great relationships with Blackstone, BlackRock, Magnetar and a number of other folks in there that help structure what looks like an asset-backed financing package that still receives a parent guarantee from CoreWeave, Inc., which is why it still shows up on our balance sheet.

And when we first did this in 2023, I think the understanding and acceptance of what this looked like was still developing or not a lot of folks are comfortable with what GPU cloud look like or GPU hyperscalers look like. Going forward, I think one of our future ambitions is basically raising these especially for our investment-grade offtake customers on a nonrecourse basis without a parent guarantee.

And there's a number of benefits to that. Number one, it's getting the look-through credit to our customer which, in many cases, is very highly rated, AA, AA- or AAA. And to the extent that we're able to achieve that outcome, we'll be able to capitalize on their credit profile, similar to how the data center ecosystem already structures project financing, right? When a big data center operator is building a site, they're usually structuring the exact same vehicle. It's a product finance box that the look-through is to the customer and they borrow against that rate.

I think we saw Meta do this recently with the Hyperion data center. I saw a couple of other folks that do this consistently in the market. Our goal is to do this for ourselves first on the GPU side. And our expectation is to be able to scale this up, especially as we build our investment-grade backlog.

Okay. And so I've heard management talk about that your balance sheet is actually a diverse set of balance sheets. So that's really what they're referring to, right?

Yes. Like we have various SPVs that support different sets of contracts and they have different sets of terms and credit spreads on them. But at the end of the day, they all flow back up to the parent, right? I think what's important to note is all of our debt is self-amortizing at the SPV level, right? Or lenders down there don't take renewal value risk.

And our parent investors or investors in our equity story or our bondholders, they get to enjoy the benefit of the levered dividend or free cash flow yield out of the SPVs back to the parent. And right now, we have an umbrella of 3 or 4 SPVs below us. We expect to continue to do them. We expect to outperform versus relative to prior deals. And hopefully, that's where we're able to rationalize some of the cost of capital advantages that we have.

Okay. So right now, the run rate EBITDA is about $3.4 billion. Your debt is $14 billion, and you got $3 billion of cash, right? So your leverage right now, 4.3, 3.4 net. Your CapEx for next year is guided to be 2x what it was in '25. That's $24 billion to $28 billion. So some of that, as we talked about, is actually really being spent this year. It's just getting recognized as CapEx next year. So nonetheless, it's a big number.

So could you talk about what are the sources? I think you've really addressed. But if you kind of pull it all together, what are the sources for funding your CapEx needs? You do have cash from operations and then you have upfront payments from some customers that you're going to put in this committed financing. Is there going to be a need potentially for more unsecured financing in your...

Not like explicitly stating it, we don't need to access the bond market, to be clear. We can fully support our $55 billion backlog. Obviously, some of that is already spoken for by the existing SPVs we've raised. But our expectation is to be able to raise more SPVs that either have more efficiency or allow us to tap into different financial instruments that once again isolate to look at the contracts as opposed to the parent itself. And that is going to enable us to look at it a bunch of different ways.

And I think what's most important to recognize is like, to be clear, our first DDTL 1 and 2 that are outstanding didn't require equity injections because of what you highlighted, which is some of our customers still contribute upfront prepayments, and they use it as a means to toggle contract value on a dollar per GPU rate. And it's a great instrument for us to maximize our credit in the outlook so we don't have to lean as heavily on parent financing means.

Does that mean we're not going to look at the high-yield bond market? Probably not at the current levels. But it absolutely does mean like there's going to be a time in place. We're probably going to try to build our portfolio in a way that's more thoughtful, that's more creditworthy and allows us to access a higher throughput capacity of capital markets.

Okay. And then so I did cite -- so right now, the gross leverage is mid-4s, right?

Yes.

Does that go up before it comes down? And I'd like you to talk about your IG North Star, like how you're going to get there.

Yes. I think this is a great point to spend some time on. We actually like to think of our debt, and I know a lot of folks in the audience know this too, we like to think of our debt on a pro forma basis. It's more fair to -- if we're going to get credit for the backlog and what the EBITDA would look like on a ramp basis, we should be burdened for the debt as well. And I think we spent some time on this back in May with a lot of folks in the room on highlighting how new contracts we sign are ultimately just deleveraging events upon stabilization.

Stabilization is basically when the cash flows are ramped, meaning sites delivered, clusters are built, customers accepted. Upon stabilization, most of our clusters are, on a ramp basis, individual SPV by SPV, represent roughly between 2 to 3.5x net leverage. So when you account for where our current leverage is, it's absolutely going to be delevering events as we announce new contracts. Now the only thing to that is, to your point, there is a little bit of a lag period in between build to rent.

Now what's important here is, I gave away a little bit of our secret sauce, that's going to come up next, which is doing nonrecourse facilities. When we do nonrecourse deals, it is functionally remote from the parent, right? Like if anything were to happen to the box itself or the SPV, there is no impact to the parent itself. And the parent does not need to cure the SPVs, which is the objective of our future financing packages for some of these big contracts. I think that's going to be a big driver in how we message and communicate to the rating agencies on our pathway to hitting investment grade.

I don't know the exact time frame, and we're obviously an interesting company just given our growth profile and how much we spend. But so long as that expenditure is being locked away in a vehicle that's away from the parent and what comes out of that SPV is an investment-grade stream of dividends, I think we're going to be able to chart a fairly good narrative on how do you understand the business evolution from where we are today to what we would like to look like in 3 years.

Okay. Great. So almost out of time. I skipped over a couple of topics, but we appreciate all the time you've been spending with investors. So hopefully, everyone had a chance to ask additional questions of you. And you were super efficient in this session, so I really appreciate it because we hit on a lot.

With the minute that we've got left, is there anything that we didn't touch on that you think is important? Or do you want to -- any closing comments about what you guys are most excited about?

We're super excited about the demand profile that we're staring at for the next year. I think there's a lot of talks about what's going on out there. I think you can kind of read between the lines when you're just looking at the earnings reports for a lot of our biggest customers, right? Like they're not observing a slowdown because the demand does exist out there. And I think so long as that does translate, we're going to be the beneficiaries of a lot of how this unfolds on a long-term basis.

Remember, we're infrastructure builders. We are a software company that builds infrastructure. And while we are hopeful that each and every one of our customers is individually successful, in many ways, it doesn't actually impact us on which one does get there. The reality is this current supply-demand market necessitates that each of our customer locks into longer and longer contracts with us. I think that's the most resounding supporting information or rebuttal, if you will, against what people are talking about the useful lives of the GPUs, in addition to what we're talking to you about with respect to renewals.

Okay. Great. Okay. With that, we're just out of time. So it's been perfect. Paul, thank you so much for being with us.

Thank you, Ana.

Thank you for standing by. My name is Tina, and I will be your conference operator today. At this time, I would like to welcome everyone to the CoreWeave Third Quarter 2025 Earnings Call. [Operator Instructions]

It is now my pleasure to turn the call over to CoreWeave.

Thank you. Good afternoon, and welcome to CoreWeave's Third Quarter 2025 Earnings Conference Call. Joining me today to discuss our results are Mike Intrator, CEO; and Nitin Agrawal, CFO.

Before we get started, I would like to take this opportunity to remind you that our remarks today will include forward-looking statements. Actual results may differ materially from those contemplated by these forward-looking statements. Factors that could cause these results to differ materially are set forth in today's earnings press release and in our quarterly report on Form 10-Q to be filed with the SEC. Any forward-looking statements that we make on this call are based on assumptions as of today, and we undertake no obligation to update these statements as a result of new information or future events.

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 an accompanying investor presentation are available on our website at investors.coreweave.com. A replay of this call will also be available on our Investor Relations website.

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

Good afternoon, everyone, and thank you for joining us. CoreWeave once again delivered an exceptional quarter, showcasing the accelerating momentum underlying our business as AI adoption proliferates globally across industries. We continue to operate in a highly supply-constrained environment, where the demand for CoreWeave best-in-class AI cloud platform far exceeds available capacity. This insatiable customer demand is a clear signal that the world's leading companies trust CoreWeave to power their most critical AI workloads.

In Q3, we beat expectations, delivering revenue of $1.4 billion, up 134% year-over-year. We added over $25 billion in revenue backlog in the third quarter alone, bringing us to over $55 billion in revenue backlog to end Q3, almost double Q2 and approaching 4x year-to-date. Further, CoreWeave has reached $50 billion in RPO faster than any cloud in history. These results demonstrate the deep confidence customers have in CoreWeave, the company they trust as their essential cloud for artificial intelligence.

We continue to scale aggressively even as the industry remains capacity constrained. We expanded our active power footprint by 120 megawatts sequentially to approximately 590 megawatts, while growing our contracted power capacity over 600 megawatts to 2.9 gigawatts. This leaves us well positioned for future growth with more than 1 gigawatt of contracted capacity available to be sold to customers that we expect to largely come online within the next 12 to 24 months.

In Q3, we executed large-scale compute contracts with many of our largest customers, including Meta and OpenAI. Each represents a meaningful expansion of existing relationships and a diversification away from any single customer. We also grew our relationship with a leading hyperscaler, marking the sixth contract with this customer to date. In fact, 9 of our 10 largest customers have now executed multiple agreements with us. The only exception being a new customer we onboarded in Q3.

CoreWeave is the force multiplier that empowers pioneers to accelerate breakthroughs in AI innovation. These are the world's most sophisticated AI organizations and once they experience the performance, flexibility and reliability of CoreWeave Cloud, they consistently expand with us. That is the strongest validation we could ask for. Our exceptional growth illustrates just how quickly AI adoption is progressing beyond the frontier AI labs and hyperscalers. Broader global demand and our recent large wins are driving diversification of our revenue base. For example, the number of customers that exceeded $100 million of revenue over the last 12 months tripled year-over-year. AI native and enterprises across sectors are embracing CoreWeave to transform operations and unlock new sources of innovation, productivity and growth.

At the forefront of foundation model development, Poolside selected CoreWeave to power its mission to build artificial general intelligence and enable the deployment of agents across enterprises, while Periodic Labs is using CoreWeave to push boundaries of scientific discovery and computational research. At the application layer, we added AI-native customers like Jasper, who chose CoreWeave as their cloud partner as they transform the digital marketing landscape.

We are also seeing incredible momentum within enterprises. CrowdStrike chose CoreWeave to advance the development of AI agents for cybersecurity, while Rakuten is using our platform to transform their visual language models, helping to achieve greater transparency, reproducibility and speed in their AI workloads. We also saw further expansion with a wide range of enterprise customers, including a leading software design platform and a large telco operator in the U.S.

Our reach now extends into the public sector, a market with unique performance and security requirements. We recently launched CoreWeave Federal to bring our cloud services to the U.S. government agencies and the defense industrial base. Already, NASA is leveraging our services to advance scientific exploration at its jet propulsion lab. We are honored to help strengthen America's AI infrastructure, enabling agencies to accelerate innovation and address critical missions and our national interests. These recent wins underscore that we are enterprise ready.

With our customer base broadening across verticals and geographies, we are excited to welcome John Jones as our first Chief Revenue Officer. John joins us from AWS, where he served as Global Head of Startups and Venture Capital. John is a strong addition to our team and will play an important role scaling our global revenue organization and driving expansion through this next phase of growth.

Next, as I move to discuss our growing data center footprint, I want to briefly touch on our previously proposed acquisition of Core Scientific, which was terminated in October. While the deal made sense strategically for both companies, the valuation required by their shareholders was simply not a price that was appropriate for CoreWeave, particularly because the outcome of the transaction in no way adversely impacts our ability to achieve our growth ambitions in the coming years. Instead, we will continue to work closely with Core Scientific on the approximately 590 megawatts of capacity we have already leased.

Our disciplined approach to expanding our capacity footprint ensures we are meeting the surging global demand for CoreWeave's cloud services. As I mentioned, we grew our contracted power capacity to 2.9 gigawatts this quarter as we diversified across size, geography and developers, enhancing resilience and flexibility across our portfolio. As of Q3, no single data center provider represents more than approximately 20% of our contracted power portfolio.

In the past quarter, we added 8 new data centers across the U.S., strengthening our domestic coverage with additional expansions underway across Europe, including a major new presence in Scotland, which is being developed in partnership with the U.K. government. And as we announced over the course of the summer, we have embarked on self-build projects to further accelerate our footprint and provide us greater operational control.

While we are experiencing relentless demand for our platform, data center developers across the industry are also enduring unprecedented pressure across supply chains. In our case, we are affected by temporary delays related to a third-party data center developer who is behind schedule. This impacts fourth quarter expectations, which Nitin will discuss shortly.

Having said that, the customer affected by the current delays has agreed to adjust the delivery schedule and extend the expiration date. As a result, we maintain the total value of the original contract and the customer preserves their capacity for the full duration of the initial agreement, demonstrating the confidence they have in our ability to provide the most performance solutions in market.

We are incredibly proud of our technical accomplishments and our customers continue to tell us that CoreWeave is the absolute best place to run AI workloads. In the third quarter, we continued to deliver many of the initial scale deployments of the GB200s, while once again being first to market this time with the GB300s, further highlighting our incredible track record of operational excellence.

CoreWeave's industry leadership is unmatched. We are the only cloud provider to submit MLPerf inference results for GB300, setting the benchmark for real-world AI performance. And just last week, SemiAnalysis once again recognized our dominance, awarding CoreWeave its highest possible distinction, it's Platinum Cluster Max ranking for the second time, ahead of more than 200 providers, including the hyperscalers and emerging neo clouds. No other cloud has achieved this once. CoreWeave has done it twice, underscoring yet again that CoreWeave stands alone at the forefront of the AI cloud.

Demand for AI cloud technology remains robust across generations of GPUs. For example, in Q3, we saw our first 10,000-plus H-100 contract approaching exploration. Two quarters in advance, the customer proactively recontracted for the infrastructure at a price within 5% of the original agreement. This is a powerful indicator of customer satisfaction as well as the long-term utility and differentiated value of the GPUs run on CoreWeave's platform.

CoreWeave is the world's first AI cloud at hyperscale, comprising compute, storage, networking and software purpose built for AI workloads. Our growing cloud portfolio is underpinned by an expanding suite of software and services that help our customers build, train and deploy new products faster.

In addition, to mission control our proprietary orchestration solution, which is critical to autonomously operate our AI cloud at the bleeding edge, we recently launched CoreWeave AI object storage, a fully managed storage service that eliminates any friction of moving data between regions, clouds and tiers with zero egress or transaction fees. CoreWeave's AI object storage delivers the highest amount of throughput of AI workloads while cutting the customers' cost by more than 75%. We have already seen tremendous interest in this offering, adding a number of initial customers, including Frontier AI labs like Mistral.

Across our entire storage platform, we have seen rapid customer adoption, eclipsing $100 million in ARR in Q3. Combined with our unique global network backbone, purpose built for AI, this positions CoreWeave as the hub for customers and their key AI workloads, enabling consistent best-in-class performance and seamless user experiences when utilizing CoreWeave cloud or a secondary provider. We supplemented these capabilities with further expansion of our observability and security suites to ensure that CoreWeave is best positioned to handle all of our customers' critical workloads regardless of the use case or geography.

Our role over the last few years has been to support the pioneers who are developing and improving AI. Now we are expanding our role to help put AI to work, from the tools that developers require to build AI to the solutions that the physical world requires to adopt AI. We've used M&A as a key tool to accelerate this journey including the recently announced acquisitions of OpenPipe, Marimo and Monolith. With Open Pipe, we quickly integrated their solutions into our broader fine-tuning product suite and introduced the first publicly available serverless reinforcement learning tool. With Marimo, we are expanding CoreWeave's exposure to and impact within the open source community, starting with entry-level exploration and prototyping. Both OpenPipe and Marimo fit seamlessly with the capabilities of weights and biases, where we are rapidly growing the developer base reliant on CoreWeave's holistic platform.

With Monolith, we are expanding these capabilities into the physical world to unlock the monetization of AI today. initially focusing on industrial use cases with an established enterprise customer base and mature workloads, including leading auto OEMs like Nissan and Stellantis. Through the rapid and successful launch of new products and services, we are expanding our addressable market and growing with our customers. We are fundamentally evolving the capabilities of CoreWeave, which is creating beachheads and expansion opportunities into new markets, all in the service of further supporting the rapid growth of AI and enabling AI builders and innovators to get to market faster and more reliably and drive ROI.

Our engagements are getting more sophisticated as evidenced by our partnership with CrowdStrike, which will unlock and accelerate partner-driven growth. Our new storage product and partnership with VAST Data is another example of accelerating both our product portfolio and partner go-to-market motions and allows us to compete in new markets where we previously had limited or no offerings. This facilitates customer-driven platform adoption and product-led growth, creating tailwinds for our business.

As I close, I want to emphasize what truly sets CoreWeave apart. We are the essential cloud for AI, combining unmatched technical and operational excellence with a rapidly diversifying customer base. We deliver the most performing infrastructure, the fastest time to market and the most advanced capabilities in the industry. The world's leading AI innovators choose CoreWeave because we enable them to move faster, scale smarter and achieve outcomes that simply are not possible anywhere else. Our momentum has never been stronger, and the opportunities ahead continue to expand, powered by exceptional products, an extraordinary team and unrivaled execution CoreWeave is ready to enter the next phase of growth as a full-stack AI service provider and hyperscale. The future runs on CoreWeave, and we are just getting started.

With that, here's Nitin.

Thanks, Mike, and good afternoon, everyone. Our impressive third quarter results reinforce the relentless demand for CoreWeave and our focused execution in building the essential cloud for AI. As Mike shared, we continue to execute within a highly supply-constrained environment, which we expect to persist for an extended period of time. Our continued focus on delivering the most performance solution in the market and investing up and down the stack is spurring growth and diversification across our customer base from new enterprises and AI natives to expansion with existing customers.

Now turning to Q3 results. Q3 revenue was $1.4 billion, up 134% year-over-year, driven by robust customer demand and strong execution. Revenue backlog for the quarter ended at $55.6 billion, almost doubling in the third quarter alone. Demand remains robust for not just the Blackwell platform but across our GPU portfolio. In the third quarter, we signed a number of deals for older generations of GPUs, adding new customers and recontracting existing capacity.

The breadth of demand for CoreWeave's cloud services has enabled us to reduce our customer concentration significantly. Today, no single customer represents more than approximately 35% of our revenue backlog, down from approximately 50% last quarter and even more meaningfully from approximately 85% to begin the year. Additionally, as of Q3, more than 60% of our revenue backlog is tied to investment-grade customers. This is what successful execution against our stated goal of platform and customer diversification looks like.

Operating expenses in the third quarter were $1.3 billion, including stock-based compensation expense of $144 million. We continue to ramp our investments in data center and server infrastructure to execute against our growing revenue backlog, which contributed to the increase in our cost of revenue and technology and infrastructure spend in Q3. In addition, the increase in sales and marketing was driven by investments in marketing and scaling our go-to-market organization to capture the rapid growth of AI opportunities across enterprises and AI natives. The increase in G&A was driven by professional services and head count.

Adjusted operating income for Q3 was $217 million compared to $125 million in Q3 of 2024. Our Q3 adjusted operating margin was 16%. Adjusted operating income was better than expected due to higher revenue, lower costs due to timing of data center deliveries from our third-party partners and improved fleet efficiencies. Net loss for the third quarter was $110 million, compared to $360 million net loss in Q3 of 2024.

Interest expense for Q3 was $311 million compared to $104 million in Q3 of 2024 due to increased debt to support the scaling of our infrastructure, partly offset by the benefit from better interest rates on our debt as we make further progress in lowering our cost of capital. Adjusted net loss for Q3 was $41 million compared to approximately breakeven in Q3 of 2024. While adjusted EBITDA for Q3 was $838 million compared to $379 million in Q3 of 2024, increasing more than 2x year-over-year. Our adjusted EBITDA margin was 61%.

Turning to capital expenditures. CapEx in Q3 totaled $1.9 billion, lower than anticipated due to the delays Mike mentioned related to deliveries from a third-party data center provider. The meaningful growth in construction in progress to $6.9 billion, an increase of $2.8 billion quarter-over-quarter is a direct result. As a reminder, construction in progress represents infrastructure, not yet in service and is excluded from CapEx until it is deployed.

Now let's turn to our balance sheet and strong liquidity position. As of September 30, we had $3 billion in cash, cash equivalents, restricted cash and marketable securities, growing rapidly and operating at scale demands a strategic approach to securing capital. CoreWeave has established itself as the leading AI cloud and the leading innovator in financing the infrastructure required to power the world's most advanced workloads for enterprises and AI labs.

We continue to make great progress in strengthening our capital structure and lowering our cost of capital. In Q3, we amended the DDTL 2.0 facility by increasing its remaining drawable capacity by over $400 million to create a new $3 billion tranche at SOFR plus 425, which is significantly below the original cost of the facility. As we discussed previously, we also closed DDTL 3.0 in the third quarter priced at SOFR plus 400, which represents a 900 basis point decrease from the noninvestment-grade portion of our prior facility. Going forward, we expect to continue to be able to finance at lower spreads as our capital providers increasingly appreciate our best-in-class execution as well as durable cash flow and visibility that underpin our take-or-pay customer contracts.

Further, we raised $1.75 billion in senior notes in July, extending our exposure to the high-yield market at a cost 25 basis points lower than our inaugural offering in May. Year-to-date, CoreWeave has successfully secured $14 billion in debt and equity transactions to support our execution on our rapidly growing backlog and efficient scaling for long-term growth.

Other than payments related to OEM vendor financing and self-amortizing debt through committed contract payments, we have no debt maturities until 2028.

Turning to tax. In Q3, we recorded a noncash tax benefit, primarily due to the impact of One Big Beautiful Bill. While the size of the impact to Q3 was onetime in nature due to a year-to-date catch-up, we expect the change in law to enable CoreWeave to realize cash tax savings in future periods.

Now turning to guidance. As mentioned, the delays in power shell delivery associated with the data center provider will have an impact on our fourth quarter results. These delays are temporary, and as Mike noted, the affected customer has agreed to adjust the delivery schedule to preserve their capacity for the full duration and the total value of the original agreement. With that backdrop, we now expect 2025 revenue in the range of $5.05 billion to $5.15 billion. In addition, we anticipate 2025 adjusted operating income between $690 million to $720 million and expect to end the year with over 850 megawatts of active power.

In Q4, we will be bringing online some of the largest scale deployment in our company's history. This will have a near-term impact on adjusted operating margin due to the timing difference between when data center costs are first incurred and when we start recognizing revenue. We expect 2025 interest expense in the range of $1.21 billion to $1.25 billion, driven by increased debt to support our demand-led CapEx growth, partly offset by an increasingly lower cost of capital.

Moving to CapEx. We now expect 2025 CapEx in the range of $12 billion to $14 billion. We expect this reduction in CapEx from our prior guidance will be mostly reflected by a corresponding increase in construction in progress due to the buildup of infrastructure waiting to be deployed following the delivery of powered shell capacity. As such, the vast majority of the remaining CapEx we had previously anticipated to land in Q4 will now be recognized in Q1.

In addition, given the significant growth in our backlog and continued insatiable demand for our cloud services, we expect CapEx in 2026 to be well in excess of double that of 2025. These investments in our infrastructure platform will strengthen our competitive moats and support our continued hyper growth.

In closing, we delivered a record third quarter and remain more confident than ever in the long-term trajectory of our business. Over the course of this year, we've made tremendous progress. Accelerating our revenue backlog growth that now exceeds $55 billion while diversifying our customer base, executing strategic partnerships and acquisitions to strengthen and broaden our platform. accessing new capital pools that meaningfully reduce our cost of capital and scaling both our capacity and organization at an unprecedented pace. This progress enables us to seize the opportunities in front of us today and create a strong foundation for years to come.

Our addressable market continues to expand, not only as AI adoption proliferates across industries and use cases, but also through deliberate business decisions we've made to broaden our product portfolio and capture greater wallet share across the industry. CoreWeave is reaching escape velocity, scaling more rapidly and efficiently and solidifying our leadership as the essential cloud for AI.

Thank you to our investors and analysts for your support. -- and engagement. We look forward to updating you on our progress in the quarters to come.

With that, we move to Q&A.

[Operator Instructions] Our first question comes from the line of Mark Murphy with JPMorgan.

Michael, every discussion we have across the AI landscape, we hear that bookings are booming, and obviously, that applies to CoreWeave. But the bottlenecks around power and manpower are just becoming so severe. Can you speak to that situation relating to the third-party provider. Specifically, is it a shortage of power or manpower? Is it something outside of that with GPUs, remember your storage?

And then have you spoken to your other third-party providers to get a sense of their own trending relative to schedule and whether they think they can hold on their deliver on their commitments into early next year.

Let me kind of take that question apart a few different ways, right? So first of all, you're correct. It is very frustrating for our clients. It's very frustrating for us. because of the kind of systemic challenges that exist within the supply chains that are necessary to deliver the global infrastructure. that's required for artificial intelligence.

Having said that, we have taken a number of steps along the way here to really drive home our ability to manage that environment, which is going to be challenging into the future. We've really spent a lot of time diversifying our data center providers. We have created a significant portion the company dedicated to being able to facilitate and assist with the operational component of delivering infrastructure. We set up our own self-build efforts, including Kenilworth and Lancaster, Pennsylvania. So you see us kind of really spreading out our -- and ensuring that we're doing everything that is possible to limit the damage associated with or the delays associated with delivering this infrastructure, which is just overwhelming the supply chains.

Now when you have a diversified portfolio of paths to infrastructure, the relative impact of each delay becomes smaller. You'd just be able to draw on different data centers as you're getting delivered. And so we really look at this as this is a significant block of infrastructure that's come on late. But that the ultimate end customer that's going to be consuming this infrastructure has shifted the contract back to allow us to be able to deliver the full contract value in spite of the delays really speaks to the value that the customers get out of our infrastructure. So you're going to be hearing this theme repeated again and again as you talk to not just CoreWeave, but you talk across the space. And it is a real challenge at the powered shell level. It's not a challenge for power, right? There's plenty of power right now, and we believe that there will be ample power for the next couple of years. But really where the challenges is the powered shell.

And so Michael, does this not relate to Core Scientific in any way? Or is this totally removed from that situation that you've gone through?

So I'm not going to speak to any specific 1 of the data center providers. We're working with all of our data center providers to do everything we can to facilitate the ultimate delivery of the infrastructure that they're going to deliver to us. We've had some incredible success getting infrastructure delivered to us as you continue to see us scaling. You saw us hit 500 -- approximately 590 megawatts. We're up 120 megawatts since the last call. So you are seeing a significant amount of success as we continue to scale, deliver. But I don't think it really matters who the individual data center provider is. This is a systemic problem that the industry is going to have to deal with for the foreseeable future.

The important part here is that -- or the important part from my seat is that the infrastructure, which is undergoing a delay is not going to impact our backlog and our ability to extract the full value from the contracts that we're going to deliver on.

Our next question is from the line of Keith Weiss with Morgan Stanley.

Congratulations on another super impressive quarter in terms of building out that backlog. You're right, we just -- we've never seen this in terms of any cloud provider being able to build out that quickly. Mike, I wanted to ask you a question that's been asked to us a lot that we're hearing a lot on CNBC. And it's really about sort of the risk of overcapacity. But I think it's more narrow than that people are worried about overcapacity from -- of what's being contracted by AI labs out there.

The question I want to ask you, though, is how we should think about your guys' infrastructure and the infrastructure that you build? And how fungible that infrastructure really is? When you're building out for a particular customer, do those data centers, is that usable for any customer? Is it usable for inference and training? Or do you really build to suit to a certain customer that would lock you in and give you kind of less degrees of freedom, if you will, if one customer is doing better or worse?

Yes, Keith, that's an excellent question. It's actually something that we've spent a lot of time thinking about here as we kind of proceed with our relationships with all our customers. And so in short, the infrastructure is fungible. It would be able to be transferred from one client to another, -- the infrastructure is built to the most demanding specs. So it's able to be used for training. It's able to be used for inference. We really have thought a lot about making sure that we maintain as much optionality as much flexibility within our infrastructure build as possible.

And I want to highlight for everyone that a lot of that flexibility, a lot of that fungibility really does tie back to the incredible software suite that we provide that allows for such effective use of the infrastructure, right? Like when SemiAnalysis did their annual kind of review of the alternatives out there, there's a reason that CoreWeave has come back time and time again as singular as the best solution for this type of infrastructure that exists in the world. And that includes the hyperscalers, the neo clouds and everyone else that's trying to deliver this infrastructure. We just do a great job, and we believe that there's a lot of value that we are protecting by providing such a robust software suite to be able to deliver infrastructure.

Our next question comes from the line of Kash Rangan with Goldman Sachs.

Impressive backlog growth. Two things that I wanted to just touch upon. One is, Mike, I think you've talked about how you're going to be diversifying your contractors and the latest to decide. Maybe you could give us honestly goodness update on how far are we away from potentially reaching a point where any disruptions that have nothing to do with your business should not affect your revenue outlook? How far away are we from that point?

And secondly, when you look at the developments, I mean, nobody expected maybe some did, but a $250 billion contract for open AI with Microsoft, nobody expected a contract opening with Oracle. All of a sudden, the -- certainly, CoreWeave has got a unique value proposition being able to stand up GPU clusters very quickly and very effectively at the speed of thought almost. But in a landscape, we're talking hundreds of millions of dollars being awarded to the hyperscaler giants, what gives you the uniqueness 3 to 4 years from now when things have sort of settled into a supply equals demand? When we look back at CoreWeave, what will be the shining value proposition that keeps you in the game at that point in time?

Let me break that question into 2 pieces, right? The first question you asked is about diversification and when does it start -- when does it stop kind of causing dislocation in our numbers as we're delivering them quarter-to-quarter. And what I would like to focus you on here is as the individual builds becomes smaller relative to the size of the entire portfolio of data centers that we are running, the impact of being a couple of weeks late will become less and less meaningful in the general accounting of what's going on, right? So when you're delivering 590 megawatts of power, and you have a step function of 200 or 300. It's a material percentage that's going to be delivered over the next quarter, right? As we become larger and larger and start to build out the full 2.9 gigawatts of power that we have, having a data center that's 100 megawatts delayed a week or 2 is not going to have a material impact.

And as Nitin said, we expect the overwhelming majority of that 2.9 gigawatts of power to be brought into service over the next 12 to 24 months. And so that will give you a really good idea of how the curve begins to become more smooth as we get larger and the relative impact of each data center becomes smaller. And so that's the first part on the scaling side.

The second part is the question you're asking has been asked of us since we started this business. Why is CoreWeave going to be able to deliver GPUs faster? Why are we going to be able to deliver the GPUs that NVIDIA uses to run its MLPerf? Why are we going to be able to create software that is going to define the space? And with each quarter, you see us extending the lead with which we have because of the customization of our cloud to the use case that is required. And once again, you saw us in the SemiAnalysis, like we're singular in this. We are out there building our product offering. We're building or buying additional capacity to further decommoditize the compute that we're delivering. And so a company that's built singularly to deliver this type of compute will be effective on a go-forward basis.

Our next question is from the line of Amit Daryanani.

Amit, are you on mute?

Operator, let's go to the next question, and we'll come back to Amit. Operator, can we go to the next question and we'll come back to Amit.

Our next question is from Tyler Radke with Citi.

Hopefully, you can hear me okay. So double-clicking on some of the delays that you called out in the quarter. Can you just help us understand the implications on 2026? I know, Nitin, you provided some high-level commentary on CapEx. But I mean, just given the visibility you have, particularly on the 24-month component of RPO, how should we be thinking about sort of the revenue implications of the shift? Is this a delay that you think kind of gets fully resolved into Q1? And should we see sort of a step up in growth rate next year relative to this year? Just any color on that would be helpful.

Yes. So I'll start and then I'll hand it over to Nitin. I think it's important to understand that the ramp that we are seeing is that a -- is associated with the infrastructure from a single provider. And we are parallel pathing with other providers for other contracts. And so you are going to see a short-term impact associated with this delivery. And then what you're going to see is the -- our ability to accelerate through the year back schedule. So the overwhelming majority of the delay that you're seeing should be taken care of within Q1 of next year.

Yes, Tyler, that is correct. The vast majority of the CapEx pushout that we experienced in Q4 will be done in Q1. And as you can imagine, we're going to ramp the capacity through the course of Q1 for this. As Mike earlier mentioned, the impact on the total revenue associated with the customer is not impacted here because we've been able to adjust the delivery dates associated with the customer so that the customer keeps the full capacity as well as the contract value associated with it. We will share more details around the 2026 build and our revenue plan associated in the next earnings. But as we highlighted in this quarter, given the strong customer demand that you see, given that is demonstrated in our revenue backlog growth as well as the continued customer demand we see, we expect 2026 CapEx to be well more than double of that of 2025.

Our next question is from the line of Michael Turrin with Wells Fargo.

I want to just try to tie some of the commentary together because the bookings growth clearly stands out, and there are a lot of questions just around the sequencing. And so it sounds like what you're saying is the supply chain impacts you're seeing are more single customer specific. What I'm trying to get a better sense of is does this at all impact the cadence at which you're able to sign on new customers? Or is this more tied to post ramp signing and one more specific customer environment?

And just as a small follow-up, does the NVIDIA deal specifically show up in the backlog metric? It might be useful to hear you expand on what that deal opens up given it's a bit of a different structure there as well.

Sure. So the -- there is no impact on our ability to bring on more clients. I think it's important to understand that we're parallelizing the build of infrastructure. There was a problem at 1 data center that's impacting us. But there are 32 data centers in our portfolio, all of them are progressing to 1 extent or another. And so that is -- each one of those is independent.

And as Nitin spoke earlier, we have 2.9 gigawatts worth of contracted power that will come on in the next 12 to 24 months. And we are going to be looking to fill that with clients that are going to be using that, which will have a substantial impact on our revenue on a go-forward basis. This 1 data center will catch up and then we will move forward from there.

Do you want to talk about the NVIDIA deal?

Yes. Michael, from the NVIDIA deal perspective, we're really excited about this deal. This contract allows for the capacity contracted and reserved for NVIDIA to be interrupted and resold to different customers. So the nature of this contract allows us to offer our services profitably to a wide range of smaller customers, such as high-growth AI labs that prefer shorter and lower upfront commitments, while eliminating any utilization risks for capacity from our side. So we're really excited about this.

Given the flexibility in the contract to interrupt and to resell capacity, accounting rules require us that we exclude the amount we expect to be resold to other customers from RPO. To be clear, if not resold, this capacity will remain committed to NVIDIA and will be recognized as revenue. So you see this NVIDIA contract in our revenue backlog, but not in our RPO to a large extent.

So just to follow up with that for a moment there. As Nitin said, we're extremely excited about this because what this contract is going to allow us to do is to provide infrastructure to emerging companies, startups, companies that are struggling to get access to the computing infrastructure that they require to be able to build their business. And so the interruptibility here is an incredibly powerful tool for the resiliency and opportunities for new companies to become part of CoreWeave's broader offering. And we're really excited about this. We think it's a great structure. It is a deal with NVIDIA. They fully underwrite the economics because we will sell the compute to them.

And I want to be clear that this really does represent an incredibly disciplined way of financing the compute in order to be able to reach parts of the market that we have been unable to reach or anyone for that matter, has been unable to reach up to this point.

Our next question comes from the line of Brent Thill with Jefferies.

Nitin, I just wanted to be clear. You cut CapEx by 40% for the year. And just to be clear, this is from one customer, correct? This is -- you're not assuming other delays across the board, correct?

That's correct. So this is associated with a single provider data center provider partner and the delays associated with that. And as we talked in our prepared remarks, most of it, a vast majority of it is going to be recognized in Q1. And in Q4, we're going to see a major impact on buildup of construction and progress associated with the buildup related to it.

And our next question comes from the line of Raimo Lenschow with Barclays.

Perfect. As we think about CapEx next year, Nitin and Michael, can you speak as well about the sources of funding a little bit because what we've seen for a lot of the other players is that leasing is coming up a lot more? You talked about CapEx, which is kind of what you need to do. How do you think about that path for you going forward between the different ways of kind of funding the business, which might give you even more flexibility?

Thank you. So look, we've driven innovation on the technology side, and we've driven innovation on the financing side, right? The way that I look at this is that we will look at the full suite of potential ways of financing and expanding our footprint. And then we will choose whatever is the most cost-effective way of increasing our scale and serving our clients. And so if leasing is the path, that's the path we'll go. But we've seen a lot of different structures. We've created a lot of different structures that have given us access to capital over the past 3 years and we believe that we're going to explore the full suite of those as we look forward.

We don't sign customers without knowing where the financing is going to come from. We go deal by deal, and we make sure that we have the physical data center spoken for. We have the power spoken for. We have the GPUs spoken for, and we have the financing spoken for in order to ensure that we are able to successfully deliver compute to them.

Our next question comes from the line of Amit Daryanani with Evercore.

Hopefully, this works better.

We got you.

All right. Perfect. Mike, I was hoping if you could just talk about as you shift from third-party data center providers to perhaps do more of your own self-build. How does that impact your CapEx and time to market for power as you go forward? I would love to just understand how do you think that optimal mix looks like and what the CapEx requirements could be as you perhaps go more to its self-build versus third-party data center providers?

Yes. So I want to be clear. We're not saying that we're going to go self-build and not use third-party data center providers. What we are saying is that self-build is a component of the way that you go about derisking delivery across the broader portfolio. And so we're going to go ahead, and we're going to continue to work with our partners who provide data center capacity that allow us to co-locate at their facilities that build facilities for us. All of that is going to continue to be true. We need that capacity in order to be able to continue to move and operate at the speed and scale that we are.

We just look at self-build as an additional piece of the puzzle. It puts us closer to the physical infrastructure. It embeds us deeper into the supply chain around the world so that we have firsthand information. We just think that you need to be on both sides of this fence, in order to be as effective as you can be derisking what is a complicated supply chain invent.

Our next question comes from the line of Brad Zelnick with Deutsche Bank.

Mike, with 2.9 gigawatts in committed power, and over 1 gigawatt yet to be contracted out to customers. Meanwhile, we continue to see a number of other large deals get announced industry-wide, how do you think about and how might you frame for us the pacing on contracting out the remaining capacity, given the demand is insatiable out there?

Yes. So look -- thanks for the question here. The fact that there are other deals getting contracted out there is incredible validation for the supply-demand environment that we have been describing for years now, right? There is no entity that has the capacity to be able to deliver infrastructure globally in order to meet the demand that's being driven by the largest technology companies in the world by the largest AI labs in the world by government, by enterprise, all of these things are coming to bear. And so your -- the fact that there are other deals going to other players is part and parcel for the fact that we like the hyperscalers, like the AI labs, like the data centers are being overwhelmed by demand. It is just reinforcing and validating the theme that we've been talking about.

We think that at the end of the day, the product that we deliver, right, which is a full stack everything from the hardware, all the way through the software is the most valuable representation of this infrastructure that can be delivered to the market. And we continue to think that, that will drive a significant amount of demand for our infrastructure. As far as the remaining capacity goes, we're being very thoughtful about continuing to drive diversification across our cloud. we're continuing to think about different applications that are going to be meaningful contributors to the way the world will work in the future. And we are allocating that infrastructure to those parties as quickly as we can in order to ensure that they are successfully able to launch their products, their enterprises.

And Brad, a couple of things to kind of keep in mind here as we kind of talked about in our prepared remarks. Today, approximately no customer represents greater than approximately 35% of our revenue backlog, which is meaningfully down from where we began the year at 85%. And 60% of our revenue backlog is with investment-grade customers. So vectors that we are very thoughtful around as we take care of the capacity that we have available to be sold.

We have time for one final question. Our final question comes from the line of Brad Sills with Bank of America.

I did want to ask a question around this concept of the power shell a bottleneck here, Mike. Is there any IP that CoreWeave has that you contribute to the build-out of these data centers any learnings from this delay that you might be able to apply to other contracts? Just trying to get a sense for how much is in your control here to kind of solve for this bottleneck issue that you're experiencing with this one contract itself?

Yes. What I would say is, Brad, I don't think that I would say that our learning has come from this one delay. We've been operating in a systemically supply-constrained market globally now for 3 years. We understand how difficult it is. And with each additional wave of demand, the market gets tighter and tight. So when you ask what are we doing to position ourselves on a go-forward basis, what I would really encourage you to think about is the fact that we've built out an entire organization within CoreWeave that is capable of helping us build and deliver additional capacity on the self-build side. That's where you embed yourself into the supply chain. You understand where the power is, how it's being contracted. You understand what it takes to build the powered shells because you're doing it yourself in addition to the fact that you're using other third-party providers, those are the type of relationships that will enable us to be as successful as possible in what is going to be a challenging environment for quite a while.

And that concludes our question-and-answer session for today. I would now like to turn the conference over to Michael Intrator for closing remarks.

Thank you all for joining us today. As we wrap up, I want to emphasize how proud we are of the strong foundation we built this year and the incredible momentum driving our business forward. Our team's exceptional execution to build the essential AI cloud has positioned CoreWeave to capture a significant and expanding market opportunity. We appreciate your support and engagement, and we look forward to updating you on progress next quarter. Thank you. Have a good night.

This does conclude today's conference call. You may now disconnect.