What's an LLM? It's just a fancy word. It's a fancy word retrieval system. It predicts the next word in the chain. And it's great. I'm not saying it's... but that's not thinking.
Yeah. Why is it going to kill us? That seems...
Well, it has to be sentient. It's not sentient. It doesn't think, right? It just retrieves words or creates software, which is another form of words.
So my thesis as to why they've done this is that Dario Amodei and Sam Altman right now are petrified. They are petrified because token maxing is done. So for those of you who don't know, token maxing was basically people who ran software and enterprises were ordered by their companies to use AI 24/7, right?
The hell with the cost, and then the cost, the price went up, and Uber went through its entire AI budget in three months, right?
Was that this year or last year? I can't remember. This year was this year.
It was in the spring. So token maxing is done. You've got open-weight models that are a tenth the price. They're not as good as the models, but you know, you don't need a Mercedes-Benz for every road trip, right?
So they're losing market share. There's pretty good evidence that a price war is starting, and they're petrified. And so they created this: the world's going to end unless you regulate us, because they think if they get regulated, they can manipulate the regulation, or they could stop the open-weight models from taking too much share. They could manipulate the regulation to keep the Chinese open-weight models out, and so they could have a duopoly. And the basic problem with the entire AI enterprise is that there are no moats.
Absolutely none. One model is good one day, and then people switch to the next. It's not like Google. Everybody was on Google forever. You were on Google forever so much, the concept of not being on Google, you couldn't even think that. Today you use Claude one day, you don't like that, you go to Gemini, you don't like that, you use it for a while, you go to ChatGPT. There are no moats. People are spending a trillion dollars.
Are you surprised by that? That how easy it is for people to switch from one to another?
I'm not surprised or surprised. It is what it is.
Right. Okay. Because I always thought the one that would win was the one that remembers the most.
Or the one that could recall the most about you every time you use it.
But they're all like that.
Which one? What's your favorite LLM for the work that you do?
I don't have one. For my purposes, it's just research. They're all fine.
They're all okay. You don't have one that you're... So we standardized on Claude Enterprise for the firm.
And what ends up happening is like everything I used to do in ChatGPT, now I just do it in Claude because we're paying for it anyway, and you just get used to it.
I sort of think the enterprise LLM is a sneaky way to build a moat because you get all the workers of the Fortune 500 accustomed to using it, and then why would they go use another?
Well, they'll use another if it's a lot cheaper. If an open-weight model can do more or less the same thing for one-tenth the price, they'll use the open-weight model.
That's the point. It's not that they won't use Claude. It's just I think that Claude's going to have to cut its price. I'm already hearing that it is cutting its price.
Okay, so then the founders of these AI labs are coming out and they're saying, 'Please regulate us.' I mean, they did a skit on Saturday Night Live, so it's now like in the mainstream that this is a thing, right? And you're saying like, 'Yeah, because they don't want any competition.' And so once you're big enough, you basically write the rules.
The statement that some of these guys have made that LLMs are going to destroy planet Earth is a lie. Plain and simple. It's just a lie.
Okay. All right. Let's start the show. John, I like it. Let's go. All right. We ready? We'll get it started. All right. My friends, episode 864.
This podcast is brought to you by VanEck. Look, we're in an era of rising debt, deglobalization, and a massive physical buildout to power the AI economy. Look, that means more demand for energy, raw materials, and infrastructure. Historically, that's been a really strong backdrop for real assets. And VanEck has an ETF for that. It's called RAX, the VanEck Real Assets ETF. It's actively managed and shifts exposure across gold, energy, infrastructure, and natural resources based on what the macro environment is actually doing, not just what it's done in the past. RAX is your one-stop shop for real assets. Head over to vaneck.com/compound to learn more. That's vaneck.com/compound.
Thank you, John. All right, guys. Welcome to The Compound and Friends. We are one of the best podcasts in the country. We're so happy to be able to tell you that our guest today also runs one of the best investing podcasts. I was listening to you yesterday, and it occurred to me now that some of your episodes are premium. I might have to... I think I have to pay to pay for it.
All right. Oh, by the way, before it's called The Real Eisman Playbook, just so we get that out there.
The real... Well, I mean, it seems like you don't need much help. The numbers on YouTube are huge.
Are you shocked by the success of the show so soon after launching? It's a big show.
Okay. But it's, I mean, it's cool.
All right, guys. Steve is an investor and the host of The Real Eisman Playbook, a weekly podcast he launched in 2025. Steve has gained fame for betting against the subprime mortgage market ahead of the 2008 financial crisis while running FrontPoint Partners, a trade that was chronicled in Michael Lewis's The Big Short. He later founded Emmeris Partners in 2012 and joined Neuberger Berman in 2014 as a managing director and portfolio manager. Live audience, Miami U. Round of applause for Steve Eisman. They love you already. They love you already.
I like it. You know, University of Miami is a really nice school. And it has a pool in the middle of the school, which is kind of nice.
I know. That's why my son looked around the campus.
Because your son is no fool.
You know, when I remember visiting my daughter at University of Miami, and what a wonderful time she had, both educationally and as just being there. I said to my wife Valerie, because Valerie and I both went to Penn, I said, 'We're idiots.' Like we went to this school in the middle of Philadelphia where we froze our butts off. We could have been in Miami.
What did she study at Miami?
My daughter studied art history.
Art history. Okay. All right. I want to ask you the question that I think is the only question that matters right now for the market. And we've had this plan to have you on for a while, so we didn't time it, but it's almost perfect timing. Is the funding mechanism for AI products going to break or not? We had this event in the market yesterday with CoreWeave. So, for people that don't know, CoreWeave is one of these things that's called a NeoCloud, and it's basically a data center. Not owned... they run data centers, right? Not owned by a hyperscaler. It doesn't have the cash flows coming from advertising like Alphabet, for example, right? The debt has shot up to, I think it's a 15% yield right now. The stock price is, I'm going to say, in a 60% drawdown.
It's bad. All right. So, what you're looking at here are spreads and yields blowing out on CoreWeave, just sort of as emblematic of all these companies that need to sell debt in order to fund these projects. Projects that you can't half-build and then just leave them. Like you've got to...
Well, you could, but you'd be bankrupt.
You could, but it would be a bad idea. I'd love to, as somebody that on a firsthand basis was in the midst of the great financial crisis and you had seen it coming, you had bet on it, I guess you bet everything on it at one point. To me, this doesn't look like that. But I'd love to get your take on the discourse around what's happening with bond yields and whether or not these projects can go forward.
Well, look, if you want to draw parallels to the great financial crisis, there's a big difference, which is, I mean, let's just hypothetically say General Motors goes bankrupt. So what happens? People get laid off. All the suppliers to General Motors, some of them go bankrupt. A lot of people lose their jobs, right? Almost certainly there's a recession in that part of the country. Maybe if there's contagion, there's a recession in the rest of the country. And that's what happens.
When JP Morgan goes down, planet Earth burns, right? That's the end.
If you can't, it's the end.
Yeah. So the reason why the government had to bail everybody out was you couldn't have a situation where people can't get their money out of the bank, right? This is more like the General Motors story.
I mean, it's a very big story.
What if you can't get your chips out of the data center? Then what?
Well, that's bad for you.
You know, but it's not the end of the world. I mean, we obviously will go more into the AI story, but if, just for example, OpenAI went bankrupt tomorrow, and I'm not making that prediction, but let's just say that it did, I think the US economy would go into a recession immediately. I do.
It's not the same. It's not the same thing as a great financial crisis. The banks are fine.
How did they become so systemically important? This is a non-public company. I know in the private market the valuation is a trillion, a trillion and a half, whatever it is. How did they become so systemically important in three years to the point where they could send the entire US economy into a recession?
They went out and made, they and Anthropic both went out and made commitments to basically everybody who would take their pen, right? And so, you know, if you want to think about the problem with the AI story, I'll give you the short version of it. The problem with the AI story is a concentration problem. Two giants, which is...
So Nvidia grew revenue this most recent quarter by over 100%. That's an astonishing statement. The largest company on planet Earth grew revenue by over 100%. It's awe-inspiring. But in the 10-Q there's a sentence that says that 70% of Nvidia's accounts receivable, which is a proxy for revenue, is from five customers.
You don't want such concentration. What happens if one of those customers goes down? Then you just lost a whole bunch of your business.
Yeah, of course. No, I'm sort of asking a dumb question. I'm just curious to hear your thoughts.
You want to have... I gave you an example. In the 90s when I was at Oppenheimer, we had a software analyst who was very talented, and I remember having a conversation with her where she said that Oppenheimer wanted to bring public a software company that had rapid growth but it only had two customers, and she said no. And the reason why she said no is if one of those customers goes away, that's the end of the company.
So Nvidia's five customers is basically everybody. Now, those are probably the hyperscalers.
Which themselves have hundreds of millions of customers.
Yes. But if you go to the hyperscalers, 70% of their AI revenue is from Anthropic and OpenAI, right? And then if that number is something like 25 to 30% of their total cloud revenue, right? And then Oracle is much worse because half of Oracle's backlog is just OpenAI, right? So the whole...
These are the commitments. So OpenAI says to Oracle, 'We're going to utilize your cloud services, we're going to utilize the compute that you're making available, and by the way, we need 10 times more. Go build all these data centers.' So all this building is really because Anthropic and OpenAI have gone crazy.
So we learned a few things yesterday. Number one, we had another hot inflation print.
And we had one this morning.
And we had another one this morning, and yet interestingly, we'll talk about later, rates are down a lot, which tells you where the pressure is going.
Because people are scared. Tells you where the pressure is coming from. But yesterday felt like a moment. CoreWeave's debt through the roof, right? They're now spending $640 million just to service the debt, which is extremely problematic given their cash flow situation. We had $2.1 trillion come out of the stocks over the past seven days. Yesterday, did yesterday feel like the bubble popping, or is this another moment to buy that we're going to look back on and say, 'You stupid sons of... everybody so eager.' Before you answer, take a look at this. This is $2.1 trillion lost in the stocks over seven days, and you can see how extreme that is on the screen right there. Like, that does not happen in a normal market.
Do you think that people are too eager to, and at this point there's a sea of people so it's too many, but to become the next Steve Eisman, to be the hero for calling the next big short? Like, I feel like...
You know, I got to tell you, it's a bit of a disease.
Because I said this on CNBC once, I'll say it again because it's such a good line that I really like it, which was a day when Ray Dalio, who I respect greatly, was basically predicting the end of the world because of the deficit.
He did it in 2015, and he's doing it again.
So I was on the 5:00 show, and Melissa Lee asked me what I thought, and I said, 'You know what the problem with all these people is? They all want to be me.' And I have news for them: the role of Steve Eisman is taken.
And it's funny, but the point is that people want to be known as the person who predicted the next calamity.
They want that glory. You know what? I had that glory. I don't need it again.
So I honestly don't know. I don't think CoreWeave is important enough to sink the AI story. I think Anthropic and OpenAI are important. Now, the couple of data points that were interesting about them is last week OpenAI came out and said that their annualized revenue at the end of September was $50 billion. A month ago they said 70. What happened to 20? Where'd the 20 go? Disappeared. And by the way, I had absolutely no idea what annualized revenue means. Nobody does.
That's not apples to apples, but...
And then today there was a report in the FT about Anthropic that said that its revenue, I'll be off by a little...
I got the numbers. It jumped 50%...
To $16.8 billion for the...
Right. So it jumped 43% in three months, but that's down from a growth rate in the second quarter of 140%.
Well, we know that's not sustainable, of course.
Well, yeah, but the point is I think the end of token maxing is starting to have an impact.
So they were making more money as a result of token maxing than they reasonably could have made on an ongoing basis once companies looked at their budgets and said, 'We obviously can't spend at this rate on AI, as great as the technology is.'
So one interesting wrinkle is that with rates where they are, the supply of financing for new data centers is obviously going to be harder to come by. Harder to justify the demand side of it, though, even though it might be changing a little bit with the open-weight models. There are still supply constraints all over the place. So if the supply gets limited and the demand continues to increase, that could make for an interesting dynamic of the marketplace.
It could. I mean, part of the problem right now is that there's a lot of pushback on data centers all over the country. It's very hard to quantify it.
I was about to make a bad-taste joke, but...
Yeah. Yeah. Fine. Yeah. But there's so much pushback that all this, the buildout of data centers seems to be going to slow by a lot.
Which would be bullish for people that sell compute. Less data center...
If it would be bullish because it would mean you're going to buy less compute because...
You can't build a data center. If you can't build a data center, you got no place to put your compute.
But what compute exists becomes more in demand.
That's probably the case.
Okay. So I actually wanted to ask you about that because I think that's part of the story. I don't think the demand changes for AI. I think like this group in front of us, they've just spent a couple days...
I agree. It's the price. It's the price of the AI that matters.
But so this idea, this group in front of us has just spent two days going into all the investment banks on Wall Street. Every one of them is going to have a substantially larger budget for AI and AI-related projects and AI-related hiring in '27 than they had in '26 and '25. Like that part of it at the enterprise level is not going to change. The other thing that probably won't change is consumer demand, which I think is mostly ChatGPT and Muse, and now Muse on the agentic side, and Gemini. Every Google search is now an AI de facto. So that's not going away either. Nobody that's using this technology is going to say, 'You know what, I don't really need this, or I haven't made this a part of my life, and I'm just going to go back to the way I was doing things in 2022.' I don't think any of that's really going to happen. So the price of the bonds, the difficulty in selling debt, is not going to really change the demand side. I think what changes is who gets to build the capacity to serve that demand. And arguably, if the CoreWeaves and the Nebiuses of the world are not able to sell debt and supply as much compute as they want to, the ball just goes back into the court of the hyperscalers. And Amazon, Alphabet, Microsoft, they're in an even stronger position than they were in a year ago when maybe debt was cheaper or there was a bigger appetite for bonds. What do you think about that idea from an investing standpoint? The hyperscalers' position just strengthens.
I agree that the hyperscaler position has strengthened. What I'm most focused on is this business is so concentrated from Anthropic and OpenAI that if there's a price war, none of that matters. The whole thing will correct. Now, if three, four years from now that happens, it probably won't matter because there'll be a lot of other companies that take the compute. But right now it's so concentrated that you better hope that those two companies do well.
It seems that the pricing is based on usage now more so than the last 15 years. The way enterprise software was sold was SaaS. You pay us per head. So such and such company has a thousand employees. Salesforce is going to take $80 per employee, and that's the SaaS contract. Now it's migrated more toward a pay-what-you-use model. That feels more sustainable to me than the way we thought they were going to sell these services.
Well, it's also a lot more expensive.
Yes, but I think that companies have more control over those, or maybe it's not that way.
Well, I think it's more expensive, and so companies are going to be more cost-conscious, and they're going to get their employees to use the cheaper models much more, which is going to put pressure on Anthropic and OpenAI to cut price.
Okay. And you think that's something that we're witnessing the beginning of?
I think it's potentially we're seeing the beginning of it.
Okay. What's the transmission mechanism by which OpenAI has a problem, things happen, and then we're in an economic recession, like economy-wide recession?
So, let's imagine that there's a full-blown price war.
Okay. Let's assume it. I don't know if it's going to happen, but let's just say that it does. Anthropic's revenue growth turns negative. Nobody's going to fund them. They got to cut back on their commitments. Oracle goes right down the toilet because half of Oracle's backlog is just from OpenAI.
By the way, its stock price is trading like...
Like it's in the toilet. Yes. Exactly.
So, just for those of you who don't know, last year, like around the third quarter before Oracle reported, the stock was 230, and then they reported and they showed that their backlog had gone to like 300 or I think 400 billion. It was up 300 billion in three months, and everybody went nuts, and the stock went from 230 to 330, like you remember this, like in two days. Like I never saw a stock go up 100 points like that.
One of the biggest stocks in the market.
And then I don't know who came out. It might have been Gil Luria from Davidson, put out a report to say that the entire 300 billion was just OpenAI, and everybody just freaked out.
And then the stock went from 330 to 230, and now it's at 130.
Because people don't trust...
They don't trust that you can't rely on a company that loses money. You've hinged your future on a company that bleeds cash like a lunatic.
Right.
This has been a very interesting couple of days in the stock market.
So, as I said yesterday, a hot inflation print, and it really felt like people were like, 'Oh, like this is... it's not getting...'
It's not getting better. We saw producer prices today. Yesterday was consumer. Today was the producer prices, also not good. Yesterday you had the memory stocks, which were the belle of the ball, killed. Fun fact: Western Digital, it's in a 54% drawdown.
54. Wow.
And Steve, it's still up 100% year to date.
Right. I've never heard of that—a stock getting cut in half and still being up double percent year to date. It's unbelievable.
Just to show you how crazy the run was.
So you had yields spike immediately yesterday. By the way, the person who bought it at the peak—that sentence doesn't help them.
Yeah. Doesn't help that it still has 100%.
You had yields spike and Anthropic—obviously it's not a publicly listed stock, but there is this blockchain-based company called Hyperliquid that does a lot of volume. Over the last 24 hours, Anthropic has traded $17.6 million worth of stock. And this thing—it's a market and it works pretty well. It worked very well for the SpaceX IPO. So this is Anthropic trading.
So this is Anthropic on Hyperliquid. I see.
So Anthropic was doing bad yesterday—obviously—and then the numbers came out after the bell of what their revenue was. It puked. It puked in the after hours. People are freaking out. Last night Twitter was black, like jet black. People are freaking out. And this morning...
