CoreWeave's Tightrope Run

Data is everywhere. When orchestrated properly, it sings. At Morningstar, we analyze and enrich data, making it actionable and powerful for you. Morningstar, where data speaks.

I think that's probably one of the more misunderstood things about our business is people will look at our debt load and say, wow, that's a lot of debt. But in reality, when we are entering CapEx, it is all success-based CapEx.

Hello and welcome to the Barron's StreetWise Podcast. I'm Jack Howe, and the voice you just heard is Brannon McBee. He's the Chief Development Officer at CoreWeave and one of the company's co-founders. CoreWeave is a recent IPO and a big stock market gainer. We'll learn about what the company does, and we'll also hear from Barron's reporter Adam Levine. He says CoreWeave is walking a tightrope. I think he means risky. He could mean circusy. We'll find out next. Listening in is our audio producer, Alexis. Hi, Alexis.

Hey, Jack.

There's no time. We're in an extreme hurry. We've got these two conversations to get to. Nobody needs to hear me prattling on about CoreWeave and setting up what the company does. We'll get to that in the conversation. It's an AI computing company. They lease computing power to others. It's an IPO, came public at $40 earlier this year, was recently about $160. So really an extreme stock market performer. And that's it, right? The rest we're going to hear about from Brannon. He's going to make a case for the company. And then later we'll hear from Adam at Barron's about, you know, some things investors ought to keep in mind. So what do you think, Alexis? Jump right in. I mean, I could do a quick 187-minute history of cloud computing if you want.

Uh, 187 is a really specific number.

Well, I've rehearsed it a few times in the mirror, so, you know.

Well, unfortunately we don't have time for that.

So I think, although it's tempting, we should just get to the conversation. Okay, let's do it.

Don't look so sad.

I'll be okay.

Here's part of my conversation with Brannon McBee, Chief Development Officer and co-founder at CoreWeave. Being that your business is newly public, there might be people out there who still aren't familiar with, um, what you do. Can you give us a, just a quick run through of the business model? Sure.

So we are a cloud platform that is purpose-built for powering AI innovations. What does that mean? That's everything from data centers, physical infrastructure, the cloud platform through which our clients access compute, including networking, storage. Like, it's, it's everything nuts and bolts. Our peer group, who we mostly compete with, their product was designed around a different engineering problem. And that engineering problem was, how do you host websites, store data lakes, and run web apps? And how do you do it at scale, truly global scale? And they've built this huge global platform for solving that engineering problem. But the engineering problem of artificial intelligence is entirely different. And largely it's not translatable to take the solutions that were designed for that problem and apply them to AI. At the end of the day, you're building supercomputers for artificial intelligence versus the infrastructure that was run for hosting websites and storing data lakes. And they built these products around a tolerance of allowing infrastructure to fail. AI workloads, as you know, are infrastructure failure intolerant. It's very bad if any component fails in these massive supercomputers that are being built. And you go from like allowing infrastructure to fail to you can never have it fail. And that's a different engineering solution to come up with that problem. And that's where Qorvig's product really sits, is that we're a purpose-built platform for artificial intelligence workloads, and it's a fundamentally different engineering problem and allows us to do things that our competition is struggling with doing.

I feel super flattered that you use the phrase, as you know, when you're talking about the intolerance of AI infrastructures. I assure you, I don't know, but I appreciate that just the same. Who is, Describe the customer to me. First of all, is it oversimplifying to say that this is kind of a, like a rent-not-own model? Like a customer can go to you and pay as they go for AI computing and services? And what is the type of customer who would prefer to do that rather than build something on their own? What are, what are their needs?

Gosh, so there, there's a lot that's in there. Let's start with who the customer is, and I'm going to be kind of generalized in that side. We can't be too explicit about who our customers are aside from the ones that we've kind of publicly announced. But that's people like OpenAI, like IBM, right? It represents the world's leading AI labs and then your enterprise adopters of AI, like all those enterprises who are coming into AI and want to run their own models, whether it's training or inference on our platform, everything we do, CoreWave owns the infrastructure and then we lease it out to the clients, right? It looks really similar to other kind of like cloud infrastructure models that you've seen like Google, Amazon, Microsoft, et cetera. It's that same concept, but just for AI. Those clients with us, they'll enter these multi-year contracts for the infrastructure. Think like 3 to 5 years in duration or so, and they're saying, We're gonna pay for it whether or not we use it, right? It's called a take-or-pay contract, and that gives them committed, discrete access to that infrastructure for that entire period of time. That's the normal type of relationship that we have with our clients. Something that we're really proud of is we were first to market with the NVL72GB200 deployments this year, which is a wildly complicated, super exciting piece of infrastructure that NVIDIA released. We're really excited to be delivering it to clients and to keep bringing it online this year.

So there are already so many fascinating things happening with AI behind the scenes and a few things in front of the scenes. Like there's some tools that, you know, people can use on their phones or they can use, you know, the ChatGPT and this sort of thing. But when is the moment when you think that the general public, people who do not look at this stuff every day and don't pay much attention, When is the moment when you think that they're going to be blown away by what has happened and where we are and what the capabilities are?

I think the introduction of reasoning models, which really started happening in last June or so with OpenAI's, uh, O-1 model was where I see my peers, my friends, et cetera, having that aha moment. With AI. And that aha moment is when do you start using AI for answers versus using a search tool to just get links, right? And I've mapped, you know, that like fifth button on your iPhone, the action button, I've mapped that to ChatGPT now to where I hit that thing, go in, and that's how I search for things now. And the answers you're getting out of this, I mean, it's wild and it's in-depth and it's thoughtful. And so I think that that is really happening right now. And you have this transition of like replacing search and search is just a massive product globally right now. And to have something that is able to disrupt that and to be an improvement on it is a super powerful thing.

I totally know the little button you mean on the iPhone, and I feel lame because mine is on like mute or unmute, but I'm gonna remap it. This is a great idea.

Just try it, yes.

And I wanna be more like, you know, the people in the know. I'm gonna map mine to ChatGPT. Just tell me about the company and the long-term trajectory on the financials. Yours is a company where you have to spend a lot of money upfront,, right? You have to build this infrastructure so that you can lease it out. And so as I look, free cash flow is negative this year. It's expected to be, you know, negative by a considerable amount. And so I, what comes to my mind, you could tell me if this is way off, but I think an investor might be thinking about Netflix, right? There was a time when we said about Netflix, wait a second, it's burning all this cash on something that is so, you know, fleeting on, on movies and, you know, How's that gonna pay off? And now of course we look at the thing and it's a cash flow colossus. So that's a, that's certainly a happy ending for investors. What is, you know, describe to me as much as you can a, about your sort of long-term thought on how these investments pay off down the road.

Sure. So what do we look at ourselves as like today and in the future? It's the AI hyperscaler. Everything has to kind of be rebuilt from scratch for AI, right? Like there truly is no existing infrastructure from what has been built over the last 15 years that is translatable to these workloads. Like this is a new kind of like industrialization era. The question is like, how quickly can it be built? So the way that we approach that, like recognizing that capital intensity, was to be focused on committed revenue contracts with our clients. So we'll call it the average revenue weighted duration is 4 years, and that's committed revenue on a specific type of infrastructure at specific economics in a take-or-pay form so that we know that no matter what, we are being paid over that 4-year period. And then that allows for us to underwrite the economics of our infrastructure accordingly. We take that revenue and that infrastructure, pair it with debt, and that debt self-amortizes within the contract period. And as the contract is finished, we've paid for all the infrastructure once it is done and are generating a return off of it as well. So I, I think that's probably one of the more misunderstood things about our business is people will look at our debt load and say, wow, that's a lot of debt. But in reality, when we are entering CapEx, it is all success-based CapEx. We're not buying infrastructure and hoping that people come and use it. We're only buying infrastructure when our clients are saying, hey, we want that and we want it for 4 years, which allows us to turn around and finance it. And we just do that on lots of contracts with lots of different clients. It allows for us to de-risk the business from the perspective of having such an immense CapEx exposure, right? Because again, it's all committed revenue that sits behind it and sits behind the debt load associated with it. And that debt load is decreasing during the contract term because it's self-amortizing, or it's paying for itself over that 4 to 5-year period.

Last question that I have for you. You've been generous with your time. Thank you. I know you have AI to build, so I won't keep you longer than I, than I have to, but it's about competition, right? There are some big players out there in cloud computing. You mentioned some Amazon and Google and so forth. And I suppose that they have to be thinking about this new frontier of, of AI computing too. And they would probably like to do similar things that you're doing, what are your competitive advantages? Is it a part of the market that you target? Is it a relationship with a key player, with a relationship with Nvidia? What, what are the most important elements to your competitive, uh, advantages? What's going to keep you safe for investors?

Yeah, so look, the scale of infrastructure we're talking about, it's— there's no way it's going to be built by one company. I think that Core is one of those important companies, and certainly at scale contributing to that build, but the product that we bring to market is differentiated from what our competition has. And it's differentiated because our competition is held back by the technology debt of the product that they've been delivering into the market for the past 15 years. And they're trying to take that existing product and modify it so that it's consumable within AI, but it typically comes with compromises. The best analogy I found for this, it's sort of like walking into— try not to pick names— an auto manufacturer and saying, why can't you produce a Model Y? And they're going to look at their existing fleet, find some car, put a battery in it, and say, well, here's our competition. But we all know that there is an entirely different process, and in product there. And that's the scale of change that needs to take place for our competitors to produce a similar performing product that we have. And our product is just truly purpose-built for AI. And I think that's recognized by our clients, by our suppliers, and by the investor space as well in both the equity and the credit markets.

Thank you, Brandon. Let's take a quick break. We'll come back and hear from my colleague Adam Levine about some additional things investors should think about when looking at CoreWeave.

Data is everywhere, but is it ready for consumption? Morningstar developed the language of global investment data so you have the right ingredients to help you shine. Morningstar.

Where Data Speaks. Welcome back. For another perspective on CoreWeave, I turned to Barron's reporter Adam Levine. Let's pick up with part of that conversation. Adam, you described this company CoreWeave as a tightrope walk. What do you mean by that?

It sounds dangerous. It is dangerous, but if you keep moving forward, you get to the other side. Oh, I see. So what they are is it's 100% pure AI cloud play. They only rent out servers built around NVIDIA hardware. NVIDIA is also— they're an investor, they're a supplier, and they're a customer. They own 7% of CoreWeave. So they're kind of blessed by NVIDIA, which is part of the excitement around them, is that they're kind of NVIDIA's child.

You have to have the NVIDIA chips. That's what the customers want. This company has them. It has a close relationship there. Is it oversimplifying to say that it gets its hands on these chips, it, uh, you know, builds data centers around them, and then it just rents out access to these data centers for AI computing specifically?

Is that it? That's exactly right. But first thing to understand is what NVIDIA basically did was they sold a bunch of GPUs to CoreWeave instead of Microsoft. Forcing Microsoft to rent them from CoreWeave. Microsoft was 72% of CoreWeave's revenue in the first quarter.

GPUs are these chips that we're talking about, these graphics processing units or graphical processing units. Those are the things that, uh, we used to talk about for video games, and now they're at the heart of AI.

That's exactly right. So this took them from $16 million, with an M, in 2022 revenue, to $1.9 billion with a B last year, and analysts are expecting $5 billion in 2025. Sales grew by 420% in the first quarter. And so it's this flywheel.

It's a shocking increase, but it makes me wonder, you make it sound as though Microsoft, maybe like this wasn't their first choice in terms of the arrangement, or is this what they wanted all along? Is Microsoft enjoying this deal? Is, are they a steady and stable customer that could be counted on for years to come, or is it a risk to have so much of your money coming from one customer?

It's a huge risk. We can basically think of that 72% of revenue as Microsoft overflow capacity. That demand for AI computing is so high that Microsoft has to go out the door to rent some of the supply.

Uh, so they do this, they do this same thing on their own. They just can't handle all of their own business. So they send some of it to CoreWeave.

Right. They're, they're one of NVIDIA's biggest customers along with Amazon and Google. And they run very large clouds that compete with CoreWeave. So it's a bit of an incestuous relationship. And it's built on a mountain of debt and capital expenditures.

Nvidia has this lavish free cash flow. CoreWeave is, is well into the negative.

It's burning through cash in order to keep the growth going. It's a flywheel. Debt turns into CapEx, turns into growth.

CapEx is capital expenditure, meaning the stuff they're spending the money on, right?

Which is a substantial amount. They said they're going to do $20 to $23 billion of that. They already have $12 billion in debt at the end of March. They just added $2 billion in senior notes to that, so that takes them up to $14 billion, and they're probably going to have to do at least another $15 billion this year in debt. All that capital expenditure, all that debt shows up in the income statement as depreciation, and interest payments, and that was 72% of their revenue in the quarter, was those two expenses.

There's not another company like this out there, or is there? Like, there are the, there are the, the other tech giants that are doing their own things in-house, their own cloud businesses, but there's not this pure-play AI capacity rental company or leasing company that is this heavy into the game as CoreWeave. So how do we figure out whether this thing is going to succeed or struggle or what? I mean, what's your, what's your best sense?

It's a binary thing. This is back to the tightrope analogy. Either they keep going forward or they fall to one side or the other. If demand for AI computing continues to grow and outstrip supply, they can keep going. They can keep growing at very high rates. It won't be 420%, but it'll be,, you know, 20s, 30s, 40s, something like that. That can keep going and it'll keep them aloft. But if there's any sort of blip in that, the debt, the interest payments, and the depreciation expenses are gonna catch up with them.

Makes me nervous. I mean, tightrope. I'm more of, I think my risk tolerance is more of like a, let's say a balance beam, maybe a foot off the ground, foot and a half I'll go, but I'm not looking for a tightrope walk., but it has since paid off spectacularly well for investors. The thing has multiplied in price. What's that about? Is that about people coming around to the story, people gaining confidence in CoreWeave? Is it about the macro backdrop and the sort of soft stock market that we had?

What, what's happening here? I mean, it's hard to say. It has gone up quite a bit.

Um, what's it at now? Let me pull up a quote lightning fast here. This is the first week of June when you're asking me, and I'm looking at it, it's about $160. And it came public at $40. So you've made— you've got no complaints if you bought it at the offering price.

And look, if you believe that this is going to keep going, you know, all this hardware and the cloud companies and all this investment, all this capital expenditure, if that keeps going forward, then CoreWeave keeps going forward. But if that doesn't, then, you know, they tip over. Now they have mitigated the risk to some extent. They get large customer deposits at the beginning of a contract. They got $4 billion on the balance sheet of deposits at the end of March. They only build the data center when the contract starts, and they have take-or-pay deals, meaning either you take the capacity they're selling you or you pay penalty. And the fact that they're doing more CapEx and more debt is actually good news if you think of it this way. That means they're already seeing the demand because they've already signed the deals. They have a pending deal with OpenAI that's in the early stages. OpenAI was Microsoft-backed, right?

Yes. This is a whole— so how, so how well does that diversify? I mean, on one hand it's customer diversification, but kind of ish, right?

It's more that Microsoft sells computing to OpenAI, and so now does CoreWeave. CoreWeave also sells to Microsoft.

So again, it's a little—

it's all a little incestuous. All right. So again, the debt and the CapEx going up so much this year is actually good news. Investors should be scared if there's— if there's none.

That's when they should get scared.

If growth really slows? If they gave a CapEx guidance that was very low, or even the same as last year, as the same as the year before, that would be bad.

Thank you, Adam and Brannon, and thank all of you for listening. If you have a question you'd like played and answered on the podcast, send it in. It could be in a future episode. Just use the voice memo app on your phone and send it to jack.hough, that's H-O-U-G-H, at barons.com. Alexis Moore is our producer. You can subscribe to the podcast on Apple Podcasts, Spotify, or wherever you listen. If you listen on Apple, you can write us a review.

See you next week. Data is everywhere, but is it ready for consumption? Morningstar developed the language of global investment data so you have the right ingredients to help you shine.

Morningstar, where data speaks.