AI Bubble Special Report: Debt, Depreciation, and Losses — With Gil Luria

Thanks for having me, Alex.

There's a lot to unpack there. And the framework that I do it with is to say both things are true. So AI is the most revolutionary technology that we've had in a really long time, whether it's back to the internet or back to the industrial revolution. We'll only know in retrospect, but clearly the tools are very powerful and are getting better. All you need to know in order to realize that is just use them. You may ask ChatGPT to do things for you that are hard, that you would ask other people to do, whether it's summarizing, writing, giving you advice, and you see that not only is it incredibly capable, but it's much better than it was a year ago and it's much better than the year before that. So yes, there is insatiable demand for this product. That is true. There's a lot of healthy behavior around that capability, and the healthy behavior are reasonable, thoughtful business leaders like the ones at Microsoft, Amazon and Google that are making sound investments in growing the capacity to deliver AI. And the reason they can make sound investments is that they have all the customers. They have all the business customers, and by extension of their relationships with OpenAI and Anthropic, they have all the consumer relationships as well. And so when they make investments, they're using cash on their balance sheet. They have tremendous cash flow to back it up. They understand that it's a risky investment and they balance it out. So all of that is true. At the same time, we are seeing behaviors that are unhealthy. And that's where those of us that have lived through financial bubbles or technology bubbles are recognizing patterns that we've seen in the past and are saying, "Hold on, this is unhealthy behavior." And there are companies that are exercising unhealthy behavior. And that's what we're trying to call out. And that's why there is a good, reasonable debate here, is what's healthy and what's unhealthy.

Palantir is the best company in the world right now. I'm not going to even say the best software company. They're the best company in the world right now. Because what they are able to do is go into a company and ask them, "What's your biggest need right now? What is it that you think AI can do for you?" And then do it soup to nuts. And that's why you're seeing extraordinary growth rates there, and they're being incredibly successful both there and on the government side, where they're doing similar things in just a more clandestine way. Then there's companies like CoreWeave, which is the poster child for the bad behavior that I'm talking about. We're talking about a startup that is borrowing money to build data centers for another startup. They're both losing tremendous amounts of cash, and yet they're somehow being able to raise this debt capital in order to fund this buildout again, without having the customers or the visibility into those investments paying off. So there's a whole range of behaviors between healthy and unhealthy, and we just need to sort that out so we don't make the mistakes of the past. And we can delve into why I think debt is an unhealthy way to invest in data centers.

So, we have to go back to finance 101, right? There's certain things we finance through equity, through ownership, and there are certain things we finance through debt, through an obligation to pay down interest over time. And as a society, for the longest time, we've had those two pieces in their right place, right? So debt is when I have a predictable cash flow and an asset, or an asset that can back that loan, and then it makes sense for me to exchange capital now for future cash flows to the lender. So again, the conditions are an asset that is longstanding that can back the loan, and/or predictable cash flows to support the loan payments. Right? That's why we have a mortgage. A mortgage is an example of both. All right, a mortgage is — wait a second. If I stop paying my mortgage payments, the bank owns the house. And since they only lend me 80% of the value of the house, even if the value of the house goes down a little bit, they'll be fine. And they have access to my income, which is relatively predictable even on Wall Street. And so they know that I'll pay my mortgage payment. That's a loan that should be there, right? We use equity for investing in more speculative things, for when we want to grow and we want to own that growth, but we're not sure about what the cash flow is going to be. That's how a normal economy functions. When you start confusing the two, you get yourself in trouble. And that's what, to your point, Oracle is doing. They're saying, "Well, I have this startup that's promising me $300 billion of revenue at a high margin over the next five years. So I'm going to go borrow money to build out the infrastructure in order to deliver that." And what Oracle has been exposed as is, hold on, OpenAI promised me 300 billion. They also promised Microsoft 200 billion, Amazon 38 billion, CoreWeave 25 billion. In total, 1.44 trillion. And who's this company that just promised all that? A company that at best will have $15 billion of revenue this year and will be losing more than that. So we'll be losing probably more than $20 billion this year. So are they in a position for me to borrow money because I have certainty around those cash flows? No. That's bad behavior. And that's what we're talking about here, is if you're borrowing money to make a speculative investment based on a speculative customer, that's bad behavior. And frankly, that's what's dragged the market down over the last few days, is the realization that this bad behavior is happening. And nobody wants a piece of that.

The answer to that is it depends on what magnitude we're talking about. Again, lessons of previous cycles and especially the financial cycle. If we have tens of billions of dollars of debt into an asset that stops being a productive asset, then if there's a problem, it's the people that issued that debt or own that debt that lose money. And the people that own stock in the companies that made those loans, Oracle, CoreWeave, etc., they'll lose if it's tens of billions of dollars. Some financial firms will lose, and mostly the owners of that debt and that equity will lose. The problem starts happening when you get into hundreds of billions of dollars of debt, which is where we were headed, at least as of a couple of weeks ago. Again, OpenAI, this startup, great startup, great product, a startup committed $1.44 trillion to all these entities. So those entities as well as OpenAI could go out and raise debt capital, which means they and their customers were seeking hundreds of billions of dollars of debt. If we are here two years from now and there's hundreds of billions of dollars of debt, and the demand for AI stabilizes, or we've built enough data centers to support the demand we have at that point in time, and the price for leasing pieces of AI, for renting access to GPUs, goes down, all of those assets then can't pay enough to pay the interest expense on that debt. All of that debt defaults at once. Now, we're talking about systemic risk. That's what folks are warning about right now, is it's okay if some financial investors lose tens of billions of dollars here and there. If we have hundreds of billions of dollars of debt into what is really just one product with one price, and that price goes down and all those assets become worthless, now we're going to drag the entire economy down. And again, we're all saying this from experience. Everybody should rewatch The Big Short, and not just to see Christian Bale and Margot Robbie and Selena Gomez. It's a great movie that talks about how it's a little problem until everybody does it, and then it's a big problem that affects everybody, right?

Well, the short answer is the largest institutions in the land. US Bank, JPMorgan, Mitsubishi Bank, those are the companies lending to CoreWeave. And again, the math they're doing, we believe, isn't the right math. Let me dig into that, what I mean by that, a little bit. Again, this is a speculative asset. Just because we're all using it and excited about it — which we laid out at the beginning, we are, and it is exciting, and we need a lot more compute — it's still a speculative asset in the sense that we don't know how much of it we're really going to need in two to five years because we don't have experience doing that. This is brand new. We don't know how much a GPU is going to rent in five years. And so when we get the revenue projections from OpenAI that they're going to make like, you know, a hundred billion dollars a year — I guess I'm exaggerating a bit — you can't trust them because you just don't — how do they know? Exactly. So, one, AI may turn out as well as we expect, but it may not. And two, OpenAI is not in a vacuum. They're competing. Part of the reason they're overpromising and creating this too-big-to-fail and fake-it-till-you-make-it and getting everybody else to have skin in the game — part of the reason they're doing that is because they know they're competing with Meta and with Google and with Elon, people that have a lot more resources than they do. So for them to say, "Oh, we're going to have a hundred billion dollars of revenue by 2027," which Sam Altman just did, is completely disingenuous. He has no idea. He's competing against much bigger, more powerful companies that have technology that's at least as good as his. So lending money based on that is dangerous because again, these GPUs — you're building a data center, you're renting out GPUs, and right now maybe you're renting out a GPU for $4 an hour. And maybe that way the business makes sense. But these GPUs keep getting so much better every year that that same GPU in just three years may be only renting out at 40 cents an hour, at which point the data center is literally worthless because that won't be enough to cover the expense of operating the data center. So this is where we get in trouble when somebody underwriting at JPMorgan or US Bank or Mitsubishi Bank ignores that. And to answer your question, why would they do that? These are professionals. It's because they don't have the downside, right? They have a mandate to deploy capital into AI. They got an order from their boss who got an order from their boss that says, "We don't have enough AI in our portfolio. Go find me AI to invest in." And so somebody comes to them and says, "Hey, look, I'm building a data center. Lend me money. I'll pay you 9%." That's fantastic interest. And you sign up for it. You get a big bonus that year based on signing that deal. If the deal goes sour, if the data center is worthless in three years, you don't care. You're not giving your bonus back. That's the world we had back in the financial crisis. That's how we got in trouble then, and that's how we could get in trouble now if we don't do something about it.

He's spot on. He's spot on. By the way, Big Short is a story of how he was spot on, but he almost didn't make it. Right. A lot of the movie is about how long it takes to play out, and you can be right, but if you're right too early, you don't make it. And the story is about him and the handful of people that did make it. There were a lot of people that were short the market for a long time and lost everything because they couldn't wait long enough. He was just in a position to wait. And he's spot on right now. And look, depreciation gets wonky. So let me just hit it at a high level because it is really important to this conversation, right? Depreciation is based on an accounting standard that helps companies say, well, I have an asset. How long is it used for? How long can that asset generate revenue for me? And if it can generate revenue for me over five years, then I should take the cost of acquiring that asset and spread it over five years as an expense for accounting purposes. Right? That's what accountants are there to do. And these accountants spent time three to five years ago with companies like Microsoft and Amazon and said, you know what, based on where the technology is now, we're looking at these chips and it looks to us like they can generate revenue for you for about five or six years. And that's why we're going to allow you to extend your depreciation to five to six years, because then you have less expenses and you look more profitable. So we're going to allow you to do that. But what's happened over the last three years is the technology has taken huge leaps forward. Jensen Huang has been preparing for this for decades, and here we are, and we can make the most of the brilliant chips that he's designed, and now he can make one every year that's ten times better than the one he made the year before. And that's great. That's why we have all these great tools, and that's why they're getting so much better. But back to those accountants: if you ask the accountants today how long will this asset generate meaningful revenue, they would not answer five or six years, they would probably answer three years. And to Mr. Burry's point, if you told Amazon and Microsoft, and certainly if you told companies like Oracle and CoreWeave, no, no, no, these chips will only generate meaningful revenue for three years, their profitability would decline very dramatically. So again, for Microsoft, Amazon, Google, I don't worry about that too much. They can handle it. For companies like CoreWeave and Oracle, it means they'll never be able to raise any more capital again, which means they would all go away. So that's why this point is important. Even though it's a little wonky, it's important, because when these companies come back and tell you, "Oh, no, no, no, I have five-year-old chips that work just fine," that's not the same thing. Saying "I have a five-year-old chip that works just fine" doesn't mean that it can generate the same revenue that it did five years ago, which is the accounting question. So that sleight of hand by these companies to tell you that the chip still works — if it's only generating 1% of the revenue it generated five years ago, it's by all intents and purposes worthless. And so that's where we have to ask the accountants the right question, and I think we will be over the next couple of years, and we're going to be correcting this dislocation, which is what Mr. Burry is betting on.

A couple of things here. So first of all, I like that Nvidia picks fun names like Vera Rubin and Richard Feynman to name their chips. I like that as a naming convention. He's clearly a dork, right? In the best way possible. The other thing to note is that Dylan and Dwarkesh are roommates, and boy is that a fun apartment to hang out in, right? Those two are some of the smartest people around, and I love hearing them speak. Now that I've said that, I'll bring you back to the fact that what they just said is sleight of hand, right? The fact that the chip works after three years doesn't mean it's going to generate the same revenue, right? You could have a working chip. I could have a ten-year-old Mac or a ten-year-old PC that still turns on. I wouldn't want to use it because it wouldn't be able to do the things I need it to do. So just the fact that it doesn't break after three years doesn't mean that it can generate the same revenue that it did three years ago. So that's one level of sleight of hand. The other thing they point to is, oh, don't worry about it, we do have five-year-old chips that we're still renting out at decent prices. And that's really just a function of where we are right now in the expansion cycle. We are so short on chips to process these AI transaction — these AI token generation inference transactions — that people are renting out anything they can get their hands on. This is like used cars during COVID. People would pay a premium to a new car to buy a used car because there were just no cars around. That doesn't mean that that used car was worth more than a new car. It just means there was so much scarcity that people overpaid, and that's where we are right now. That's not a sustainable situation because everybody is building out data centers. And again, even if you took out the bad players that are borrowing money to build data centers, you still have Amazon, Microsoft, Google, Meta, Elon using cash flow to build data centers. So we're building a tremendous amount of capacity. Once that capacity even gets close to catching up, then the old chips that can't do as many calculations for you will be worth a fraction. And this is where the market's going to end up right here. Markets are efficient. And here's where the equilibrium is. Here's where we'll get to the balance of supply and demand, is on dollar per flop. Dollar per flop is to say dollar per calculation, right? Remember what these AI chips do is they generate tokens, which is what we call words or numbers or images. And if a Richard Feynman chip can generate X tokens per second, and an H100 can generate one-millionth of a token per second, or take a million seconds to generate a token, then the Richard Feynman chip is worth a million times more than an H100. Even if the H100's working, we won't use it because it can't keep up. It's not worth it because we'd rather use a chip that can generate the amount of tokens that we need. And so this is two different conversations. Will the chip work? It might. But will it be worth something? Will it be able to do enough computation to generate revenue? That's a completely different question. And that's where we think in a three-year time frame, three-year-old chips will just not be able to do enough computations to be worth keeping them on. So either we replace them or we use them for much less important things that will generate much less revenue. So again, we have to be careful not confusing does it work with can it generate revenue.

That's exactly it, is that we're one accounting conversation away from having all these companies have to report much lower profits. And you know, we use profit multiples to value companies. So if a company has to depreciate most of its assets over three years instead of five years, that means their profitability is going to go down.

Proportionally, and we could have, in a stylized case, the value of a company decline by 40% because an accountant said you have to depreciate this over three years instead of five years. That's why this is very real. Sounds wonky, but this is very real. If you reduce a company's profitability by 40%, their value will go down by 40%.

That's right. And again, I go back to Microsoft, Amazon, Google, they can handle it. They have a big business. It's diversified. This is only one part of the business. They can literally stop on a dime. Again, they're deploying cash because right now that cash is sitting on their balance sheet and generating 4% returns. And so they're saying, 'Well, this AI thing is huge. We think it can generate 15% returns based on our math. Let's use the 4% cash and deploy it here. We think we can get those returns. But by the way, the second we think that stops, we could stop our capex on a dime, go through a couple years where we don't do any capex while we absorb the previous investment.' And those companies will be just fine. It's the companies that either have the debt or are lending the money or have equity investments in highly leveraged entities. Those are going to be the ones that can't handle that kind of a transition.

Yeah, maybe it's worth having the... Our mental framework for looking at AI is that there's three camps, right? We have to take a weighted probability of three outcomes. There's the pessimist outcome, which is AI is cute and it's useful, but it's cute like the metaverse, or it's cute like social media, in which case it'll be a useful technology, but really we're spending way too much money on it. And a lot of people are in that camp right now. I'm not sure I agree with that. There's the optimist scenario, which is AI is the most powerful technology in a long time. It will be so powerful that it will make us so much more productive that it will drive an acceleration in GDP growth. That is where Microsoft, Amazon, Google are at from their perspective. And then there's a maximalist scenario, which is we are maybe as close as a couple of years away from superintelligence, a technology that can do anything a human can do better than any human. In which case it's going to replace us en masse, create untold wealth to whoever owns a piece of that, and therefore no investment we could possibly make is enough in order to get there, especially if you believe that only the one entity that gets there first will own everything. This is where Mark Zuckerberg lives. Sam Altman, Elon Musk, Dario Amodei, they live in this maximalist camp of this is a race that we can't afford to lose, and therefore we need to build everything we can. Now, all three of these things are possible, and so you have to plan ahead for that. But most of us are in this optimist camp, which is we should invest a lot. We just have to be thoughtful and careful about how we do it. So if we're wrong and it's the pessimist scenario, we don't bring everything down with us.