'Big Short' Investor Explains How the AI Bubble Will Burst
- Steve Eisman sees AI’s immediate danger as concentration, not lack of demand. Nvidia grew revenue roughly 110%, yet 70% of its accounts receivable came from five customers; meanwhile, OpenAI and Anthropic reportedly generate 70% of hyperscaler AI revenue, equal to 25-30% of total cloud revenue. “The entire chain from Nvidia to the hyperscalers…rests on the future health and growth of Anthropic and OpenAI.”
- OpenAI looks like the weaker link because its costs are outrunning already-rapid revenue growth. Eisman cited roughly $6.5 billion of quarterly revenue against about $12 billion of costs, with revenue rising $1 billion while costs rose $3 billion; Anthropic’s revenue, by contrast, had grown more than 100% in three months. If OpenAI failed, he argued, “the whole United States of America would go into a recession overnight.”
- The capital loop can hold only while the model companies keep growing and their strategic backers keep funding them. Eisman had assumed venture capital supplied most of the money, but said Amazon, Google, Microsoft, Nvidia and SoftBank are the major funders—creating “a circularity to the financing.” It breaks if either leading lab stumbles and investors stop “ponying up.”
- LLM providers have no durable moat because customers can switch models and are becoming intensely token-cost conscious. After OpenAI and Anthropic raised prices, “token maxing” faded: Uber reportedly exhausted its AI budget in three or four months, while another company spent $500 million before realizing it. Users can reserve premium models for critical work and move routine tasks to open-weight alternatives such as “Kimi K3”: “You don’t need a Cadillac for everything.”
- Eisman suspects existential-risk rhetoric is partly an attempt to manufacture the regulation that would create an otherwise-missing moat. His theory is that incumbents want rules excluding Chinese open-weight models and leaving a protected duopoly. He discounts calls to slow development, arguing that if Anthropic CEO Dario Amodei truly believed his product was dangerous, Anthropic could postpone its IPO; it has not. “I take people seriously when they put their money at risk.”
- Hyperscalers retain scale moats, but AI capex has transformed their cash-flow profile and exposed them to fragile customers. Eisman cited approximately $700 billion of hyperscaler AI spending this year, Oracle’s heavy dependence on OpenAI commitments, and Google raising $85 billion of equity despite historically generating abundant cash. His suggested picks are the recipients of that spending—Nvidia, Micron, GE Vernova, Arista Networks, Cisco and Eaton: “Picks and shovels.”
- For the broader market, Eisman watches the 10-year Treasury rather than the Fed’s overnight rate, with 5% his tentative “Rubicon.” He previously thought 4.5% would trigger trouble and admits, “I was wrong”; with the yield around 4.98% after crossing 5%, he thinks a sustained breakout could produce a correction. On federal debt he wishes the deficit were smaller but doubts an imminent debt spiral because Treasuries underpin the global financial system; the host added that there is currently no alternative.
1. Concentration holds up the AI capital stack
Eisman’s starting point is customer concentration: an established software company with thousands of customers is resilient, while one with two can be “dead” if either fails. AI increasingly resembles the second company, despite the scale of the spending around it.
Nvidia’s roughly 110% revenue growth is extraordinary—“the largest company on planet Earth” growing above 100%—but Note 7 of its filing showed 70% of accounts receivable at July-end coming from five customers. The host called that a warning flag, not the end of the world; Eisman used it to illustrate the concentration risk.
The dependency intensifies downstream. Eisman said OpenAI and Anthropic account for 70% of AI revenue at Microsoft, Amazon and Google, equivalent to 25-30% of their total cloud revenue: “The whole chain” therefore rests on those two labs maintaining growth and paying their commitments.
OpenAI appears weaker for now, though Eisman stresses that disclosure is inadequate and the ranking could change. He cited roughly $6.5 billion of quarterly revenue, approximately $12 billion of costs, and a three-month period in which revenue rose $1 billion while costs rose $3 billion: “We like to say [that] is upside down.”
2. Circular financing can persist—until confidence disappears
Eisman originally believed venture capital supplied most model-company funding; he now says the principal backers are Amazon, Google, Microsoft, Nvidia and SoftBank. The host’s pushback—worth keeping—is that vendors appear to fund customers who then make commitments back to those same vendors. “That sounds like a circle.”
Eisman concedes similarities to the financing loops dramatized in The Big Short, but rejects an automatic collapse call. As long as OpenAI and Anthropic grow “very, very rapidly” and investors keep supplying capital, the chain can hold; the break comes when one lab seriously disappoints and new money retreats.
Oracle makes the exposure concrete. Its RPO, effectively backlog, reportedly jumped from roughly $150 billion to $400 billion in three months. Eisman said sell-side work attributed about half to OpenAI. He later said Oracle itself was over $600 billion, while still describing roughly 50% of its future revenue or backlog as tied to OpenAI—a company that “bleed[s] money.”
3. LLMs have usage growth but no customer captivity
“Token maxing” meant employees were instructed to use AI constantly, needed or not. OpenAI and Anthropic initially undercharged relative to token costs, then raised prices; customers became much more cost-conscious.
The budget shocks were vivid: Uber reportedly exhausted its AI allocation in three or four months, while another company spent $500 million before recognizing the bill. Cost awareness pushed routine workloads toward cheaper open-weight models, leaving premium systems for the highest-value tasks.
That destroys the loyalty needed for a moat. Developers using Claude today can switch tomorrow if another model is better; leadership rotates among Gemini, ChatGPT and Claude. Unlike Google search at its height, no provider commands habitual, near-unquestioned usage: “There are no moats.”
Eisman’s explicitly labeled conspiracy theory is that AI-doom rhetoric seeks government intervention. Regulation could bar Chinese models and manufacture a duopoly—a “legal barrier” where no commercial moat existed. Regarding Dario Amodei’s call for a slowdown, Eisman said that if Amodei truly believed his product was dangerous, he could postpone Anthropic’s IPO; it has not been postponed.
4. Hyperscaler strength now finances model-layer weakness
Hyperscalers do possess moats: almost nobody has “a hundred billion dollars to spend on building data centers.” Their vulnerability is that these capital-heavy, defensible platforms increasingly depend on LLM customers that lack switching costs and remain structurally cash-hungry.
Microsoft, Amazon and Google once generated more cash than they could reinvest and returned the surplus through buybacks. Data-center spending has now consumed much of that cash flow—sometimes making it negative—and Eisman described Google raising $85 billion in equity as unimaginable only one or two years earlier: “World changed.”
Oracle shows the market’s changing tolerance. After rising from roughly $230 to $330, the stock later fell through $200 to about $150; even after decent results produced a 7% opening gain, it closed down. The host noted a downgrade and an S&P debt rating of BBB-minus; Eisman said investors were nervous about Oracle’s debt and had no specific news explanation for the decline.
5. Chip depreciation is secondary to whether the chain reverses
The host raises Michael Burry’s argument that hyperscalers extended chip depreciation lives from roughly three-to-four years to five-to-six, mechanically lowering reported expense even as new generations risk making existing chips obsolete. Eisman grants that the accounting change creates a large profitability benefit.
His disagreement is practical rather than mathematical: current demand remains so strong that older chips still command demand and higher pricing. “With all due respect to Michael,” the depreciation case is “too academic” for the present market.
If AI succeeds, changing depreciation schedules will not determine the outcome; if OpenAI fails and commitments reverse, the correction will overwhelm any accounting effect. Burry may be sensing that “something [is] wrong,” Eisman says, but he is pointing to a factor too small to drive the decisive move.
6. Power creates the clearest picks-and-shovels exposure
On whether US power availability will cap AI deployment, Eisman’s honest answer is: “Get in line. Nobody knows.” Some specialists describe serious delays and others say supply is manageable; he cannot yet reconcile the evidence into a confident bottleneck call.
GE Vernova, ticker GEV, illustrates the demand shock. A business once assigned roughly negative $3 billion in sum-of-the-parts value now faces US electricity growth of 3-4% annually—incremental demand Eisman compares with adding “two large cities” each year.
Gas turbines sit at the center of that buildout, alongside the electrical equipment needed to connect new generation. Eisman recalled orders rising about 85%; the machines are enormous and take years to manufacture. The host said GE Vernova’s backlog extends through 2035, making backlog and orders more useful indicators than narrative alone.
The host’s Memphis example sharpens the physical constraint: when the grid could not serve the Colossus data center quickly enough, 35 portable gas turbines were installed on site. Alternatives such as nuclear and Bloom Energy exist, but Eisman expects most near-term supply to remain gas-turbine-led.
7. Own the spending recipients and demand resilient stories
Eisman would not invest in an LLM provider because there is no moat, and he is wary of hyperscalers after the erosion of their free cash flow. His suggested picks are the companies receiving the capex: Nvidia, Micron, GE Vernova, Arista Networks, Cisco and electrification supplier Eaton—“picks and shovels” rather than glamorous software.
The software risk is selective, not universal. The host described a Meta agent she called “Muse” that could book flights and hotels without users visiting travel sites, weakening customer ownership and sidebar advertising; Eisman agreed that payments may also become “dicey.” Deeply embedded enterprise products such as Salesforce or ServiceNow retain stronger switching moats.
His core method is “story but rational”: he wants real growth tailwinds, dislikes management selling stock and generally avoids cyclicals because that becomes an economic forecast. Moats help explain holdings such as Moody’s and Visa, but are not mandatory. The host said she would not buy Palantir at its current price because it was too expensive, and Eisman agreed.
On Scott Bessent’s Treasury buyback plan, Eisman said the aim was to buy long-term Treasuries, bring rates down and ease the government’s interest burden. He said the announced $6 billion was negligible relative to a $40 trillion deficit and had failed so far; he suspects another measure may follow but does not know what it will be.
In macro, the 10-year yield matters more than Fed funds because it determines long-term borrowing costs. Eisman now guesses 5% is the market’s “Rubicon,” after incorrectly choosing 4.5%; by contrast, he doubts an imminent federal debt spiral while Treasuries remain the global financial system’s backbone. He has never owned precious metals because they are unproductive, and regards Bitcoin as worse: it trades opposite the behavior implied by its inflation-hedge thesis, making it “a way to speculate about speculating.”
Full transcript
Steve, welcome back. You are known for spotting a bubble before anyone else does. Michael Lewis wrote a whole book on it. So my first question to you is: Where on earth are we in AI right now?
There was a great movie with Sean Connery where he played—I can't remember the name—Finding Forrester. I remember the young character asks him a very complicated question, and his response, as he's eating soup, is, “It's not exactly a soup question. It's a complicated question.” [laughter] I never forgot that line. I thought it was one of the best lines in the history of movies. It's not a soup question.
Not exactly a soup question, is it?
This is how I look at it. The concentration risks here are awe-inspiring. You take a step back, and someone said to me, “Why don't you analyze this software company?” Forget about what it does; it's a software company, an established software company. If it turned out the company had thousands of customers, that would be great. If it turned out the company only had 2 customers, you'd say, “I don't want to invest in that, because if something bad happens to 1 of those customers, this company is dead.” There's something of that going on in the whole AI story.
So let's start just with Nvidia. Nvidia, God bless them—and I own the stock, okay? When they reported a few weeks ago, I think the revenue growth was like 110%.
It's a lot.
So let's just take a step back for a second and say to ourselves, “Wait a minute. The largest company on planet Earth just had—forget about earnings growth, which was great too—revenue growth of over 100%.” That's insane. That would say the AI story is great until you read the 10-Q, which came out that night. I'm going to impress your viewers by saying, if they look at it and go to Note 7—
Oh, okay.
Okay, yeah. Note 7 says that 70% of Nvidia's accounts receivable, as of the end of July, came from 5 customers.
Warning flag. Not the end of the world, but a warning flag.
Now let's go to the hyperscalers. You're talking about Google, Meta, Amazon, Microsoft, and let's throw Oracle in for good measure: massive companies spending massive amounts of money buying Nvidia's chips and everything else under the sun. Seventy percent of their AI revenue, which equals something like 25% to 30% of their entire cloud revenue, comes solely from Anthropic and OpenAI.
Right?
Let me just say it again so your viewers get it. If you look at Microsoft, Amazon, and Google—leaving Oracle out for a second—70% of their AI revenue, which is equivalent to 25% to 30% of their total cloud revenue, is just from Anthropic and OpenAI. If you go to Oracle, Oracle puts out a data point called RPO, which is basically a form of backlog. When they reported earnings last year in October for their August quarter, their RPO went from like $150 billion to like $400 billion—
In 3 months.
It was a massive jump.
And the stock went crazy.
The stock went crazy. It went from $230—
To $330 in 2 days.
Yeah. Then some of the sell-side analysts who are very good, who did some digging, came out with reports that said 50% of that RPO is just from OpenAI. Now Oracle is over $600 billion, and it's still like 50% from OpenAI. Basically, 50% of Oracle's future revenue is from a company that loses money like crazy.
Well, that's what I was going to ask you. Maybe you can explain how this works, but from my perspective, I don't understand where this money is coming from.
We're coming to that. Let me just finish the chain.
So now we come all the way to Anthropic and OpenAI. My view is that the entire chain, from Nvidia to the hyperscalers and all the way down, rests on the future health and growth of Anthropic and OpenAI because they're creating the commitments to the hyperscalers. If they don't grow and have the money to pay for those commitments, the whole chain slows.
Yeah.
So that's the risk. Between the 2, I'd say OpenAI is probably the weaker entity, but it's not clear because we really don't have enough numbers. The only thing we do know, because The Wall Street Journal reported this, so I'm assuming it's true, is that OpenAI had something like $6.5 billion in revenue in the second quarter of this year.
Okay.
Anthropic was at $11 billion-plus—
But this is just revenue.
Just revenue.
Okay.
OpenAI had costs of something like $12 billion. So the way the math worked was that in 3 months, OpenAI's revenue went up $1 billion and its costs went up $3 billion, which we like to say is upside down.
You want the reverse. You want the other way to go.
Great business model. Their revenue grew 18% in 3 months, whereas Anthropic's revenue grew over 100% in 3 months. So they're the weaker company at this point. That could change. My only point is that this whole industry makes me nervous. Let's just imagine that OpenAI fails—
Could happen.
The whole United States of America would go into a recession overnight. Eventually, there'll be a lot more diversification and a lot more companies, but that's going to take time. Within the next year or so, those companies have got to stay healthy. That's where I think we are. There's a concentration risk.
There's a massive concentration risk.
And that brings me to the question I was asking before. My understanding is that OpenAI and Anthropic are not making money. They're losing money.
Excuse me. To say that they're losing money would be kind. They bleed money.
What's the nicest way?
They lose a lot of money.
We're pre-profit, right? Pre-pit. [laughter] Something like that. They make money if you exclude all costs.
That's how they like to think about it.
That's something we should start reporting as a metric.
Yes, we should. [laughter]
So where does the money come from?
That's actually a very interesting question. I had thought that most of the money was coming from venture capital, and that happens not to be true. Most of the money is coming from Amazon, Google, Microsoft, Nvidia, and SoftBank investing in these companies.
Okay. So, taking equity stakes—
Taking equity stakes. They raise capital, and those guys have ponied up money. Whether they want to continue to pony up money, I don't know.
That sounds like a circle. It sounds like money or commitments are going one way, and then commitments are coming back the other way.
Yeah, it does have that tendency, does it? This kind of reminds me of a scene from The Big Short with your character.
You had to work that in there, didn't you?
I did have to. I remember you saying, “Oh, it's kind of like a CDO, and then they put parts of that into a CDO-squared, and those 2 put parts into a CDO-cubed.”
There's some similarity, obviously, but in their defense, there is a circularity to the financing. As long as Anthropic and OpenAI keep growing very, very rapidly and people keep giving them money, the chain will hold. It's if 1 of those 2 companies really messes up and people pull money or don't want to invest anymore that the chain doesn't hold. That's where the concentration-risk problem comes in.
If the industry was much more diversified—if, instead of 70% of hyperscaler AI revenue coming from Anthropic and OpenAI, that number had been 10%—we'd be having an entirely different conversation, because clearly there are a lot more customers out there of size.
Let's just plug that podcast for a second and call it the Real Eisman Playbook.
The Real Eisman Playbook. Yeah, absolutely. It's very, very good, and it's gaining a lot of traction. I think it comes down to the authenticity of it.
Well, I appreciate that. I really enjoy it. I think it's very, very high-quality.
My wife and I work on it every day.
Both of you? I didn't realize.
My wife is my partner in this. She does all the editing, God bless her.
Oh, really?
This gives you an insight into things. I am a very linear thinker, which I would define as 1, 2, 3, 4, 5, conclusion.
Got it.
Too often I write that way. We have this thing called the Weekly Wrap, which we put out on Friday, where I sum up the week. Too often, when I write it, I'll do 1, 2, 3, 4, 5, 6, and Valerie, my wife, who's the editor, will always say, “You buried the lead again.”
Yeah.
And she'll flip it.
Ah, okay.
And so, because she says most people never get to the bottom. People get your conclusions at the bottom, and it takes 10 minutes to get to the bottom. So she edits it, and she runs the business.
1. Chinese Competition
Right? Okay, that's cool. I didn't know that. That's a little behind the scenes. I like that. Back to the AI train of thought before I forget my train of thought again. Going on from what you're saying, if something happens to OpenAI or Anthropic, that's where the problems could be at this moment in time, when we've got the concentration risk. How do you see things like DeepSeek's new model or Kimi—what is it called now? Kimi?
Best name in the biz.
Kimi K3.
So this is where the industry has, I think, real business weakness. Let's leave concentration risk aside; that's its own risk. Here's the business risk. There was something going on for a while called token maxing, which is where, for lack of a better term, I work for a company, and I'm a software engineer, and I've been told by management, “You're to use AI 24/7 whether you need it or not.”
Just do it.
Just do it. Right. What happened was, last year, OpenAI and Anthropic were dramatically undercharging for their services. Then they raised prices because they were so undercharging for the cost of tokens. It was killing them, right? So they increased the prices so that the customer was bearing more of the cost of the token.
So you get people in, and once you've got people in, raise the price.
Raise the price. Uber, I think, blew through its entire AI budget in 3 or 4 months. Another company that I read about spent $500 million before it even knew it had spent $500 million. So what's happened is people have gotten a lot more cost-conscious. Token maxing has gone away, and people are using these open-weight models a lot more.
You don't need a Cadillac for everything. People will use Anthropic and OpenAI models only for the super-important tasks. Everything else they'll use Kimi K3 or whatever. So what I like to say about this is that this industry, despite all the hundreds of billions of dollars that's been spent on it, has no moats.
There are no moats.
This is Google with its search. Until very recently, that was a moat. Everybody used Google. I mean—
90%.
90% of planet Earth used Google, and no one even thought about it because it was just better and nobody could approach it. Here, one day it's Gemini, one day it's ChatGPT, another day it's Claude. They just rotate.
2. Manufactured Hysteria
We'll come to it—about my conspiracy theory about the end of the world.
Conspiracy theory. All right.
Okay. Okay. So, as all your viewers know, the world's going to end at some point. It has to.
At some point, it has to. Maybe it could be 5 billion years from now or in a couple of weeks.
Yeah. CNBC and Wall Street. I think this entire “AI is going to end the world” thing is a complete subterfuge.
Okay.
What I think is really going on is that these companies are nervous. They're nervous that token maxing has ended. They're nervous that there are no moats. They're nervous that these Chinese open-weight models are taking massive market share. So they're manufacturing a hysteria.
What they're hoping for is for the government to come in and regulate the industry, and they think that they could manipulate that regulation to create a duopoly. The regulation will create the moats. The regulation will say, “No Chinese AI models. There's too big a risk.”
I see.
And then all of a sudden there's a moat that didn't exist before—a legal barrier. That's what I think is actually happening here.
So you think that's the reason behind it? I've noticed Elon Musk is always talking about it, but recently the Anthropic CEO, Dario Amodei—
And they're also talking about slowing down.
Slowing down.
That cannot be taken seriously. If I was Dario Amodei and I really, really thought that my product was dangerous and I really needed to slow down, I'd postpone my IPO.
You'd have to postpone your IPO.
You could just take the steps. Take a step back and, you know, fix what we need to fix, and we'll come back. Are they postponing their IPO? No.
Elon Musk had a very funny quote the other day, which I think he did on X, where he said something like—I’m going to paraphrase; I don't have the exact quote—“This is some messed-up 4D chess where you're saying that your product's going to end. Oh, and by the way, how much can I allocate to you for the IPO?”
Yeah.
Seriously, I take people seriously when they put their money at risk, right? This statement about a slowdown is just all part of this hysteria that they're trying to manufacture, right?
3. No Moats
So they're aware that there may not be moats. For those that don't know, explain what you mean. We're talking about moats specifically in the LLM providers, not the hyperscalers, because there—I guess there's still—
The hyperscalers have moats.
Okay, hyperscalers. So what's a moat?
Your grocery store doesn't have a moat because somebody could open up a grocery store across the street tomorrow. But there are some businesses that have real moats around them, like a competitive advantage. It's a competitive advantage that is eternal, or at least very long-lasting.
Nvidia makes GPUs. Well, nobody else really makes GPUs. That's a moat.
Mhm.
Certain software companies have them. If you're Salesforce or ServiceNow, which are two massive software companies, your product is embedded in the companies that you service. Those companies that use your product have used it for so long, they can't function without your product.
Hard to switch away.
Trying to switch out of that is brutal.
Yeah.
That's a moat. With the hyperscalers, what's the moat? Well, I don't know about you, but I know that I don't have $100 billion to spend on building data centers.
Okay.
They just don't have it. So there is some moat.
So, just in terms of size and money, there are only certain companies that can actually build data centers.
They're just that expensive. LLMs—the creation of the models is expensive, but there's so much competition and there's no loyalty. If you're a software developer and you're using Claude, if tomorrow another LLM shows up that's better than Claude, you'll switch.
Yeah.
You switch. You're not stuck. That's the problem with the LLM model.
Gotcha. So there's no moats in LLMs.
Yes.
Hyperscalers have diversified business models, so there's—
And they have moats.
But they're dependent upon the LLMs in their cloud businesses. That's their weakness right now. So there's kind of an argument for hyperscalers having a moat and against having a moat. In some ways, there's the scale, the cost, and the barrier-to-entry moat of the investment, and then there's also the dependency on the people that are buying—
That have no moats.
Yes.
That's the problem.
Interesting.
4. Changing Business Models
The other problem with the hyperscalers—this may be temporary, but then again, maybe it's not—is that Microsoft, Amazon, and Google, 3 or 4 years ago and way prior to that, were incredibly profitable. More importantly, they just threw off cash like crazy. They had so much cash they didn't even know what to do with it, so they just bought back stock because they literally didn't have enough investments to pour back into their own businesses.
Today, because of the incredible amount of money that they're spending on these data centers, their cash flow is gone.
I noticed that.
In some cases, negative.
Negative now. Very little.
I mean, Google raised $85 billion in equity capital. If you had said to me a year or 2 years ago, “Hey, Steve, I'm going to make a prediction. Google, which hasn't raised capital since it went public, is going to raise $85 billion—not in debt, in equity capital,” I'd have said, “You're out of your mind. You're insane. What are you talking about?” They create $85 billion in cash overnight. Why would they raise equity capital? Well, the world changed.
It's a very dramatic shift. As you know, I'm an investor who has held Google for probably 8 years, and the company that I hold now is very different from the company that I held then.
And that's an interesting point. We've seen the market get a little bit jittery with the amount of spending that's happening. There could be a payoff; maybe there's not. By the way, let me just jump you for one second. Let's go back to Oracle.
Okay.
Because after Oracle—I didn't finish. After Oracle, people said that 50% of Oracle's RPO, the backlog, is from OpenAI.
The stock, which had gone from $230 to $330 over the next 2 or 3 months, went to $200, and today it’s $150. And what’s fascinating—
Oracle just reported, and the numbers were pretty good. The stock was up 4% or 5% after hours, was up 7% at the open, and closed down on the day.
Right.
There was no news, so I don’t have a news explanation. Nothing happened.
Clearly, people are very nervous about Oracle because Oracle was downgraded, and its debt rating is triple-B minus by S&P, which is, I think, maybe just 1 level above junk.
So people are nervous about Oracle and how much debt they have.
Well, it seems investors are nervous about all of these hyperscalers now that are investing literally—
Hundreds.
Hundreds of billions. It’s insane. It’s staggering. The numbers are just so big.
I think the number that I heard this year is that if you just look at the hyperscalers, they will spend $700 billion on AI capex this year.
That’s such a huge number. It’s hard to even get your mind around it.
It’s so enormous.
I’m interested in your perspective on what Michael Burry has been saying. He’s concerned that the data centers are taking too long to come online, and they’re buying so many chips. His opinion is that the chips become obsolete way faster than the depreciation schedules.
I understand his argument. So, let me give it its due first. What he pointed out last year, I think in November, was that the hyperscalers had changed the depreciation schedule of the chips from 3 to 4 years to 5 to 6 years. If you did the calculation—I can’t remember exactly what it was—it’s an enormous increase in profitability just from the change in that accounting.
By changing your depreciation schedule from 3 to 4 years to 5 to 6 years, by definition, your depreciation expense, which you report, is going to be lower, all other things being equal. He also said that there are so many new chips coming that they become obsolete.
Where I think he’s wrong, for the moment, is that there’s such demand for chips right now that there’s still huge demand for the older chips, whose prices have gone up with all the other chips.
Right.
So, with all due respect to Michael, I think his argument is too academic. If AI succeeds because Anthropic and OpenAI grow like crazy, and the hyperscalers do well, it’s not going to matter if the depreciation schedule changed from 3 to 4 years to 5 to 6 years.
Right.
At the same time, if OpenAI fails and the whole chain goes in reverse, we’ll have a massive correction that has nothing to do with the depreciation schedule. I think what he’s trying to point out, deep down, is that maybe there’s something wrong here. But I don’t think the thing he’s pointing to as being wrong is important enough.
5. AI's Power Problem
Right. There are much bigger factors that play in both directions.
Much bigger factors.
Okay, interesting. So, we’ve spoken about no moats, and we’ve spoken about China potentially coming in as competition. Another headwind that I’ve been trying to wrap my head around is the power element, because this is another one of those big things that we’re talking about—earnings and chips and this and that.
When I started to look at power, I think it was Elon Musk’s interview with The Economist that opened my eyes to it. He said, “China has a chip problem. The U.S. has a power problem.”
He’s right.
However, in his view, China can solve its chip problem. It might take some time, but a harder one to solve is the power problem because power is physical infrastructure. It takes a long time.
Correct.
I don’t know if I have the expertise or the understanding to know how big of a restraint or bottleneck power in the United States is actually going to be.
Get in line. Nobody knows.
Right? I mean, I keep looking. There are people who say it’s a big problem, that things are slow. There are other people who say things are fine. I can’t nail it down yet.
I do know that the companies involved with power are doing great.
Like GE Vernova, for example. That stock has gone nuts. I’ll pat myself a little bit on the back: I bought that stock really early.
Oh, really?
But I bought it because the sell-side analyst I’m very friendly with told me I should buy it, and I just took a flyer on it.
For those who don’t know GE Vernova, tell me about it.
Let me tell you. It’s very interesting. GE used to be composed basically of 3 massive divisions: healthcare, aerospace, where they basically make the jet engines for planes and then service them, and what they called energy.
If you ever saw a jet engine and looked at a gas turbine, which is what goes into a utility that creates electricity, they look exactly the same. It’s just that the gas turbine is much bigger.
Yes.
It’s basically the same technology. The energy division of GE makes gas turbines. They have all this electrical equipment that they sell, and then they have a wind division, which does terribly.
Now, this should show you how fast the world can change. The energy division was created around 2015. GE bought a company in Europe called Alstom. Alstom was an energy company that also created gas turbines, and GE had a business that created gas turbines. So, they mushed them together just in time for the entire gas turbine business to fall apart.
Right.
This is why Jeff Immelt, who was the CEO of GE, finally lost his job, because that was enough already. Eventually, all 3 divisions got spun out. There’s GE Healthcare—I think its symbol is GEHC. There’s the energy division, which is called GE Vernova, and its symbol is GEV. Then there’s GE Aerospace, which is the aerospace division.
Gotcha.
A year before GE Vernova got spun out, which would have been maybe 2022, if you were to read sell-side reports about the industry, the energy business was so bad that they ascribed negative value to Vernova—negative value. They said it was worth negative $3 billion.
Oh, my gosh.
That was in terms of a sum-of-the-parts analysis.
Right.
Now, what’s happened is that even before the whole data center thing, electrical production in the United States finally started to increase for the first time in 15 years. Add on top of that the data centers, and you’re talking about U.S. electricity growing 3% to 4% per year.
That may not sound like such a huge number, but 3% to 4% off the base of the United States is the equivalent of 2 large cities.
Okay. That’s a lot.
It’s huge.
Yeah. A company like GE Vernova has a backlog through 2035. That’s how crazy things are.
Because this is how these data centers are being powered: with these gas turbines, right?
Mostly.
Mostly. There are some alternatives. People are talking about nuclear, and there’s a company called Bloom Energy, which makes its own little turbine that you could hook up to a data center.
But most of it is going to be through gas turbines. That’s what Elon had to do with the Colossus data center that he built in Memphis. The grid was too slow; it wasn’t ready. So, he ended up getting 35 portable gas turbines—
Right, and hooked them up to his—
Hooked them up. He created his own mini power center on-site, right next to it.
Okay. So, most of GE Vernova is—
Well, it’s the gas turbines. It’s all the electrical equipment. Think about it: when you’re building a new utility plant, it isn’t just the turbine. There’s all this other electrical equipment that has to get hooked up. They make that, too. The wind business will always, I think, be a crappy business.
And that’s fine, right? But you would say the thing to look out for in GE Vernova’s case is the gas turbines. Is that the core of the business?
Yeah, that’s the core. You would just want to look at the orders, which I think in the last quarter were up like 85%—something insane.
I mean, those gas turbines, in the room that we’re in, would be like 5 of these rooms combined. That’s how big these things are. They’re huge, and it takes years to build them.
Yes. I think I read something that their backlog is stretching past 2030, or something like that.
People want to line it up as much as they can.
But is that even more of an argument for this power problem? If people are making orders now because they want these turbines now, and the response is, “Hang on, we’ve got 2030”?
I just don’t know. I really don’t know. I don’t have enough information.
Yeah. I can’t remember who it was that you interviewed—the man who knew a lot about power—but I found that to be a really good interview. I might leave it linked on screen right now. I thought that was a really good explanation.
Okay, so we’ve covered a lot of headwinds when it comes to AI.
6. How to Invest in AI
I think you went on the record saying that if you try and predict what's going to happen, you're a fool. So don't try and predict it. I think there are a lot of people out there who feel compelled to look at these AI plays and look into the realm of AI. If someone comes to you and says, “Oh, look, Steve, I really have to get into AI somehow,” what are some of the safer ways to play AI? And what would you say are the high-risk ways to play AI, talking about just investing in stuff?
I wouldn't invest in an LLM. As I said, there are no moats.
Because there are no moats.
Yeah, that makes sense. I might be a little wary of the hyperscalers at this point, just because their businesses have lost all their cash flow.
But I would be looking at the companies that are getting that cash flow, right?
So that would be Nvidia. Maybe you'd want to own Micron, GE Vernova, Arista Networks, Cisco, and then, if you want to get into the industrial side, you could talk about Eaton, which is an electrification company. That's what I would do.
So it's more picks and shovels.
Picks and shovels, right?
As opposed to the flashy software side, right?
Right.
Okay. The software—the whole software industry. I'm sure you've heard the word “SaaS apocalypse.”
I have. All I know is there will be software companies that will have problems.
Because, you know, take this new agentic AI, Muse—I think it's called—that Meta put out. If I want to book a flight, I say to my Muse, “I want to book a flight to Miami on such-and-such a date. Book me in the best hotel in Bickl [?].”
Okay.
And it goes and does it. Well, how does it do it? It goes on all the travel sites, finds the best price, and books it.
Mm-hmm.
Well, that kind of makes the travel sites worth less because you're not going to Booking.com or Travelocity or whatever directly anymore.
You're not their customer anymore. Your AI is.
Your AI—my AI agent—is my customer. It does the work, and the sidebar ads don't work on that.
So stuff like that—I think stuff in the payment world could get dicey. On the other hand, software that's deeply embedded in enterprises is probably okay.
Well, that's what I was going to ask you. It sounds like the most important thing to look at is the switching moat and how resilient it is.
How resilient is it? Okay. I mean, for Booking.com, what's the switching moat? I go on the Booking.com website and book a trip. So now I don't go on the Booking.com website; my agentic AI finds just the best deal.
7. Michael Burry's Thesis
So that kind of makes the online travel companies worth less. I think it's a little early, but I think that's a possibility. Yeah. Okay. So I guess another argument that I've heard, and I'm interested to hear your overarching thoughts on this around AI, is that people are very quick to liken it to 1999—a tech bubble 2.0.
Oh, no. That's what the media will say, right?
I'm very interested in what your thoughts are on this. I have my own opinion, but I'm interested to hear what you think. We're in the same setup: new technology and speculation in financial markets. Similar setup. Is it different this time? Is there anything fundamentally different?
I don't know if it's different or not. I think it's too early. If OpenAI or Anthropic fail, you'll have a real correction, and then the next generation of people will come up and pick up the pieces. I don't know if that's going to happen or not, so I just don't know.
8. How Steve Invests
Fair enough. Going back to the investing argument, I'm actually interested because I didn't ask you last time, and I had some subscribers who were interested in understanding how you actually go about your investing—not what stocks you're picking or anything like that, but when it comes to the real Eisman playbook, what is the real Eisman playbook? How do you analyze companies? Do you stick within a circle of competence? Do you go down rabbit holes? Do you look at certain financial metrics that you really love to see or not?
Well, a couple of things. I'm very story-oriented.
Story. Okay.
I am not a quote-unquote value player, I think. Stocks are cheap. They're probably cheap for a reason.
Okay. But I can't see you being someone who would grossly overpay for something either. I see you as personally very rational. I mean, I would have loved to have owned Palantir, but I won't buy it now because it's so expensive.
Yeah.
So, story but rational.
Very story. Story but rational.
Okay. Are there any metrics on your checklist or anything that you love to look at that might be a red flag? I'm interested to see what you really look for. Is a moat a must-have for you?
Not necessarily. I like a moat. That's why I own Moody's, for example. That's why I own Visa. But it's not a complete requirement.
Is there anything that you particularly hate to see in a company? What are some things that will instantly turn you off?
Management selling stock. I generally don't like cyclical companies, because then you're just predicting the economy. There are exceptions, but I like companies that have a real story with real growth tailwinds. That's what I like.
Okay. Interesting. Do you mind if I finish off by asking you some questions from the audience?
Sure.
Is that all right?
9. US Debt
All right. I had a quick screen, but I might have forgotten some.
All right, let's—oh, this is a really interesting one. I did want to get your opinion on this: US debt. This is such an interesting topic, and it's very highly covered. It's at $40 trillion now. The average interest rate on it has gone from 1.77% in 2020 to 3.45% today. The interest expense has risen from $523 billion a year to $1.22 trillion. Is that something investors need to be genuinely worried about? Is there a real risk of a debt spiral in the future?
All things being equal, I wish the deficit was smaller. I have my doubts about a debt spiral. Number one, we are the reserve currency of the world. But maybe even more importantly, US Treasuries are the financial system of planet Earth. For example, banks all over the world do something called repos, where they lend to each other overnight. They do it through overnight Treasuries. So, as long as the US Treasury is the backbone of the financial system of the world, I tend not to worry about the deficit too much, although I'd like it to be smaller.
Okay.
If there was an alternative, we'd be in trouble.
But there is no alternative at this point.
10. Interest Rates
Let me ask you this. The Fed just raised rates for the first time in 3 years. Are rates likely to be a showstopper for the market, and in particular for the AI narrative? The rate to look at is the 10-year.
It's the long-term rates, not the short-term rates that the Fed controls, because that's what people borrow.
Okay.
You know, the Fed is just—
The Fed funds rate is the rate at which the Fed lends to banks overnight.
You don't have access to that. Neither do I.
Yeah.
So, just today, for example, the markets rallied because, despite the Fed raising rates, the 10-year yield went down. I'm getting the feeling that 5% is sort of the Rubicon for the market. As long as we're below that, we'll probably be okay. But if something were to happen and we blow through that, I think we get a correction.
Okay. Because it's around—
That's just my guess. I actually thought originally the number was 4.5%, and I was wrong.
Okay.
But 5% feels more significant.
Do you know where it is now? It's around there.
It's 4.98%.
4.98%. Okay.
But it was over 5% yesterday.
And it's come back down. People are watching every tick.
Yes.
11. Bessent's Buyback Plan
Interesting. I'm interested to hear Steve's thoughts on rising yields of long-term Treasuries. Is Scott Bessent's buyback plan really designed to increase liquidity in older long-term bonds, or is the government trying to manipulate long-term interest rates to ease its interest problem?
The latter. He's trying to buy long-term Treasuries to bring rates down and ease the cost of money for the United States of America. He's failed miserably at this point. Rates are higher than when he made his announcement. I have a suspicion that he's going to come with something else because he announced $6 billion. $6 billion is nothing. I mean, it's a $40 trillion deficit.
That was my thought.
I don't think he's an idiot, so I think he's going to come with something else. What else that is, I don't know.
Okay. So, a different plan of attack to do the same thing?
Yes.
12. Owning Precious Metals
Okay. I've always been skeptical about precious metals, but Steve's recent discussion with Porter Collins and Vincent Daniel made me second-guess that. I would like to know if he has any conflicting feelings about precious metals, or if he's still firmly opposed to the asset class.
I'm not opposed, but I've never owned it.
It's not something that has ever really enticed me one way or the other. Similar thinking to Warren Buffett: it just sits there and looks pretty.
Sits there. Does nothing.
Does nothing. It's not productive.
It has value because people say it has value.
That's all. It doesn't pay you an interest rate.
13. Bitcoin
Similar argument, I'm guessing, to Bitcoin and anything else that sits there.
Oh, definitely.
Baseball cards, whatever sits there.
Well, Bitcoin is worse.
Bitcoin is worse. Yeah.
Bitcoin is worse because it trades inversely to its own thesis.
Yes, I have noticed that. That is quite strange.
For your viewers, just to explain what I mean by that: if you went to a Bitcoiner—
—and you said, “Dude, why do you want Bitcoin?”
The answer you would invariably get is that fiat currency, which is government currency, has been debased, inflation is coming, and you want to hedge against this, so buy Bitcoin.
Okay, that sounds reasonable.
The problem is that if that were the case, on days when people worry about inflation, rates are going up, and the stock market goes down, Bitcoin should be up. On days when the Nasdaq is up like crazy and rates are down, Bitcoin should be down. But it does the opposite.
Yes.
You say that Bitcoin is going to go up when the sky is blue, and then the sky is blue and it goes down. Why do I own it? You have no thesis. It seems to be just a speculative asset.
It's just a way to speculate about speculating.
“Speculate about speculating.” I like that.
14. Investing Outside the US
All right, let me ask you this. Dear Steve, you seem to be very much centered on the U.S. stock market. Have you ever tried to broaden your investing or trading horizon geographically? Do you have any interest in companies held outside the U.S.?
Great question. When I used to run my hedge funds, I used to invest overseas.
Mm-hmm.
And I found that there was no night and there was no day.
Okay.
So, for the last many years, all I do is the U.S., and I'm perfectly happy. There's plenty to do in the U.S. I don't feel the need to invest overseas.
Fair enough. Plenty of opportunities here, I like it.
What's his view on when all these wars would end? I guess particularly the Iran war. Do you see inflation coming down anytime soon?
I have no more insight into the war than anyone else, so I can't answer the question.
15. The US Taking its Medicine
Fair enough. Well, I think that is just about all we've got. Is there one more? Is there one or a set of numerical indicators, whether related to interest rates, inflation, unemployment, or otherwise, that Steve could see as being the tipping point for the U.S. to take its medicine—cutting benefits or hiking rates up, or whatever he thinks that medicine might be?
I've seen it said that oil will rise until stocks fall, meaning that the U.S. won't get out of Iran until stocks really take a beating. But this is a much broader question about getting the debt back well under control. I guess we kind of touched on that.
I mean, all I would say is that's a total political question, and there's no political appetite in Washington right now to cut the deficit by a dollar from either side.
We'll see where that goes. I think that's just about how we finished our last interview. I was like, “On that cheery note.” So, again, on that cheery note, Steve, thank you very much for coming on.
For those that don't know, we talked about it earlier, but The Real Eisman Playbook is what you're currently working on. That's your podcast. Can you tell us a little bit more? What can people expect from that? It's on YouTube. I guess you can get it on podcast platforms as well.
Well, we do two free podcasts a week. One is an interview, and the other one is a market wrap where on Friday I put out a summary of the whole week. If you're willing to pay for the paywall on Substack, we do an additional podcast, which is sometimes an interview. This week was part one of a two-part master lecture of how to analyze banks.
All right.
But you have to subscribe.
Okay, I will subscribe. [laughter] No free lunch.
No free lunch there city here.
I will see.
Yeah, but if you want to know how to analyze banks, go there. That's where you should go.
Awesome, Steve. Thank you very much. Appreciate it.
Great. Bye.
Yes, you're the man now, dog.