CoreWeave, Inc. (CRWV) Earnings Call Transcript & Summary

Okay. If you don't mind taking your seats, we're going to get started in a second. Thank you. Thank you. Thank you. Thank you. 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 that is from Deborah Crawford, an IR, who is sitting somewhere here. With that introduction, welcome to Goldman Sachs Communacopia and Technology Conference 2025. We're delighted to host you, Mike and Brannin. Brannin, you have been to one of our private software company conferences, I think in the past couple of years, you're a small company back and now you're a big company. So Mike, welcome to the conference. I wanted to just start off with the big picture. Congrats on the IPO, first of all. And wow, what is stock performance since that debut.

It's been exciting.

Yes. It's been super-exciting. What is your vision? Where do you see CoreWeave going in the next 5 years?

Yes. So first of all, thank you for having us. It's exciting to be here and speaking as a public company, and I'm sharing the stage with Brannin, which is the first time that Brannin and I have done this, so we'll see how that goes.

It's going to go great. I'm going to assure you that.

So look, the strategy of the company, in many ways, has been relatively simple right? It's to provide the leading edge infrastructure that is required to serve evolving compute requirements from use cases period, right? And like I said, it's pretty simple. So when I think about the company 5 years forward, I think that, that is a pretty resilient vision for what the company is going to try to do. What we have really focused on as a company, what I think we've been exceptionally good at as a management team is thinking about the tactical steps that are required to allow a company to be effective working towards that vision in an incredibly fluid dynamic and high-growth industry like parallelized computing has been for the last several years and how it clearly will continue to be for the next several years. And one of the things that's incredibly important to focus on when you're thinking about our company, our industry, parallelized computing, artificial intelligence, regardless of how you deconstruct your thought process around this is just we have been unrelenting in our assessment and how we've positioned the company tactically around the demand signals for the infrastructure that our clients are telling us they need. And there are -- I think about it very much along the lines of the forest and the trees kind of approach to, right? There are lots of noise in the space, right? That kind of make observers, investors, entrepreneurs consumers alter their view of the space on a day-to-day basis and that's normal and expected. But in order to be effective in this space and in order to be able to function in this space for our company, you have to take a -- you have to zoom out a little bit and understand that we have been positioning our company, we have been talking to the market. We have been investing across the stack because we believe and this is certainly true at this moment that there is enormous staggering and unrelenting demand for compute as the world builds a global computing infrastructure to serve artificial intelligence at a size and scale and magnitude that the world has never seen before. And that's the job of our company, right? Our company is to position ourselves. You've seen it in the deal flow in the headlines within our company, within other companies that occupy the space. I was joking in one of the earlier meetings, like we haven't been kidding, right? Like we've been saying the demand is overwhelming. The demand is overwhelming. The demand is overwhelming and then there's been information Microsoft stepped away from a transaction and the market kind of moved sideways for a moment. But like that's a company, the space has never wavered.

And just before we got on stage, I was saying -- telling somebody that tomorrow is going to be amazing. But today is amazing because of this. And we had a Google public cloud just present earlier today. we're going to have open AI, we're going to chat with them later today, and we're going to have a SaaS, Defend and Salesforce CEO wrap up the end of the day. So today is absolutely an amazing day. I'm curious to hear since we first met, I mean, Brannin was first executive at CoreWeave that we got on a couple of years ago and then subsequently later you. This is the one question that we used to get on the IPO road show. Okay, I get it. CoreWeave does this. Pure first AI only pure hyperscaler. Well, why do they win? Why do you think you win in this marketplace that everybody knows is populated with the giants, the $1 trillion multitrillion-dollar market cap giants?

Yes. Look our IPO wasn't particularly funny. We kind of came out into the teeth of Liberation Day, tariffs. And I believe that was the right decision, although it did take some -- tested our fortitude and those of our bankers and our investors and a few other people along the way. But I believe that the company needed to be a public company. And that is why we followed through on the IPO. And when you think about the company and why the company succeeds. I really kind of think about it in 3 buckets. And I've organized the company or we've organized the company around those 3 buckets, right? The first component of why we are successful is we built a beautiful technical solution to a problem. It didn't have a solution, right? We built a software architecture around the serving of parallelized computing at scale. And the cloud, up until that point had been built around serving sequential computing with this wonderful solution that just wasn't appropriate for the way that compute was going to be consumed in an age of artificial intelligence that required massive scale parallelized computing. And so we had a enormous advantage. And it's hard to say that as a start-up company when you're phasing down the largest, most powerful companies in the world. But we did have an advantage in that we had no legacy business, and we were able to sit down with a blank sheet of paper and say, okay, deconstruct the problem. Now that we've deconstructed the problem, solve the problem and don't solve the problem with any legacy baggage, just solve the problem, like in its simplest, most elegant form solve the problem. And we did that. And we have a great technical solution. We have a great software stack. We have an incredible ability to provide the consumers of this compute a uniquely fantastic environment to be able to consume compute that is difficult to consume at a size and scale that is difficult to serve. It's just a great environment. When you talk to our clients, they will say it over and over again, like, "Oh, it's a CoreWeave cluster, right? Like it's special. And it is special, right? Because it really works so well for what their use case is. So that's the first thing you've got to do. And that's like the starting point, right? The second thing you got to be able to do is realize that Man, it's grizzly physical, right? It is a very physical business you need to be able to stand up data centers, you need to be able to manage power and you need to be able to manage the delivery of a physical product, which largely within the tech sector, there's huge portions of it that have just outsourced that to the cloud. And when we decided we were going to be a cloud, we were taking on that mantle, and we did. And we really have focused enormous amounts of our company, of our energy, of our resources on being able to build and deliver physical infrastructure at massive scale. And our clients depend upon us to be able to do that. And then the third kind of area that I think we were almost uniquely good at. And this is really where Brannin spends most of his time is like this is a capital markets problem, right? This is a -- I was sitting with [ Masa ] and we were talking to it, and he was just like this is a war of capital. And I just thought it was like an incredibly insightful moment, right, where like I was going through all my things, and he just boiled it down to a language that he spoke. And I think that's true. And what we have done, right, like not throwing shade at Silicon Valley or West Coast Capital. But I do believe that the capital markets kind of break in 2 directions, right? West Coast Capital is wonderful. Venture capital, they change the world. They invest in technology that's going to reconstitute the way we live our lives for good or for -- and they're looking to invest money into companies that can do that and can do that at a massive scale. But you can't build an infrastructure company that way. In order to build an infrastructure, and this was a unique skill set that we had because of our background is you had to go back to the East Coast and the East Coast capital, which is governed by the debt markets, has 1 rule, which is do not lose my money, right? And that's it. Like that's kind of...

Pay your rent on time.

Pay your rent. And so what we were able to do is we were able to marry up these 2 pools of capital in a way that really hasn't been done very frequently. And create and engineer products that had the capacity to move the world forward but also could be underwritten by the debt markets. And we have gone in, and like I said, Brannin's team has gone in again and again and again and tapped into the capital markets in ways to enable us to punch above our weight. And we have built incredible thoughtful debt structures that support SPVs and amortize within the 4 walls of the contract so that we can be able to do what differentiates compute, which is scale, right? Like everybody can run a GPU. The question is, can you run a supercomputer? A matter of fact, can you run all the supercomputers, right? Like that's the part that decommoditizes the compute that decommoditizes swaps is being able to do and deliver this at scale. And so those are the 3 pieces.

And I would acknowledge within those 3 pieces, it's a 3-legged stool, right? You cannot execute where we're at and where we're going with all 3 of those working in tandem. You could have all the capacity in the world and the right products, but if you can't finance it. It just doesn't work. And of those 3 variables, I think that the financing aspect is probably the most underappreciated in terms of just how critical it is to allowing our business to continue to accelerate. And we've exited the ability for a single balance sheet to solve this problem, right? You now have to assemble many balance sheets across many different types of securities to approach this market, and we'll continue to be innovative on that front and finding those opportunities.

CoreWeave has raised over $25 billion in the last 18 months to drive our business forward across equity markets, high yield markets, debt markets, it's been an incredible accumulation and delivery of capital to solve this problem.

When you talk to investors, not that you talked to investors all the time, but what is well understood increasingly in the last few months since you've been a public company. What is like a source of frustration, why don't people get it? It's such a misunderstanding about CoreWeave. And then you feel so high conviction about something that's clearly misunderstood what would that be?

We'll probably both have answers to this. I think that technology piece is better understood now. We had a phenomenal report come out of semi analysis that did this first benchmarking of what is a GPU cloud and what drives these differences because ultimately, there isn't fungibility in this infrastructure. And H100 is -- doesn't have the same performance no matter where you plug it in. It matters deeply how it's operated and how it's configured and especially to be able to do so at scale in a supercompute configuration. And I thought that, that was the first report that succinctly and went into a deep technological dive of why is CoreWeave singular in its performance? Why do the world's leading AI labs and enterprise adopters of AI choose to work with CoreWeave and keep coming back to our platform. So I think we've made a lot of progress there, and we're happy with it, but there's still room to go, right? Like it's a question that still comes up. We're continuously innovating and driving our product forward, and I think there will be a lot of opportunities for us to continue bringing that technology advantage to our clients through our proprietary software solution.

I'm glad you mentioned the proprietary software solution. Would you care to elaborate and maybe help educate our investor clients. And maybe me too, what is the secret sauce with your software? So when we worked on the IPO, we had brilliant presentations by some of the engineers, top-notch engineers you had from Google, Meta, et cetera. Tell us what's happened to the secret sauce that ties it all together?

If we had a couple of hours, we can go the full thing here. It's funny. It's not a silver bullet solution. It's not like that one thing, we got it. These are tens of thousands of engineering decisions that are being made of how do you optimize around the workload. That's the engineering problem. At the end of the day, it's parallelized workloads at scale on supercomputers. And that's very different engineering problem than what our peers are presented with, right? They solve around how do you run host websites and store data lakes at scale. And they built a phenomenal solution for that and did it globally. But now this thing is just fundamentally different. And again, I thought that, that -- the analysis board did a great job of going through it, but it's in how you provision and operate the infrastructure.

Kash, I'm going to -- you asked the question, and I'm going to take it -- since I'm sitting with a room of investors, the part that...

Room full of investors.

Wonderful investors. Great investors. So I've said this a few times, and I think it warrants repeating, right, which is we have a different business model than most tech investors have encountered in a very long time. And what I have said again and again is that if you give me a chance to live with the investors and they get to watch us build our company and they get to watch us execute and they get to watch the clients' return and use us, they will increasingly become comfortable with what we do. And so there's this period of time where the initial response is, "Oh my gosh, look how much debt do you guys have, right? And now like all of a sudden, a lot of people are stepping back and saying, "Oh, wait, but the debt sort of makes sense. I understand how it works. And so the business model is a work in progress as we continue to introduce it and give our investors an opportunity to live with it. And the way that I've talked about it internally is like when you roll back the clock, right, when we go back to when we initially began to really scale the business. And I think about how long it took myself and Brannin and our team to kind of work with the Blackstone team so that they could understand like what a GPU is and what the technology is and what the client...

It does not love at first sight.

No.

Nobody loves me at first sight.

So it took us literally 9 months. When I go back and think of like our like early investors, like Fidelity, like it took us a year of explaining and reexplaining and reexplaining how it works. It required us to show them not just on the spreadsheets, but actually in the product before they became comfortable with it. And like I think that over time, as more and more people get to watch the model work as more and more people get to watch us execute on the business at scale. More and more people will begin to get comfortable with that. We'll get comfortable with the capital structures, the business model and all that. And so like my biggest frustration is it's hard to watch the clock on that. But we're making progress every day. Things like this help us make progress every day.

Got it. I remember working on the Salesforce IPO in 2004 or so. People didn't believe there wasn't -- we didn't call it the cloud back that we call it software-as-a-service and people did not want to believe that model. It took 4 or 5 years then was a 15-year run.

Yes.

I'm curious to get your take on the current demand environment for AI compute. Where are you seeing the strongest demand signals?

And Mike articulated this earlier, we've been consistent in this messaging of there is no ability to solve the demand profile that is in the market with the capacity that's available today, right? And capacity being powered shell data center infrastructure. That problem is continuing to persist and is honestly worsening. I would say what we've observed over the past 4 to 6 weeks is yet another inflection in demand. And that demand is...

I think that I'm curious.

It's inference. It's the actual consumption of this product. And I know that that's been a market topic for 18, 24 months at this point. It's like when we see the revenue and like what -- when do we see users interfacing with this product like that is happening and it's accelerating and the clients are coming to us saying that they need this infrastructure. And what's interesting is it's not thousands or tens of thousands of GPUs. It's hundreds to millions of GPUs. It's at massive scale and at these volumes that no one ever anticipated before. So what is the demand climate today? I'd say it's as tight as it's ever been. There is no solution in sight to be able to bring enough infrastructure into the market to solve what it needs to continue scaling.

That's great. And if you could characterize where we are in the innings of training versus inference and how do you win an inference as you won in training?

Look, I already said we won in training. Our clients don't come to us for training or inference infrastructure. They come to us for AI infrastructure and AI infrastructure can run anything and our client profile moves smoothly between training and inference. And I'd say it has a lot more to do with the life cycle of the architecture, right? Like at what point are you at? And the first 6 to 18 months, it's a heavy focus on model training because all these entities that get access to our infrastructure, generally ahead of the rest of the market because we're first to market to bring it online, focus intensely on advancing their models, their next-gen foundation models. And then they transition over to inference. And that's what we've seen within Hopper. And as we brought Blackwell online, there's been a very intense focus on training and then move to inference. But importantly, like it's not training, it's not one or the other. They use it for both, and that's an incredibly important aspect to our clients to be able to move in between the 2 functions.

Got it.

You're seeing like the compounding effect of different portions of the economy moving through AI. And in an environment like think of it from like a power market, right? It's the marginal unit that defines price. It's the marginal unit that's going to define demand. Like you've got your AI labs that are building the future of the world. You've got enterprise that is starting to incorporate it and finding real value and incorporating it into their business models. You've got sovereigns beginning to move in a meaningful way to try and get their arms around the infrastructure. And it's just -- the truth is with each additional demand signal that comes in, the market is further on tilt. And I've always said markets are really good at solving these problems. Like that's what they're built to do. They're incredible efficiency machines. But time and infrastructure are not matching up, right? And so the speed with which the space moving, right? I always say price moves faster than infrastructure can be built. In this case, it's -- demand is moving faster than infrastructure can be built and the situation is, I would say, 100%. We are losing ground on our ability to deliver the infrastructure that's required.

Excellent. How would you characterize the current pipeline, the deal pipeline? What kind of customers are you in discussions with? And is there still an appetite for large multibillion-dollar deals?

I'd characterize it as we support the world-leading institutional consumers of AI, right, whether it's AI Labs, our peers within the hyperscaler network, enterprises, they all have come to us for our product and its differentiation. How is it evolving? I think people are increasingly focused on longer duration exposure to the infrastructure whereas a year or 2 ago, it was a 3-year contract that entities were looking for. Now it's 4, 5, even 6-year contracts for the exact same set of compute. And just as a reminder, the way that we want...

We just want to be locked in and give that -- get that assurance.

Yes, it's the assurance, but it drives home a lot of the market anxiety around just how long will the infrastructure be used for. Clients are coming to us saying, I want that GB200 infrastructure for 6 years, and they're willing to enter into a fixed take-or-pay contract at fixed economics for that entire duration. That's a really astounding thing when you think about it from the way that technology is being consumed, it's a really interesting tool for us to work with on a financing aspect as well, but longer duration, larger scale are the 2 variables that are driving our pipeline right now.

I didn't intend to ask, but you said it has a financing implication. So the longer duration, larger contract, what does it allow you to do from a financing perspective? What are the opportunities that are available to you to fund?

Sure. It just continues to expand our pool of tranches of capital that we can get into. I mean, look, our focus is investment grade, right? That is our guiding North Star on how we are managing our balance sheet and our risk exposure is how do we go achieve that? I mean in the last 12 months, we dropped our non-IG cost of capital by over 900 basis points.

Astounding.

Astounding for us to be able to do that right? And we're going to continue classing it so that we have as close to parity as possible within our peer group, some of the largest companies on the planet. So it's affording us an ability to move into like the insurance tranche. It's affording us an ability to move into the public debt markets and work with a diverse set of balance sheet to fund the AI infrastructure build-out.

Leverage compounding at multiple levels, again, demand signals are strong. Revenue growth is good. Return on assets is strong. Cost of debt coming down, so leverage returned even higher. And thanks to Nitin, great job on operating margins, the profitability on a non-GAAP and even GAAP basis seems to be.

You can do my next earnings report. I'm good.

Let's talk about -- by the way, are we -- Deborah, is it okay to take questions from the audience. I can't see Deborah.

Yes, that's fine. We are fine.

Anybody has any questions, please just raise your hand. And if you don't, I have like 1 million questions. No, I don't have 1 million, but I have a few. Yes, please.

Yes. amazing deal. gargantuan scale. There is no way to interpret that, that I think is reasonable other than holy cow, they want to buy infrastructure at scale wherever they can find it on whatever time line they can get it. And that is a deal. My expectation based on my conversations with the big buyers in the market is that, that didn't fix the problem. Let's put it that way. And that there is a huge and sustained backlog for additional infrastructure. There's a huge and sustained backlog by many of the players that need to build compute. So I think great job for them. And it's a fantastic deal, and it is a clear indication of the depth of the demand that we're seeing in the market.

Which brings me the natural question. Your capacity profile today and how do you see that -- how do you see your plans to scale your capacity in the future because we talked about the demand, and we're good on demand. What about the capacity?

Yes, I'd qualify it like we've been preparing for these inflection points in time for years right? You can't make this powered shell capacity just exists out of nothing, right? This is capacity that we've been contracting for identifying, diligencing over the past 2 years. And today, we've assembled a portfolio of -- it's about 2.2 gigawatts of capacity that's coming online. By the end of this year, we'll have 900 megawatts in operation. You've got an incremental 1,300 meg of delta there that we're able to be active in market with right now. And what we'll just keep reiterating is we are going through a period of demand inflection right now where there is only a limited amount of availability of capacity in 2025 and 2026, and we expect for it to move pretty quickly.

Got it. We talked about the debt markets several times here. You've raised cumulatively, I think you mentioned $19 billion, $25 billion.

$25 billion.

And since the IPO, the cost of financing has come down. Can you talk a little bit about the future financing needs of the company? And how do you expect to balance raising equity versus debt?

Yes. So I think we've proven that we have a pretty sophisticated approach to the capital markets. And I hope we've proven that. We think that there are certain components of our strategy that are just incredibly appropriate to continue to finance with debt structures, right? Like you're buying a depreciating asset that has a contract that will pay for the infrastructure, the OpEx, the profit margin. It's a wonderful tool to go out and use debt to build and expand your footprint on that side of the house. When you look towards other components of our business, when you look towards some of the M&A stuff, we finished up a deal with OpenPipe. We purchased Weights & Biases, great examples of us extending up the stack, providing wonderful software solutions to be able to make our clients stickier decommoditize the compute that we're delivering, do all the things that are really important for long-term success of the business. We've used our stock in those acquisitions. We did the IPO, we did several rounds before the IPO pre-IPO rounds as we were building up our capital base. And I think that we will continue to be thoughtful around how we manage our cap structure. And I think that the majority of the capital, the overwhelming majority of the capital really will live within the debt market because that's an indication of the growth of our business from a client-driven success perspective. And that's really how we've built our business, it's how we scaled our business. It's how we've been able to deliver products that the debt markets can underwrite. It's all been success-based.

I love how you're breaking new ground, breaking new barriers and breaking conventions of what is it the way to fund a start-up company, et cetera. So congratulations. And here's to the journey ahead. I hope that you will keep coming back to Goldman. You're invited to come back to 2026 Communacopia and 2027 because we want to see the 5-year vision. We're going to come back and ask you Mike, you said that like you said this in 2025, We'll do it again.

Okay. Thank you very much.

Well, a round of applause for Mike and Brannin. Thank you.