Episode summary
The AI All-You-Can-Eat Buffet Is Ending with Gary Marcus | The Real Eisman Playbook Ep 62
The Real Eisman Playbook
· · Hosts: Steve Eisman · Guests: Gary Marcus
Summary
Steve Eisman argues that the AI trade has spread far beyond Nvidia into a whole ecosystem of hyperscalers, chipmakers, networking vendors, utilities, and industrials, but he says the economics are getting harder to justify. He points to massive capex plans from Google, Amazon, Meta, Microsoft, Oracle, and others, while warning that much of the current story depends on a small number of private model providers and on data centers that may face political, power, and water constraints.
Gary Marcus says the core LLM story has always been overstated. He traces the field’s problems to distribution shift, hallucinations, and weak generalization, and argues that simply scaling models has stopped delivering the expected gains. In his view, GPT5 was disappointing, the industry is increasingly leaning on tools, harnesses, and other symbolic techniques, and the most genuine progress today is in narrow, formally verifiable domains like coding and math.
The discussion turns to token economics, where Marcus says inference is expensive because each output token consumes substantial compute in the cloud. He says agents can be especially costly, and that subscription pricing looks unsustainable if users are forced to pay the true marginal cost. Eisman frames this as a potential bubble trigger: if pricing shifts and customers reject it, or if the circular dependence between hyperscalers, OpenAI, Anthropic, and venture funding breaks, the whole AI narrative could slow sharply.
Key takeaways
- Eisman says the AI trade now extends from Nvidia into hyperscalers, networking, CPUs, power, and industrials, but the spending binge is becoming harder to defend.
- Marcus argues that LLMs still hallucinate because they are next-word predictors, not fact-based reasoners, and that this weakness has never been solved.
- Marcus says scaling stopped being the main source of progress around December 2024, with tools, harnesses, and symbolic methods doing more of the work now.
- Marcus says coding and mathematics are the clearest current use cases because they can be formally verified and improved with synthetic data.
- Marcus warns that token pricing could expose how costly agents are, making the consumer all-you-can-eat model hard to sustain.
- Eisman says the AI ecosystem is circular and vulnerable because future hyperscaler revenue may depend heavily on unprofitable private companies like OpenAI and Anthropic.
Notable quotes
“the AI story from a markets perspective began on May 25th, 2023, which was when Nvidia reported its first quarter numbers for 2023”
“In fact, in 2001, I pointed out the problem of hallucinations. And that has never been solved”
“scaling isn't really carrying the progress anymore”
“Coding good. Not perfect, but good”
How the episode opens
Hey, this is Steve Eisman and today we're going to have a interesting interview with Gary Marcus about AI. But before we get to the interview, I want to talk for a while and kind of lay the groundwork because this is really the not the issue duour, it's the issue of the age. So I'm going to do an intro on the entire AI landscape.
Think of it kind of as a mini master class on all of AI, who's benefited, who's gotten hurt, and what are the risks involved, and whether it's a bubble or not. You know, it's unusual in terms of markets where you can actually point to a date where a story actually began. And the AI story from a markets perspective began on May 25th, 2023, which was when Nvidia reported its first quarter numbers for 2023.
The actual results were very strong. That's not what was important was what was important was that Nvidia predicted second quarter revenue of 11 billion which was 50% higher than the expectations. And as a result on that day, the stock was up 24%.
Which for a market cap of that size was pretty astonishing. And that began the AI story. As you all know, Nvidia manufactures GPUs.
But the ramifications for what Nvidia was talking about with respect to AI were far broader than just itself GPU chips. So the first group that everybody focused on were the hyperscalers. Meta, Google, Amazon, Microsoft, Oracle.
These were the companies that were buying the chips from Nvidia and creating data centers for the future of AI. And the amount of money that is being spent is astonishing. Just as an example, Google spent $90 billion in 2025 and is going to spend $180 billion in 2026.
Amazon is going to spend over 220 billion.
From the episode’s captions: roughly the first two minutes, lightly cleaned. Keep watching on YouTube →
Topics
- ai industry
- llm hallucinations
- scaling limits
- token pricing economics
- ai spending and outlook
Watch the episode
Summary, takeaways and quote selection by Octagon. Quotes are short excerpts; the full conversation is in the episode.
More from The Real Eisman Playbook
-
The AI Debate is No Longer Boom vs Bust: Why It’s Time to Go Back to Stock Picking | The Weekly Wrap -
The Midterms Have Changed the AI Narrative: S&P Down 4.4%, Nasdaq Down 5.5% | The Weekly Wrap -
Why Young Americans Feel Locked Out with the Lemonade Stand Podcast | The Real Eisman Playbook Ep 82 -
Anthropic’s Growth is Slowing: Is AI Capex Next? | The Weekly Wrap -
Inflation, The Deficit & Why the Dollar Stays King with Benn Steil | The Real Eisman Playbook Ep 81 -
AI Narrative Switched, Debt Yields Spike | The Weekly Wrap
The Real Eisman Playbook, this episode and its artwork belong to Steve Eisman. Octagon is not affiliated with the show; this page is an independent summary.
