The Shaky Economics Behind CoreWeave (ft. Gil Luria)

Thanks for having me.

They still don't make any money. Let's start with that. Okay. We're not sure that this category is even a worthwhile category, right? The Neocloud category is the marginal provider of compute. And we talk about tides, right? When the tide is rising, guess which boats get lifted highest? The smallest boats, the ones that have the most leverage to an incremental buyer of compute. That's CoreWeave. So of course right now when there's an unbelievable shortage of AI compute, there's an enormous gap between demand for AI compute and supply of AI. These are the companies that are going to benefit the most. That's why they're doing so well right now when there's such a shortage of compute. The longer question is, are these worthwhile businesses? Will they generate the types of economic returns that can justify their cost of capital? The jury is out on those things. And I believe that right now the evidence is that they will not over time be able to even return the rate of capital that they borrow at, that they raise at. Certainly not above that.

They don't. And let's just go back to finance 101, right? What's the job of a business? The job of a business is to generate return on invested capital that's higher than their cost of capital. So if I'm investing in something, if I'm borrowing to invest in something, the return I get needs to be higher than the cost of borrowing. Otherwise, I am not a worthwhile endeavor. I should not exist as a business. Let's talk. CoreWeave is really, it's so vivid that it's probably an even better example than Nebius. Nebius is actually a little bit more of a nuanced story. But let's talk about CoreWeave. CoreWeave's cost of borrowing is now 9%. To get capital, CoreWeave needs to raise it at 9%. Their return on capital right now as of last quarter is less than 1%. By the fourth quarter, they're saying their margins are going to expand dramatically by the fourth quarter to what they're saying is mid-teens margins. But if you take that and you look at what that means for return, that's still around 4% return. So they're borrowing at 9% to generate a 4% return. Right? Your viewers and listeners are pretty sophisticated. I don't think any of them would take a margin loan at 9% to buy treasuries.

That's investing in CoreWeave is taking a 9% margin loan to invest in treasuries. That at high risk treasuries at that doesn't make any sense. Now the argument that CoreWeave shareholders and Nebius shareholders make is, oh, don't worry, we'll grow into it. And I think that's possible again if the demand for AI compute continues to outstrip supply for the next five years maybe. But it's not that these are small businesses, right? These are already over $10 billion. Nebius will be $10 billion next year. CoreWeave is already at a $10 billion run rate. Right? If you can't generate good returns when you're $10 billion and managing dozens of data centers, why should I believe you that you'll do it when you're bigger than that? And that's the essence of the problem. Now, does it matter right now? No. Because these companies are showing huge upside and they're selling and the price for compute at the spot market is probably four times what it was at the beginning of the year. And this is a good place to be. But, you know, you always have to remember that there's another side to this kind of spike.

They're very savvy operators, right? So if you're going to bet on somebody in this space, you want to bet on the smartest guy in the room, and that's Arkady, the CEO, right? There's a guy who built Yandex to build Russia's Google in Russia. Gosh, that's a high degree of difficulty. And then, you know, he left Russia because of the Russia-Ukraine stuff and he had to delist. And when he relisted, it's his Nebius. It's the same super savvy guy. So, yes, it's a tough business, the Neocloud business. The burden of proof is on him. But he's doing super clever stuff like, you know what, I can provide capacity to my long-term customers, but the price for long-term contracts is actually quite low compared to what I can get in the open market, in the spot market, in the short-term contract market. So, let me go around and just gather some bits and pieces of supply, sell it in the short-term market, and I can make a killing doing that. This is the kind of savvy operator that you want to be in business with. Now, again, I'm not sure that the business is long-term a winner, but if you're going to be in this business, be with the person who knows what they're doing.

So many concerns, but let's just start with, you know, Jensen Huang is in charge, right? Jensen Huang is the puppet master, right? Jensen Huang does not see the world like other CEOs do, which is, okay, how do I improve my research and development and sales and marketing to sell more chips this year. Jensen Huang sees the world as something that he can construct in the manner that will favor Nvidia. That's a very different approach to the world. And the reason he can do that is that he has $60 billion on his balance sheet and in the next 12 months he will generate more than $200 billion. Now he uses some of it to buy back stock. That's very shareholder friendly. But then he has a lot left over. What can I do with it? I'm not going to be allowed to buy any other chip company. Regulators aren't going to allow that. So what am I going to do with that cash? Let me use that cash to build a world with a lot more AI demand, with a lot of investment in generating AI supply so I can continue to perpetuate this wave, this buildout. And when that happens, I'm going to sell a lot more chips. And that's what Jensen Huang does. Now, as part of doing that, he is funding his customers. That's what we're referring to as circular financing. That is not healthy.

I want to full stop, not healthy. When you are financing your customers and they're buying your product, I want to both, we can hold both thoughts in our head at the same time. That is not healthy behavior. That is not real demand. We can hold that thought with the second one which is Jensen Huang has been and continues to be successful at creating a world where there's a lot more AI demand and his ecosystem is building out the ability to generate the supply to meet that demand that will at the end end up with a lot more Nvidia GPUs. We can hold both thoughts at the same time. We can be worried about circular financing, worried about leverage, but at the same time understand that Jensen Huang is busy building that world where we're all using AI a lot more and he's willing to invest in anybody that can help him accomplish that. And then in terms of what this particular deal is, you were spot on. He sees what we just talked about earlier, which is CoreWeave now needs to borrow at 9%. They were already down to seven and that's now gone up to 9%. That's concerning to him because he understands that whole conversation we had about the Neocloud business and he says, 'Well, what can I do to reduce their cost of borrowing? I want to bring it back down to 7%.' Because I believe long-term they will be able to generate returns better than that. But if I don't lower their cost of capital, they'll never get there. So, how about I step in again? I'm going to generate $200 billion of cash flow in the next 12 months. And so I have the resources to step in and backstop these loans, make the lenders comfortable that they can lend not at junk rates but at investment grade rates. And therefore I can continue to make this ecosystem healthy, build out the supply, more supply, better models, more models delivered to consumers and businesses, businesses using models more, thus creating more demand, thus creating more supply. And that way I sell more chips.

Yeah, leverage is bad, right? Leverage is bad. We should not be building out this highly speculative infrastructure based on leverage. Amazon, Microsoft and Google have plenty of cash. They generate plenty of cash flow. They can choose to make speculative investments in that. When we start introducing substantial amounts of debt to build what is a speculative asset, that is not a good thing. Again, we can hold both thoughts at the same time. AI is good. It's going to create a lot of productivity, generate a lot of economic growth. Building it on debt with using leverage, especially if we're not pricing that debt properly, is a bad idea. It's just not a good idea. And we have to be careful about that. We can fund some of the build out with debt, but when we start funding most of it with debt, which is the path we're headed, that's a bad idea. This is a speculative asset. To your point, how quickly the assets depreciate is of no concern here. It's that they depreciate. Whether it's three years or nine years, they depreciate. And borrowing against that when we don't know if it's three years or nine years is not a good idea. We should not be doing that at the scale we're starting to do. We haven't done it yet. We've only lent a small portion of the capex to date, but we're on a path to inflating this thing by lending too much money at too low of a rate for what is inherently a speculative investment, right? That's where blowups happen, right? Leverage and too much debt. And we saw this in previous cycles.