What does CoreWeave’s IPO say about the market?

0:05Pushkin. It is a difficult moment to set sail on the initial public offering market. There was a pretty serious correction in U.S. stocks earlier this year that hit tech stocks the hardest of all. None of this is discouraging the good ship CoreWeave, a tech company that in theory is meant to start trading tomorrow, Friday. This is Unhedged, the markets and finance podcast from the Financial Times and Pushkin. I am joined today in the studio by Lex editor, John Foley. Hi, John. Hi, Rob. It's a delight. And down the line from his linen closet in London, we are joined by swivel-eyed conspiracy theorist, Rob Smith.

1:03Thanks for that, Rob. My job title for listeners is corporate finance editor of the Financial Times. But yeah, you're not a conspiracy theorist. Six of one, half a dozen of the other, I say. You're not a conspiracy theorist if they're all out to get you. That's the old phrase case. John, will you start us off by just explaining to our listeners what this company does? What CoreWeave does? Yes. CoreWeave. I thought it was a kind of men's baldness treatment. Or like some kind of textile fashion brand sleeveless gilets. But it's not, is it? Investment bankers. No. What it is is effectively a giant heap of NVIDIA chips.

1:45So you know these chips that everyone is using for AI to train their algorithms and to spit out answers to questions. Yeah, yeah, draw pictures. Inference, as it's known. So that all relies on these NVIDIA chips. These chips are quite hard to come by. They're quite expensive. CoreWeave basically takes a bunch of NVIDIA chips directly from NVIDIA, packages them up in data centers and rents that capacity to companies that want to do AI stuff. Primarily, as I'm sure we'll discuss, Microsoft OpenAI. And that's what it is. It's an AI capacity rental company. And if I'm right, and Rob Smith, I think you might have written a little bit about this.

2:24This used to be a cryptocurrency miner. Is that correct? Is that where this came from yeah so the founders of this business are a bunch of traders so they're not the typical sort of tech founders they're not coders those kind of people i think they worked at a gas futures hedge fund and they originally set up call weave as a bitcoin miner i think it was called atlantic crypto corporation at the beginning and then it kept pivoting so it pivoted to i think it was like 3d video rendering in the first crypto one of the first crypto crashes and then it's like adding oh we do machine learning and then because it had a bunch of these nvidia gpus which became like gold dust right everyone was trying to get their hands on it it became a gpu cloud an ai hyperscaler it started renting out this stash of gpus to these sort of data hungry big tech companies to run these LLMs and everything else that has changed the world in the past few years?

3:31Well, it's interesting that the business had its start with a bunch of commodity traders, because it seems to me that the first question I would ask about a business like this is whether it is a commodity business. In other words, you buy a bunch of stuff from someone else and then you rent that stuff to other people. That sounds on the face of it to me like a kind of low margin business where you are eking out a margin because of differences in the timing of needs of different parties, different finance costs between different parties. In other words, it doesn't sound terribly low barrier to entry.

4:22Am I barking up the wrong tree? That's definitely a part of it. They're not literally just shipping raw chips out to their rental customers, though. So there is a bit of value add around that, which is creating data centers and supplying them with power and all that kind of stuff. And there's a software layer as well. Right. Cloud companies like to tell you that their value add is through software, which is in some cases true. But fundamentally, yeah, this started with a commodity-like trade, which is GPUs, hard to get. We got them. Lots of people want them. We can join the dots between A and B.

4:55Do we have a sense of the margins or the returns this company is earning at this point? I can see from the S1, the IPO filing document, that revenue is growing extremely fast. Yeah, I mean, a big part of the debate around CoreWeave has been on the metrics they present. And I understand, I'm reading this from New Street Research. CoreWeave presents a 75 % gross margin, but apparently they strip infrastructure depreciation outside of that, which is huge. So New Street Research sort of put them on blast and said this was misleading. and they backed out the actual margin at more like 45%. And of course, the reason I should say just for our normal listeners at home, the reason depreciation is so important in an asset heavy business is because the depreciation line is where the expense of buying all this stuff shows up.

5:57Depreciation is you have a big one-time asset cost and you spread the cost of buying it over time. and the depreciation line is where that cost is spread to, as it were. So in a business that is buying and renting out assets, you would, at least on first blush, you'd want to include depreciation in your estimate of the company's profitability. So I am not here to be the advocate for CoreWeave. I think there are some very odd things about this IPO. One thing that you can say for them, because they do obviously report their depreciation, not in their cost of revenue, but they report it in their, you know, it's there in the operating expenses, which is the normal way to report it.

6:38Coreweave is actually profitable on an operating profit basis, even when you take out the stock-based compensation that most companies ignore when they tell you about their adjusted operating profit. Good for them. In fairness to Coreweave, unlike most tech IPOs and most tech companies that we look at, This one is actually profitable on an operating basis, right? Now, the challenge, as I'm sure we will also discuss, is that there is a financing element to Corweave's business. It pays a lot of interest. And if you start thinking about that too, then profit kind of quickly dwindles. But again, in an asset-heavy business like this, isn't your cost of finance an absolutely crucial part of the profitability of your business?

7:23You can't not pay your interest, that's for sure. If you think of this as a financial business, as a business that has a cost of capital on the one side and a return on capital on the other, you don't want to leave out the finance. Right. Cost, that cost of capital. It's a big deal. Rob, what can you tell us about the debt structure of this company? It's gargantuan and labyrinthine. So that was how I was kind of brought into this story. So my colleague, Tabby Kinder, who's our San Francisco sort of beat reporter, has been doing amazing work on this. And Tabby's like no slouch when it comes to understanding complex financial structures.

8:03But she called me, you know, a week or so ago and was like, Rob, can you just take a look at this? Because it's hurting my head. And then I took a look at it and it hurt my head. And then we read some articles which were very factual but caused a lot of grief. And the reason we caused a lot of grief is because we did some adding up of numbers, which is something companies don't like journalists doing. No, they certainly don't. Not in my experience. No. And we basically found out that CoreWeave has to repay seven and a half billion bucks of debt obligations by the end of next year. So kind of over the next two years, which is that's a monumental number.

8:45Did CoreWeave or its financers come back after you wrote that it had this huge mountain of debt to pay back in the relatively near term? No, there was no on the record statements or denials. And I would have been surprised if there were because it was just simple addition of numbers. Okay. I want to come back to the kind of financial structure of this company, but I want to deal with the revenue question first. If I'm reading the documents from the company and what I've read in the FT correctly, this company borrows at a fairly high interest rate, double digit interest rate. It buys a lot of assets.

9:30That debt has to be serviced. what is the revenue picture of this company look like its revenue trajectory i know it made tremendous amount more money in 2024 than it did in 2023 but what do we kind of know about the earnings power of this business in the context of having to pay back this mountain of debt it's a great question i mean they have a lot they have huge obligations future interest payments as you say and also capex they need to buy more chips what you can say for the company is that its revenue is growing very quickly. It's something like 700 % in a year. They're trying to mitigate that risk by signing customers up for longish term contracts.

10:10Long is a very subjective term here, but they're signing customers up for four or five-year contracts where Microsoft and now OpenAI pledge to pay a certain amount over a certain number of years. So there is cash flow coming in from those contracts and it's not small. They have like$15 billion already signed up at the end of last year, they've just added 12 billion from OpenAI. So provided those contracts are honored, then there is money coming in. Well, let's first talk about the companies they're contracting with, and then we'll talk about the durability of the contracts, say. One issue that has been raised by the FT and a lot of other places is that this company is very dependent on a small number of customers.

10:57Is that a deep concern? I mean, if I was going to be dependent on one company in the world, I'd want it to be Microsoft probably. Is this something that gives you pause? So that's the question. It's who is the customer? Being dependent on a customer can be good or bad depending on who that customer is. So what we've seen in a lot of data center related financings, deals with hedge funds and private equity firms to build data centers is that some customers of those data centers are treated almost like sovereign credit. Microsoft is one. Microsoft is a AAA rated credit. It's unlikely that Microsoft won't be around to pay its dues.

11:34So in that sense, Microsoft is not a huge concern when it comes to honoring the contracts it's already signed. Not a huge concern. It's not no concern. It's not a huge concern when it comes to honoring the contracts it's already signed. Does that mean Microsoft will want new contracts that come after 2030, who knows? That depends on what Microsoft's own AI goals are at that time. And don't forget that if you're buying shares in a company today, you kind of hope that it's got a plan for after 2030. Five years goes by very quickly. This is not a company that is kind of used once and destroyed by 2030.

12:02So in terms of the counterparty risk for Microsoft, that's relatively low. Now that things get a bit complicated, like if you say we want you to diversify your customer base, Colweave says, great, here's a new contract with OpenAI. Is OpenAI like a sovereign credit? I would say no. Would you rely on OpenAI to service its debts in the way you would have Microsoft? Almost certainly not. So by diversifying, they've also diversified their counterparty risk. I'm glad you mentioned Microsoft's sterling credit rating, because this is one of the biggest questions I had about this company. And I put this question to both of you.

12:40Microsoft can borrow money at a price that is almost like the price the US government pays. I mean, they pay nothing, right? A spread of roughly nothing over treasuries. But they are renting assets from a company that has to borrow at a very high rate to finance those assets. So it seems that by renting from CoreWeave, Microsoft has added a huge amount of unnecessary financing costs to the whole ecosystem. In other words, why don't they just buy the darn GPUs, finance them at 4 % and use them when it needs them? Rob Smith, any thoughts? Yeah, well, I think you've got, again, it goes back to like thinking about this in terms of capital structure and debt it's because it's off balance sheet for them right for microsoft or for yeah core weaves like enough balance cheat structure for them right where i mean they're 60 of this company's revenues right like they're kind of the gorilla in the room behind this and by doing this they can keep some of this capex over there and have some optionality around it and it not show up on its balance sheet Right.

13:58And they're, you know, they're smart people. They're always tinkering with things. They obviously see some value in doing this and having not just themselves building everything, buying everything, just putting some stuff over there and doing it over there and renting it. Optionality is just the point. I'm sorry to interrupt, but optionality is just the point. If they have optionality, right, that raises again, the question of the sustainability of these contractual relationships. The point of renting something is that if you don't need it anymore, you can stop paying the rent. Right. Isn't that a right point?

14:36So this is a great point, though, because Microsoft, it does indeed keep this stuff off Microsoft's balance sheet. Why would Microsoft, which is an absolutely enormous company, I'm now currently scrolling through its 10K trying to get the number of how big its total assets are, and I'm not going to find it in time. But why would it care whether this stuff is on its balance sheet or not? It's impossible to imagine that investors or creditors of Microsoft would be concerned by this tiny addition to its balance sheet. So if you don't like the off balance sheet theory, John, it's incumbent upon you to give us another theory.

15:08Well, so here's a theory. Okay. Here's a theory, and it is only a theory. Yes. Based on some observations of how the market is working at the moment. So NVIDIA chips are in high demand, rightly or wrongly. People think that there's no such thing as too many of them. It's hard to get as much supply as you want. So NVIDIA unusually has some say over who gets its chips. Now, NVIDIA would argue strenuously that it does not play customers off against one another. And we've heard no evidence from customers that they feel like they get played off one against the other. If I were NVIDIA, let's put it that way.

15:46If I were NVIDIA, I might say to myself, I don't really want to be beholden to one customer. Why should Microsoft get more chips when I can instead encourage them to rent chips from this third party that diversifies my customer base? But let me just ask you one question. I thought everybody was buying NVIDIA's chips. Yeah. Meta, Google, Amazon, everybody. Why is NVIDIA's revenue line so dominated by Microsoft and not any of the other so-called hypersuppers? They're all big. They're all huge customers, of course. You say Microsoft is just a very big customer. Yeah. And Amazon, Google, or Alphabet are also trying to make their own chips.

16:25They're also trying to create alternatives to NVIDIA at the margins. But Microsoft is just a very big company. It's got a very big cloud business. It's a big customer. There's no suggestion that NVIDIA, as in I'm not saying that this is what happened, but if I were NVIDIA, I might try and seek ways to diversify my own customer base. Because as we're talking about with CoreWeave, reliance on one customer is a risky thing. Just this morning, we've had two pieces of news about the CoreWeave IPO in the pages of the Financial Times. First, both the size and the valuation of the IPO have been cut. And second, NVIDIA is putting several tens or even hundreds of millions of dollars of its own money into the IPO.

17:09Gentlemen, what do we think about these two facts? oh well i mean it shows that the demand is not there from the market and the nvidia which is as well as being the supplier of chips to core weave and having a current six percent equity stake is also believed to be core weave's second biggest customer by the way if we're talking about complexity you know this very much feels like nvidia having to come in to prop up an ipo which can't fail. This is like this IPO has to happen. And NVIDIA for what is like a rounding error in its quarterly cashflow is putting some of that in to help. Now you have to explain something to me, Rob, because I am a simple man.

17:59NVIDIA is a company that makes GPUs. It is also the customer of a company that buys its own GPUs and rents them to other people. Yeah. Why would NVIDIA need to rent its own chips back from CoreWeave? I do not know. And we say believe to be. We don't know. Well, so to unpack that, we know that NVIDIA is a customer. Basically, CoreWeave gave some disclosure about its customers. It disclosed who its biggest customer was, and that was clear it was Microsoft because it tallied with the size of the contract. It's second biggest customer. You can kind of work out from the maths that it must be NVIDIA, given like some disclosure around NVIDIA's contracts, basically.

18:53So we know NVIDIA's a customer, and it is strongly implied that they're the second biggest customer. And yes, it breaks my brain too. I do not understand it. I'm not going to lie to your listeners and pretend I understand it. Final question. It seems to me this business, despite the kind of highways and byways we've taken listeners down in this podcast, the basic structure of this business is actually quite simple. And what I mean by that is this. It is a leveraged play on demand for GPU processing capacity. It acquires financing at a cost, buys that capacity and rents it to other people. And the success or failure of the company depends on how strong demand is for that capacity and how price for that capacity holds up over time.

19:51Do we know anything about how that demand is going? What you've just said is exactly the story really with CoreWeave, right? There are some very odd quirks to it, but fundamentally, CoreWeave is good or bad based on will people pay to use GPUs 5, 10, 15 years from now at the price that makes a profit for CoreWeave. Do we know the answer? Of course we don't know the answer. That question is holding up the valuations of all kinds of companies right now, including NVIDIA. Rob, final word? yeah well i mean i think john's like hit upon the uncertainty in sort of like the earnings outlook shall we say but i think added to that you just have the capital structure when you run a lot of leverage you have less room for error and uncertainty so i think you layer all of that on top of what john just said and it makes that uncertainty like even scarier potentially even scarier.

20:51If I can say, Rob, also, the thing about IPOs, right? The thing about IPOs, the IPOs are risky because they're a company with no track record as a listed company coming to market. The more weirdnesses you layer on top of that, the more you should be worried. Like, IPOs are inherently a risky thing. So, every single thing that we tick through, every oddment in their filing is a reason to be more cautious of this company. Like, why would you go to market when And not only are you untested and new, but you're also stocked to the rafters with these weird quirks that it's very difficult to price in.

21:30Listeners, we'll be back after a short break.

21:46Listeners, welcome back. This is long and short. which is the portion of the show where we go long things we like and we go short things that we do not like. What I am short and I do not like today is auto tariffs, as announced by the president yesterday. I do not think this is going to go the way the president thinks it is going to go and that there is going to be collateral damage along the way, and we are going to find out unpleasant facts about auto supply chains we didn't know before. Rob Smith, do you have a long or a short for us? So I'm going short Blackstone. Blackstone is Corweave's biggest lender.

22:29And as we've reported this week, we were the first to spot this. Corweave defaulted, it was a technical default, but still a default, on a$7.6 billion facility, a loan facility. Now, why that matters to Blackstone, this was Blackstone's largest ever single loan commitment in its history. so Blackstone committed the majority of that and it defaulted you know a couple of quarters out due to breaching terms this is one of the largest private credit loans ever and this might give you an insight into some of the underwriting standards that are going on at the supposed Rolls-Royce of the private capital market and I'm not doing this Blackstone specific but it's a window into what is happening into this market, which is often quite opaque.

23:17John, do you have a long or a short? Make it fast, because Rob Smith just did a whole other story for us. There's a long short. It's very long winded short. This is a short long. Okay, a short long. Give it to us. So your point about car tariffs, I would go long Ferrari, because although Ferrari is a huge exporter of cars to the US, and its car's very expensive, its buyers, I don't think, care about the price going up. And I would argue that maybe Ferrari is a Veblen good, such that when the price goes up, people want it even more. Listeners, we will be back in your feed next week. Until then, stay sharp.

23:57Unhedged is produced by Jake Harper and edited by Bryant Erstat. Our executive producer is Jacob Goldstein. We had additional help from Topher Forhez. Cheryl Brumley is the FT's global head audio. Special thanks to Laura Clark, Alistair Mackey, Greta Cohn, and Natalie Sadler. FT Premium subscribers can get the Unhedged newsletter for free. A 30-day free trial is available to everyone else. Just go to ft.com slash unhedged offer. I'm Rob Armstrong. Thanks for listening.

24:40holistic contact