CoreWeave CEO Mike Intrator on GPU economics, cloud consolidation, and why old chips still monetize

Speaker 1: Yeah. I really enjoyed your perspective.

Speaker 2: But yeah. Have a great rest Thanks of

Speaker 6: for

Speaker 14: the You too.

Speaker 5: Thank you.

Speaker 2: Bye. Let's bring in Mike from Core Weave. He's the CEO. We've been keeping him waiting. And I'm very excited to talk to him about the future of CoreWeave. Mike, how are doing? Welcome to the show.

Speaker 10: Thanks guys. Great to

Speaker 15: be here.

Speaker 2: Thanks so much.

Speaker 1: Nice have you guys.

Speaker 10: You guys have done it with Brian and Brandon. I figured it my dear.

Speaker 2: Yes. Yes. Yes.

Speaker 1: I know. You know what we see? We sometimes see founding teams, executive teams like end up being like, no, these are these are my guys. Like and they don't like to share the TBPN. They don't like to share the TBPN visits. Okay. We like to meet we like to meet the whole team.

Speaker 10: Yeah. We we had the hey, you guys are gonna love this. Go hop on with these guys.

Speaker 2: That's great. I want to talk about some of what Elon and Jensen were talking about the compute math, the monetization of gigawatts. But first, let's get the latest from Core Weave. What's newest in your world? What is top of mind and on the frontier of your business today?

Speaker 10: So look, I'm speaking to you from Fully Connected, our our global conference here where where where we're, you know, sitting down with 5,000 of our clients, developers, companies, enterprise labs that use the cloud that that that Core Web has built to be able to drive their business. I couldn't think of a more exciting place for our team to be.

Speaker 1: Yeah.

Speaker 10: Right? To to to get to to get an opportunity to to be in the in the seat with these folks and get the feedback loop in in in going in in a first person is just, you know, I mean, that's a unique experience for a company. And, you know, in in many ways, Core Weave is really well positioned to take on the information that we're we're garnering from from these clients about, you know, what we've built and how it could be better, and, you know, all those pieces really come together nicely when when when you get a few minutes to put some of your engineers in the room, some of your product people in the room, and some of your clients in the room, and that's that's what we're really here to do. So it's fun. It was supposed to be 2,000 people. We're over 5,000.

Speaker 2: 5,000.

Speaker 10: Wow. Yeah. It's great. It's great. It's like

Speaker 2: And I imagine I imagine the customers are no longer just asking for just a bunch of GPUs. Like we are firmly in the AI agent era. They want CPUs. They want new storage solutions. Like what are they asking for you that you're saying, oh, yeah. We're we're we're we're already working on that. We'll work on that. We'll do that. Give us give us a day. Give us a week. Give us a month, and we'll be there for you as a customer.

Speaker 10: The the truth is is what the the clients are asking us for, what most of the clients, right, are asking us for is the cloud.

Speaker 2: Mhmm.

Speaker 10: Right? Like, they're not asking us for GPUs. Right?

Speaker 1: Yeah.

Speaker 10: They're you know, like, when when we think about our client base, like, 90% of our clients use two or more of the products that we deliver, and that's the software solutions that, you know, all of the different pieces that we stack up. When, you know, if we get down to, like, 78% or something like that, it's like, you know, we're we're at three or more products. So, like, we're not here anymore to say, hey. We can deliver you GPUs. What we're really here to do now is to deliver an AI cloud. A cloud that gives you the full functionality so that you are able to, you know, build the the the to the potential of your company, to the potential of your project, to be able to serve it globally, to do all the things that that we have historically thought of as what the cloud's role in in in the broader ecosystem, but really much more targeted towards the the role for a cloud in the AI era. And so, you know, we we we announced our our our our product here called called Forge, really, which is about bringing together all of that feedback, bringing together all of the tools and software rails, and delivering it to the broader universe. So like not everybody's got, you know, AI experts at the level of these frontier labs. So you've got to be able to build product to shorten the distance between what enterprise clients need to be successful in integrating artificial intelligence into their workflows.

Speaker 16: Mhmm. And

Speaker 10: that's what the AI cloud is gonna do. That's what Forge is built to do, And we're really, really excited about rolling it out because we think it's gonna be a very, very successful effort for for us as we kinda continue to diversify across the enterprise space.

Speaker 2: I have a bunch of questions, Jordan.

Speaker 1: Go for it. Okay.

Speaker 2: Yesterday, we were talking to Sam Altman and he was very excited about the OpenAI compute build out, specifically just the lack of homogeneity across the inference opportunity, everything from big beefy models that are going to run for a long time, the decisions API, the stuff that they're doing with the Jalapeno program. Like it really feels like we're fully stepping out of this era of, okay, just get a bunch of the best chips and that will do everything. There's so much economic value in choosing the right tool for the job. And it feels like that's a big piece of your opportunity is to bring What that capability to does the next year or two look like for you on that side? Obviously, you're building the software solution to harness that. But what other decisions are you making either on the hardware side or with your capital partners? What are the different decision parameters that you're wrestling with right now?

Speaker 10: So so I I love this line of questioning because I've been pillared over this for two years now. Right? Like, oh, you know, you're gonna build the GPUs and they're gonna be obsolete Yeah. Four days after that.

Speaker 2: Yeah. And the depreciation.

Speaker 10: It killed me because Yeah. The depreciation conversation was was it was a self serving conversation by a few players that had a vested interest in in that narrative.

Speaker 2: Sure.

Speaker 10: But the reality is, right, is that, you know, you need the most performing GPUs to build on the frontier. Mhmm. And then once you've built it, there is this incredibly fat tail of other use cases where you can take GPUs and keep them deployed productively for a very, very long time. And so, know, the the the the fact that that Sam is talking about it, the fact that you're talking about it, I I'm gonna take a personal victory loop here because I've been getting clubbed I've been getting clubbed in the head over this for

Speaker 2: for for, you know,

Speaker 10: you know, every time I go on CNBC, they take a whack at me on

Speaker 2: this You deserve a victory lap.

Speaker 10: The the truth is is like, look, compute is going to drive intelligence that is going to permeate every asset of civilization. Right? Like that's that's a fundamental belief that we have. And what we have seen repeatedly over the years is that every time a GPU is no longer capable of driving the frontier for it, it has a slew of other uses that will absorb it. And, you know, I talked about this in my last earnings report, the a one hundreds. That's 2020 architecture. Yeah. Right? Getting that contracted out through 2029. Right? Like, that that is a you know, like, that kind of puts that debate to close. Right? Totally. There are use cases that will absorb this, whether it's batch computing, or medical research, or

Speaker 2: Yeah.

Speaker 10: You know, there's a you know, or things that we haven't even invented yet because there's not enough compute that can drive the new businesses into existence. And so super excited about that. I love that you guys are thinking about that. I love that Sam is thinking about it also. I think it's important.

Speaker 2: Yeah. Yeah. There's just so many different opportunities now. You saw, like, ultra fast 8x the speed, but then also the JEV moment and decisions API and, like, the computer use and then image generated video. There's so many different opportunities there. Jordy, you have a question?

Speaker 1: Our friends over at Semi Analysis have you guys in the platinum tier. Do you are have you been pushing them to introduce a diamond tier?

Speaker 10: Oh, yeah.

Speaker 2: Because so that you and

Speaker 1: the team have something to grind for because I I imagine you don't wanna be stagnant.

Speaker 4: Yeah. Yeah. That is fair enough.

Speaker 10: Fair enough. You know, we we joke. I don't know if you guys have have spent any time with Peter, but we call him premium Peter. Right? Okay.

Speaker 1: The

Speaker 10: joke around here is, and he's amazing where he's the principal architect behind the the so much of the cloud that we've built. Sure. And, know, that would he would work himself to death trying to get it back in tier. I think

Speaker 2: they gotta open it up. Yeah. In the town.

Speaker 1: Well, I just I don't know what to think. Used to be you guys were the only ones there. Yeah. Now there's a player there. Gotta It's

Speaker 2: getting crowded, man. We need a new mountain.

Speaker 1: Going going off of that, when I look at the new Cluster Max three which came out I last week or maybe Monday, losing track of time. There's a bunch of names on there I've never heard of and we spend all of our time talking about these companies. Mhmm. A lot of them, you know, a lot of the other companies that I do know have been have been on the show. My Right now, there's so much, like, excitement and new capital formation around some of these neo clouds. And my question is, like, I have to imagine Like, I I I think the natural assumption is that, half of these companies, like, won't make it independently forever. And so my question is, like, what do you think when do you think the window is gonna come where these these companies that have invested, let's say, between a 100 and a bill you know, billion dollars into building something but aren't gonna quite make it. And then that becomes a sort of opportunity for the more scaled platforms to hoover up talent and maybe some more specialized products and and even down to the individual, like, site level?

Speaker 10: Yeah. It's something we we we we think about here and talk about a good bit. Right? And I I wanna be clear. There there's a lot of different paths to consolidation across the cloud industry. And, you know, one of the things is is, you know, we we don't really think of ourselves as a neo cloud anymore. We are the AI cloud. Right? Like, you know, the concept of being a new cloud is great for a while, but the truth of the matter is is our job is to deliver the functionality that the world needs, and that's the AI cloud. Right? Yeah. But you also get to a world where you get consolidation. It could be because there's an air pocket in compute demand. Right? And that's one of the things where, you know, it would be destabilizing. But there's also another scenario where, you know, power becomes more dear and the selection process for who people are going to sell to tightens, right? And they're just like, why am I going to take a shot with a third tier new provider when I can go ahead and get, you know, the same terms with a reputable well established Yeah.

Speaker 1: Just it just comes down to like Capital capacity. If like if I'm a if I'm a company I'm planning to scale my demand for compute really quickly, it doesn't make sense to sign with the third tier provider simply because I'm gonna run out of, you know, capacity or or Yeah.

Speaker 10: And and and it also causes drag on the balance sheet of the off take companies. Right? If you're a a buyer of compute and you've got 47 different, you know, third tier clouds, that can lead to drag on your balance sheet, which is not great. Right? Like, so you wanna have a efficient allocation of capital, an efficient allocation of the capacity of your balance sheet, and you do that by really working with entities that you have a high degree of confidence are gonna be able to deliver performant infrastructure. Yeah. You know, I I I think that, you know, there's an inevitability. Right? Like it's a capital intensive business at the end of the day, and capital intensive businesses end towards a consolidation. Whether the number is going to be 10 or seven, I don't know. When it's going to happen, I don't know. The the the the the priority for us is to build a resilient company, and you build a resilient company by working with high quality clients, by delivering a high quality product, by decommoditizing the compute that you're delivering, by having incredible software rails so that people are able to use it, by moving across the enterprise space. All of these things have been priorities for us, and we have done a really, really good job of staying focused on, you know, as a team on making sure that we drive the the parts of the business that will ensure that Core Weave is resilient through whatever cycles come. Mhmm. And we're pretty comfortable about that.

Speaker 2: Amazing. I have one last question. We'll let you go. For for q two, the rough dollars per watt was around $77,000,000,000 per gigawatt. It feels like that's accelerating. It feels like you might be tracking to closer to $40,000,000,000 per gigawatt. Elon and Jensen were doing some rough math around 60,000,000,070 billion Is this something that you have visibility and optimism around continuing to climb? Or is there more focus on getting to some sort of steady state where you're monetizing the energy that you have and then just scaling out horizontally?

Speaker 10: Yeah. I I I look at the problem and opportunity a little bit differently. Right? I look at it from a time horizon perspective. Mhmm. Right? And so I believe that there is going to be a market for lower cost compute that lives outside of the build cycle of infrastructure. So if you think about it, and let's say, just hypothetically, takes us two years to build a a data center. If I can sell capacity in a time frame beyond that two years, I can replace it. Right? Once I'm inside that, that's where Elon Musk and and Jensen are talking. Right? Because then it becomes truly a a supply demand dynamic where, you know, people are going to compete for that compute. And those are two separate markets, and I believe that that over time, they're going to begin to behave differently where people who can make the long term planning are going to be able to drive economies that are associated with that because it's going to allow folks on the infrastructure side, on the cloud side, to be able to plan better. And so there is it's a it's a it's a really fascinating piece of the market when you think about it in terms of time, not just in terms of dollars.

Speaker 2: That's very interesting. Well, thank you so much for coming on and breaking it down for us. Congratulations.

Speaker 1: Say hi to all 5,000 people over here. Hopefully, we can join next year

Speaker 2: in person.

Speaker 10: Love to have you guys here, man. Let's do it from the home place.