CoreWeave CEO on Nvidia, AI, and Building the Cloud

[MUSIC] Hello, everyone and welcome to AdBaron's I'm Andy Surware. And welcome to our guest, Mike and Trader's CEO of CoreWeave. Mike, great to see you. Thanks so much for joining us. >> Thanks for having me. >> So CoreWeave is right up there with Nvidia and the Mag7, one of the most consequential companies on the planet right now. So there's a lot to talk about. I guess I'd like to start by asking you Mike to explain CoreWeave's business model. How does it work? >> That was an aggressive placement of CoreWeave. We're an important part of the ecosystem for sure, but those are some of the largest companies. >> He's being modest. >> Yeah, so look, CoreWeave's role in the ecosystem, right? Like the way that we built the company, the way we conceived of the company was from a position of trying to build, trying to pull together the best silicon solution, the best software solution, and the best infrastructure solution, and deliver that to the market so that consumers of the accelerated compute would have the best experience as they build product, as they serve inference for their clients. All of those things really come from the integration of Nvidia's silicon solution, our software solution and our ability to build and deliver the infrastructure at scale, which is an incredibly important component of the debate, right? Because you are required to be able to participate in the market at scale, or else you truly don't matter, right? Like anybody can run a GPU, but can you run a super computer that has the capability of training one of the bleeding edge foundation models? And that's the part of the space that CoreWeave tries to live in, right? Serving inference at scale so that companies that have built wonderful products are able to sell them broadly and effectively. That's what it is. Okay, so maybe for a 60-something fairly ignorant journalist to break that down a little bit more, your customers are the ones that are making LLMs, developing those large language models, right? Creating AI, and they need Nvidia chips and tremendous amount of computing power. And you guys have the chips and then software layered on top of that. So instead of just buying the chips and building these sole systems, which are software and chips together, they can just go to CoreWeave for a solution. Yeah, so is that accurate? Yeah, yes. It's pretty good for a 60-something. Okay, then you got to do it. Thank you. Look, you know, running this infrastructure is very difficult, right? It is very challenging, it breaks, things need to be restarted. And the way that we have been able to build the company, the way that we've been able to kind of disrupt what had been a very, very static business. The cloud business really was dominated by obviously three hyperscalers. And it was extraordinarily difficult for the past 20 years to make a real attempt to dislodge or to provide an alternative service that could participate in the cloud at scale. And you know, it was the confluence of accelerated computing, right? So the real, the birth of these incredible GPUs that allow for certain type of computational acceleration. A new technology, right? And all of this is intertwined, right? And that's artificial intelligence and the birth of a new way of using the cloud. All three of those things came together and allowed us to fill void, which allowed us to build the company that we've been running for since 2017. And it has been a combination of the different pieces of the business that allowed us to kind of accelerate into the space the way we have. Right. Can you give an example maybe using one of your customers, your big customers? And for instance, are you running physical data centers or give us some more color if you can, like maybe maybe a good way to do this is, let me tell you how I think about the business, how I construct the business in my mind and how I try to like figure out how to build and accelerate and continue the growth trajectory we're on. And the way that you do this or the way that we did this, probably better way of saying it, is there are three kind of macro components that had to come together, right? The first one is that we had to build a technological solution to a problem, right? It is a beautiful solution. And the problem was parallelized computing, the way that artificial intelligence, CGI rendering, batch computing, medical research will be done going forward. Is through something called parallelized computing. And that is different than the way that cloud 1.0 came into existence. That was sequential computing, right? And that means basically you solve one problem, move to the next problem, move to the next problem. But parallelized computing requires you to solve simultaneously across a massive band of compute and step and solve and solve. And the GPU allowed that to happen and our approach to the cloud was, okay, this is a fundamental change. This is a totally new way of consuming compute. What are the building blocks that need to be brought to bear in order to be able to do that effectively? And that was the technical solution that is represented by the software layer that Core Weve has produced that is the best solution for solving for that type of cloud in the world. And it's an incredible solution. We are years ahead of everybody else and kind of configuring it and our clients absolutely love to use our solution because it is so much more effective than other solutions that are out there. And so that's the first piece is you have a technical problem you got to solve. The second piece of the problem is you're building compute, you're building the cloud at a size and scale that defied the imagination until we were here, right? It is so massive the build out that's going on. It is a planetary scale computing build out to be able to support artificial intelligence today and into the future. And you know, for people who come from the energy world, this is like the base load build out. And it's going to be the engine that allows us to continue to scale solutions in artificial intelligence. And that's the second piece, right? It's a physical business. And that has been one of the components that's really hard for, let's say tech investors, in particular, who have historically invested in capital light businesses to understand like this is a capital heavy business and that has become increasingly obvious to everybody as you see the hyper scale has come in with repeatedly increasing the capital budgets that they're going to allocate to building this infrastructure. And then the third piece of the business that was a necessary component was to understand the magnitude of capital that was required to build this infrastructure. And you know, when I think about our business, you cannot do and you cannot build this type of cloud without having each of those three components working in synchronization in order to drive the company forward. I especially want to talk about the financing part. And as people may be surprised, believe it or not, like is not a technical person. He's a Wall Street person basically, which is very unusual, obviously, for a tech company. But just again, to go back a little bit to what you do, and then we'll move on a little bit more. So are there core-weve data centers? Yeah, there are. Okay. How many do you have and where are they, for instance, like? So we have, when we look at the data center space, we have self-built data centers and we co-locate an enormous amount of our infrastructure in third-party providers. And so that's a host of different public and private companies that provide infrastructure to us. Oh, these are some examples. Can you give some examples? Equinox, switch. Digital Bridge has a bunch of companies. Like, the who's who within the data center space? And the, you know, when I started this business, I didn't really intend to go into the data center business. I felt like that was more of a real estate portion of the business. And it was hard enough to walk into investors and say, hey, I got this great idea. I'm going to get a bunch of folks together that I really, really want to start a company with them. We're going to tackle one of the largest, most capital intensive, static industries that exist, which is the clock. And then you have to tell that to a real estate guy and a laugh. Right. What we really wanted to do is we wanted to piggyback off of of the great providers in the space that provide physical infrastructure. However, as the energy markets have tightened, as the buildout has progressed at the pace and scale that it has, getting more control over the bricks and mortar at the data center level has become a strategic imperative for the company. And so while we will continue to use our third party partners to help us accelerate and diversify our locations, and build infrastructure as fast as possible, we will also be doing, you know, corporate data centers. And it will be a combination of both as we move forward. - Right. - And are they all in the US at this point? - No, no. We have data centers across the US and Canada. We have data centers across Western Europe up into the Nordics. You know, and we continue to build and scale the company as we try to establish our global footprint, you know, to be able to serve our clients. And one of the things that's been a mantra for me has been to allow our clients to lead us, right? They tell us where they need infrastructure. They tell us the type of infrastructure they need or want in order to serve their clients and serve their business needs. And so, you know, we are being pulled increasingly around the world by our clients who love our solution and want it to be in more and more regions to be able to support the workloads that they have. - So you said dozens or scores or how? - Oh, no, we have like, I'm trying to remember the number that we use at the last earnings. I think it's like 42. - 42 says someone from the audience who knows what they're talking about. - Well, that's dozens. - Yeah, okay. - Okay. - Yeah, we got dozens. - Okay, that dozens, come on. - Soon you'll have scores. - Yeah, I think scores, what you could say-- - We got two scores. - You could say scores now, but you kind of need more than 60 to really do a score. - I'm working on it. - I'm working on it. - Yeah, okay. - Two minutes to the next turn. - You're a masked guy, you should, come on. All right, and so you have this very close relationship within video, right? And describe that. I mean, you are a huge buyer of their chips, I mean, it's a starting point, right? But it's more than that. - Yeah, look, Nvidia has been an absolutely incredible partner for Corviv. And, you know, I always, I describe the relationship as a symbolic, but not equal. - Symbiotic. - Symbiotic, but not equal. - Yeah, and what I mean by that is that, Nvidia builds the most incredible technological solution to this problem in the world. Their GPU has driven their company to be the most valuable company in the world. And, you know, they are, you know, the folkroom on which artificial intelligence was built, you know, quite honestly. And, but their technology is complicated. It's hard to get it to work the first time. And what our role is, the software layers that we built, allow them to go ahead and give us their infrastructure at which point we build it and troubleshoot it at scale, right? So think of it as Nvidia is working in their labs. They do an incredible job building this technology and they run it in their labs at sub-production scale and then they look to push it out so that people can go out there and test it and drive it and run it in, you know, the real conditions that they're going to encounter. And because the software we built is so good, they really look to us to be able to do that, which is why when you look back over the last couple of years we're the first ones with the H100s at scale where the first ones went, you know, H200s, we were the first ones to deliver the GB200s. Jensen just came out and said, "We'll be among the first to deliver the VR Rubens' queue." And it's really just that idea that like, "Hey, you know, let's get it out there in the wild and see what happens." And we're going to get it out in the wild faster and more performant than other folks. And I think that plays a real role in the feedback loop that they need to understand what they have to do and how they need to deliver infrastructure to the world. - VR Ruben, the new chip architecture coming out in the video later this year named after the famous astronomer. - Yes. - Right. - Each generation is after a scientist. - Right, yeah. And what is this future world of AI, the AI economy going to look like to your mind, Mike? - You're going to ask the hard questions. - Yep. - So look, I think that, you know, as you spend time with different people that populate different parts of the stack, whether it's people who are at the physical data center side, people who really build inside of the data centers that compute the software layer, as you continue to move up the stack into the application layer, you begin to understand the speed and breadth with which the promise of AI is starting to get traction. And what I believe we are going to see over the next several years is a continued acceleration in the use cases we're going to see a continued acceleration in the dropping of price of tokens. We're going, that is going to cause a feedback loop of new companies coming into existence because the barrier to entry will come down. You know, I sat on a panel with Sarah Freyer from OpenAI and she said, you know, when they did a million tokens for CHAT GPT-3, it cost, I think it was $39 and now it costs nine cents, right, to give you an idea of the incredible acceleration and decreasing cost associated with these tokens that are the lifeblood of being able to build an experiment with infrastructure. And so I, you know, I just think that we're in for a, an incredibly interesting world that's just going to accelerate into it. And like, you know, there's a piece of this which is, look, I drank the Kool-Aid, right, like I'm all in on this stuff. But it doesn't take a huge amount of imagination to see that the future that I just laid out is directionally correct because the traction across companies, you know, as I say, you know, like, you know, my kids never are going to do homework the same way again, right? And it's not that the way that they do homework is good or bad or indifferent. It's just going to be done differently. And I think that's amazing. I think that's what the world needs, right? Like, because now not only is my kid gonna do that homework, but there are gonna be kids all over the world in places that wouldn't have a chance to do it to the same level, to the same quality or gonna have that opportunity. And that's great. - All right, you sort of preempted my next question, which is the bubble question. I know you're the anti-bubble guy. So I don't even have to ask you that question. But, well, now I wanna ask you this question. You can go back to the bubble a little bit. But you have, you talked about a relationship, talked about within video, you mentioned OpenAI, you've got a relationship with Meta. There's people who are sort of suggesting that there's these overlapping financial relationships, circular finance, you know where this is going. People are saying it's a big, correct sue I guess would be a phrase and not in a good way, going on with OpenAI amongst these big players. How do you respond to them? - So, I was on a call for Weff a couple of weeks ago. And interestingly, they paired me with a CEO of a mining company. And he was, you know, we were talking about our businesses and he started to talk about the challenges that exist in the copper market. And, you know, copper is going to be very tight, very tight supply. And the way that he wanted to address that was he wanted to, or the way he thought the only way that they thought they could address it was to work together, right? And so this company is gonna work on mining and that company is going to work on smelting and this company is gonna work on processing and they're all gonna work together to address what is a systemic shortage in the copper market. And I said, that's a great idea. But if I say that, man, it's gonna get ugly, right? Because we are facing a systemic shortage in compute. And when you see those designs that like the media loves to print because they're kind of cool looking designs and, you know, and it's a good story. But like, you've got to remember, like, we're trying to build an infrastructure at a pace the world has never seen before. And we are systemically pinned on demand, right? I cannot deliver enough compute. And it's not just that I can't, the Neo Clouds can't, the hyperscalers can't, the chip manufacturers, and nobody can get enough infrastructure delivered. into the market. And so when you, in an environment like that, what you're doing is you're working together to try and accelerate the business to deliver this infrastructure, this computing ultimately that the world is desperate for. And, you know, like going beyond just the physical limits, you know, where you're going to ask me about like what's, what's stopping us from building, you know, like, you know, more compute it's going to be, hey, you know, I'm going to talk about the power shell, right? Because that's the part that I feel. But like, you know, like you're seeing it everywhere, right? Like, you know, you're going to see it in energy, you're going to see it in chips, right? Like I mean, like, you know, you talk to, you know, micron technologies and, you know, they're just everyone is redlined trying to get enough of this infrastructure that the world wants to market so that they can build and develop and serve their clients. Other constraint, we've touched on this before and I mentioned your finance background. So I want to get into that a little bit more, another constraint is capital. And you came from this Wall Street background, which I love you to talk about a little bit, but it was the financing element and the tens of billions and maybe hundreds of billions plus trillion that is going to be put to use here. It couldn't be raised the way Silicon Valley was used to doing things, right? Yeah, so, when you zoom out and look at our business, right, there's two really big chunks of our business that require kind of deconstruction, right? There's one piece of it, which is we've built this incredible software layer that is best in class and allows us to extract incredible yield from the infrastructure that we built. That belongs in Silicon Valley, right? Silicon Valley likes to invest in things that are going to change the world. They're incredible at it. They allocate capital and they build things that are just moving the world forward from a technological perspective again and again it's one of the great assets of America, right? But they don't really understand, and this is in my mind a reflection of how powerful technology has been over the past 50 years. Like, they've never needed to use their balance sheet because the technology that they're building is so transformational, it's so powerful that they throw off enough cash that they can finance their entire business through equity. It's actually incredibly inefficient, but because it's so world-changing, they're able to do it. It's like sub-optimal from my seat or from the seat of let's say a finance professor who was looking at the problem and they would look at it and say, "Hey, you know, like, it should be some debt here, there should be some debt." And Silicon Valley has moved violently to the equity side of the balance sheet. What we looked at and said, "Okay, that makes sense. It makes sense because we've got to build these software, we've got to invest in our engineers, we've got to do all these things to bring all these people to bear on a problem." That's one part of the business. But then there's a second part of the business which is how do you build physical infrastructure that's capital intensive? And what we understood, right, like I come from a commodities background. I understand finance and I understand structured finance and project finance and that's where a lot of the DNA is for myself and for some of the founders, although there is enormous technical expertise within the company, staggering, overwhelming brilliance within the technical space. But what we understood from the infrastructure side is that the way that you build infrastructure is using the debt markets, right? And the debt markets view the world differently, right? The debt markets, which I always, you know, casually refer to as East Coast Capital, as opposed to West Coast Capital, really has only one rule, which is, "Kami, my goddamn money back," right? And that is a debt profile. And so what we did and the way that we configured, the financing is we said, "What are the attributes of the infrastructure we're trying to build?" And once you have the attributes in hand, you can go ahead and build a structure that will allow you to deliver a product to the debt markets that they can underwrite. Let me explain what I mean by attributes because that might help. One of the debates that's constantly going on is, "Oh, GPUs depreciate and they GPUs become obsolete." And, you know, those are true, right? Compute, everything comes obsolete. It's just a question of how long it takes for it to become obsolete. And what the insight was within the team at CoreWeave as we moved into these GPU financing is that if you enter into a contract with a Microsoft or a meta, or one of the other hyperscalings, or, you know, as long as you have within the four walls of that contract paid back the principal, the interest, the op-ex, and secured a return on capital, you can go ahead and you can take all of the GPUs that you structure in these deals, put them down into an SPV. And go to Wall Street and say, "Look, we're going to borrow money to buy these GPUs, but we have a five-year contract with Microsoft to pay this back at this rate." And we're going to place it in something we call the box, right? And it's a financing silly term. And in the box, we're going to place the collateral, the GPUs. We're going to place the off-take contract with Microsoft. We're going to place the contract with the data center provider and the Power Purchase Agreement. And all of the necessary components to deliver compute for five years. And then what's going to happen is we're going to deliver compute. Money is going to come from Microsoft, but before it comes to us, it's going to go into the box. And then from the box, it goes out in a waterfall. And the first thing it does is it pays for the power. And then it pays for the data center. And then it pays for the principal and the interest. And then after it's paid for everything, it comes back up to us. And so we don't even get to touch the money until it has completely paid off the obligation. And which point it amortizes within the five years. And at the end, we are left with profit. We are left with compute that is still useful, despite what you're hearing out there. And an opportunity to go ahead and build new infrastructure in that data center to set up the next iteration. And we don't, we're not a build that it will come. The way that we work our business is we go to Microsoft and say, hey, we've got a data center. And it's ready to go. And if you want to sign a contract with me, I will go out and I will buy the GPUs and place them in the data center through this structure and deliver to you. And so we're not building speculative infrastructure, right? We think of our risk capital as being allocated or dedicated to the long pole. And that's the physical data center. But when we think about where everybody kind of loses the plot a little bit, they're losing the plot because they think we're going out buying speculative GPUs. We're not. We're buying GPUs that have already been pre-sold to a counterparty so that we can deliver it, pay back our debt, earn a return, and be left with what I have always referred to as the equity slug. And I want to own the equity slug. I want to own those GPUs after five years because I believe that they will have value and they have option value embedded in them, which I think is very valuable. All right. Well, thank you for that impassioned answer. One question, lightning round question for a lightning round answer before we let you go. It was tough for me. That's what I'm saying. Stock was a smash hit in your IPO, which was what just last spring. Kind of cool down a little bit. You still have what about a $50 billion market cap at this point. Why should people either hold the stock now or buy it? Lightning round. I think that I want people to participate in this company that believe in the long-term value that we create. And the stock price on a day-to-day basis is going to bounce around. It's going to do whatever it's going to do. But fundamentally, we have built a business that delivers enormous value. And I think that's a good place to allocate money. It's how I allocate money to other things. I think this is where there is value. On a risk-adjusted basis, there's value. All right. Let's leave it at that. My contractor's CEO of Core Wave. Thank you so much for joining us. Thank you. This is Ed Behrins. I'm Andy Server. We'll catch you next time. [ Music ] The production team for At Barrens is LA Smilodoo, Joseph Lusby, Kinga Roycech, Makarena Karasosa, Rebecca Pizdale. The executive producer is Melissa Hagridi. We'll be back with a new episode next week.