Title: "Big Short" Investor: The AI Narrative is Falling Apart Show: New Money (YouTube) — clip commentary built around Steve Eisman's CNBC remarks; host Brandon van der Kolk Guest: Steve Eisman appears ONLY in the quoted clips (the ">>"-delimited passages). Everything else is the host's narration. Date: 2026-08-15 URL: https://youtu.be/Bz2oWtHFNr4 Length: ~11:30 Note: THIRD-PARTY COMMENTARY VIDEO — this is not an Eisman-hosted episode and not an interview with him. Only the ">>"-marked clip passages are Eisman's own words; the surrounding narration (the Artificial Analysis token-cost table, the Microsoft 45%-of-backlog figure, the Vanguard corporate-bond-issuance figure, the Phil Town workshop plug) is the HOST's, and is treated as context on the analysis page, never as an Eisman stance. Remove-only cleanup per skill Step 1: [music] artifacts and pure fillers ("um", "uh", "you know" as interjection, contentless "kind of"/"sort of") deleted; stutters collapsed ("the the the fee"→"the fee", "they they own"→"they own", "I don't I don't know"→"I don't know"). No words changed, added, reordered or paraphrased; every (mm:ss) cue is preserved exactly where it was. Auto-transcript garbles LEFT INTACT here and corrected only on the analysis page: "Deep Six V4 flash"=DeepSeek V4 Flash · "Kimmy K3"=Kimi K3 (Moonshot AI) · "ChatGPT 5.6 sole"=GPT-5.6 · "Claude's Fable 5"=Claude Fable 5 · "motes"=moats · "Open AI"=OpenAI. =====
00:00 What happens if AI doesn't succeed? >> I think we have a big correction. It's all one trade. >> Over the past few years, the investing world has been pretty crazy about the AI companies. So much so the big LLM providers like OpenAI and Anthropic are now looking to cash in with big IPOs. But is the AI narrative actually starting to crack? Well, Steve Eisman, an incredibly smart investor and one who is particularly talented at spotting bubbles, certainly seems to think so.
00:26 So much so he's getting out of his AI plays. >> I sold my Google. I wanted to reduce my exposure to AI. What scares me is that it's all one trade. So it better succeed. >> But what is he actually saying that the rest of the market is only just waking up to? Well, there are three dominos that Steve puts forward.
00:46 So in this video, let's look at how he's seeing these start to crack the AI narrative as well as the number one indicator he thinks we need to be watching to ensure that we don't get caught up in a big AI correction. Also, quick reminder that the three-day live investing workshop that I do in Atlanta with the one and only Phil Town is once again back in September.
01:05 It's on the 18th to the 20th of September. It is completely free to attend. However, it is still ticketed. So be sure to secure your ticket via the links in the description and the pinned comment. It's a fantastic three-day education on the Warren Buffett investment strategy. You'll also come away knowing how to implement option strategies in your investing, too.
01:20 Spaces are limited to just 300 though. So please secure your ticket today and I'll see you in September. So what are these three dominos that Steve has spotted that are causing the AI narrative to crack at the moment? Well, the first comes right out of the Warren Buffett playbook, but we do need a little bit of context first.
01:37 So when Wall Street talks about AI, we're normally referring to one of two different types of businesses. There's hyperscalers and then there's LLM providers. Now, hyperscalers are companies like Google or Microsoft or Amazon or Oracle that are building the data centers. So they're essentially building the factories.
01:54 Whereas the LLM providers like OpenAI with ChatGPT or Anthropic with Claude, they're building the AI models that use these hyperscalers data centers to run. Have a listen to how Steve explains it. >> There is a major difference between hyperscalers and LLM providers. The hyperscalers are the huge tech companies that are building the data centers where the LLM models are being housed.
02:16 There is overlap between LLMs and hyperscalers. Google and Microsoft are hyperscalers, but they also created their own LLM models. >> So, you have Google and Microsoft, Amazon, and Oracle, they're predominantly in the hyperscaler column, and then you have Open AI and Anthropic in the LLM column. However, as Steve said, a few of the hyperscalers, like Google, like Microsoft, are also trying their hand at the LLM side as well.
02:39 Now, why does this context matter? Well, it goes back to the gold rush analogy. During the gold rush, the people that made all the money were those selling the picks and the shovels, not the gold miners themselves. Now, in an AI context, obviously, the ultimate pick and shovel merchant is Nvidia, who is supplying the hardware to the hyperscalers, but next in line are the hyperscalers that are building the data centers for future AI use.
03:02 Then, in this example, it's the LLM providers that are the true gold miners. They could strike gold, but the business is much more risky. And this is Steve's first potential domino, the fact that the LLM providers don't really have moats. In fact, right now, they're in fierce competition with one another, and they're now in competition with drastically cheaper Chinese providers like DeepSeek.
03:23 >> The large LLM providers, Anthropic and Open AI, and partially Google and Microsoft, are much more problematic. The debate has really shifted because there just don't seem to be any moats, or at best, the moats are shallow. Enterprises are switching between models and using cheaper open-source Chinese models in order to control costs.
03:44 The future for these large LLM providers is very questionable. The Chinese models are much cheaper, and this could eventually cause a price war. >> So, this is the core problem that is the first domino in Steve's chain. And to show you just how cheap these Chinese providers are, I want to show you this work from AI research firm Artificial Analysis.
04:04 They found that Deep Six V4 flash charges 14 cents per million input tokens and 28 cents per million output tokens. So, from their testing they estimated V4 flash's average cost at 3 cents per test compared with 86 cents for Moonshot AI's Kimmy K3 versus $1.86 for OpenAI's ChatGPT 5.6 sole and $3.15 for Claude's Fable 5.
04:26 So, there is a big difference in cost there and it makes sense that Steve is concerned about a price war, which wouldn't be good for either OpenAI or Anthropic. But, this is where the second domino comes into play. And that is the fact that a lot of the hyperscaler backlog is actually from the LLM providers. AKA, if something were to hit the LLM providers, something like a big price war, it would likely have a pretty big flow-on effect into the hyperscalers.
04:53 This is Steve talking about the LLM providers, both OpenAI and Anthropic. >> If the lack of moats begins to cause them problems, then the entire AI ecosystem could go through a correction phase because so much of the hyperscaler backlogs are from these two companies. For example, of Oracle's 600-plus billion backlog, around half is from OpenAI.
05:15 >> And it doesn't stop there, either. Microsoft announced on their Q2 earnings call that 45% of their backlog is just from OpenAI. That's around 280 billion. We also know Anthropic is spending huge amounts on compute with Amazon and Google, although neither have officially stated how much backlog is just from Anthropic.
05:32 And then, of course, there's Steve's example. Oracle has around $640 billion in backlog, of which 300 billion is just OpenAI. The point here is we're talking hundreds and hundreds of billions of dollars of expected future spending into hyperscalers from the LLM providers. Future spending the market has already baked into the share prices.
05:54 But if for some reason OpenAI or Anthropic aren't able to follow through on that future spending, then that could cause shockwaves through the entire AI ecosystem. And that leads us straight to Steve's third domino. The fact that so much upward momentum in the market right now is effectively all the same AI trade.
06:15 But it goes deeper than you think. >> It's all one trade. It's literally one trade. Even people who think they're diversified because they own 60% stocks and 40% bonds are missing the fact that they're actually not diversified because of their 60% more than 50% of it is tech and AI related. And of the 40% of bonds, most of the new issuance of bonds is AI related.
06:38 Even people who think they're diversified because they own bonds are not really that diversified. What scares me is that it's all one trade. So it better succeed. >> And this is where we start to talk about the intermingling of different parts of the AI ecosystem. And some parts of it are a little bit reminiscent of the global financial crisis where Steve made his original fortune.
06:59 Remember a scene from the movie, you have parts of CDOA and CDOB and parts of CDOB and CDOA and then they both get put into CDOC. Now, granted, it's not exactly the same when it gets applied to AI. But we have that kind of situation forming where everyone is in cahoots with everyone else. Remember this chart that went around last year? Well, these days it's much more complicated than that.
07:19 Plus as Steve notes in that last clip, you've got all the world's big ETFs exposed due to these companies all up at the pointy end of the major market indexes. And even if you retreat to something like corporate bonds, well, you're still not escaping the issue as a lot of the bonds issued today are AI related.
07:35 It's not as bad as tech concentration in ETFs, but even still Vanguard says around 10 to 15% of all US corporate bond issuances this year they'll be related to tech, which is still around $400 billion worth of bonds. So, the last domino in this chain is that we all have large exposure to this theme, whether we like it or not. Doesn't matter if you hold ETFs, doesn't really matter if you hold bonds.
07:55 If something goes wrong with AI, we're all going to feel it. So, those are Steve's three dominoes that could potentially trigger a correction across the AI ecosystem. But, how do we know if the cracks are starting to appear? Well, according to Steve, there is one key indicator that investors should be watching above all else, and it is the financial health of OpenAI and Anthropic.
08:16 >> A key thing to monitor to determine a catalyst for a real sustained sell-off is the health of Anthropic and OpenAI. >> Now, of course, this is quite challenging for us to do at the current time because these companies are both still private. But, what do you know, both of these LLM providers are currently looking to go public.
08:32 And when they do, their SEC filings will be available for us all to scrutinize. So, I know I'm definitely going to be watching these S-1 filings when they eventually hit us, and it's probably a good idea to pay close attention to the hyperscalers quarterly reports as well, and particularly what they say in their earnings calls.
08:50 But, with that said, for the last part of the video, I also wanted to talk about how Steve is looking at investing in the AI narrative at the moment. Now, we heard at the start of the video that he's personally selling out right now. But, if you're interested in his broader philosophy about buying into the AI stocks, this is it. >> Anyone who thinks they can confidently predict the ultimate outcome for AI is just kidding themselves.
09:12 The story is moving too quickly. The facts change weekly, and I really don't know where this will all end up. But, here is where I think we are now. Despite the fact that the hyperscalers have become incredibly capital-intensive businesses, they do have businesses that have some level of motes. Anyone who wants to do anything with AI, whether it is an LLM model or an agentic AI or something else, will have to house it with a hyperscaler and they're only going to be a few hyperscalers.
09:42 The amount of money it takes to be a hyperscaler is insane and that expenditure itself is a moat. So, the hyperscalers like Google, Amazon, Microsoft, and Oracle have real businesses here. What the returns will look like, I don't know yet, but they have real businesses. >> So, Steve is much more inclined to look at the hyperscalers if you're going to play at all and it really comes down to three main reasons.
10:04 Number one, there's a stronger argument for these hyperscalers having moats. Plus two, the hyperscalers, they do have diversified business models, so it kind of protects them. And then three, ultimately, that gives them the financial strength to play the high capex game without too much downside risk through what is quite an uncertain time.
10:24 But on the other hand, the companies he's less bullish on are the LLM providers, specifically the ones that exclusively offer LLMs. So, we're talking about Open AI and Anthropic. >> Anthropic and Open AI are also problematic because they don't have the breadth of revenue streams of Google and Microsoft. Google and Microsoft have multiple revenue streams from established businesses which are very unlikely to simply disappear.
10:46 They also have hyperscaler businesses to balance their vulnerability, but their LLM businesses are also questionable. >> So, that's Steve's thinking in a nutshell. The hyperscalers are much more protected from the potential downsides than the LLM providers. But as he says, it's also worth remembering that nobody knows who the long-term winners will be in AI.
11:06 So, you just have to be careful with how you play. But with that said, I hope you found this video useful. Remember to get your free ticket to the upcoming live workshop and I will see you in Peachtree City on the 18th of September with Phil. As I said, spaces are starting to fill up now, so please, if you're interested in coming along, I'd love to see you there.
11:23 Get your tickets today via the links in the description and the pinned comment. But with that said, I'll see you all in the next video.