Steve Eisman says AI hype is repricing stocks, and investors are buying the wrong picks

Steve Eisman is back with his favorite hobby: taking an industry consensus, squinting hard at the numbers, and then calling it nonsense. In a half-hour phone call, the Big Short legend skewered what he says is a haze of “hype and hysteria” around the AI trade, and his most pointed example was SpaceX. His punchline is unusually specific: “SpaceX has the revenues of Kellogg's, which makes Froot Loops,” Eisman said, adding that SpaceX is being valued at “over 100 times revenue,” while “no company of any size has ever had that kind of valuation.”

To anchor the comparison, the source notes Kellogg’s cereal business was bought by the Ferrero Group of Italy, and Ferrero’s 2025 revenues are $21 billion, which are indeed close to SpaceX’s $19 billion. Eisman also used another valuation datapoint to drive home the point: “By comparison, Palantir is at 50x.” In other words, this is not a vague complaint about “irrational exuberance.” It is a valuation argument: investors are paying a level of revenue multiple that, in his view, does not pass the basic sanity test.

Now, it’s easy to stop there and call it contrarian theater. But Eisman insists the SpaceX case is really about what investors are funding and what they are ignoring. He argues that Musk and Silicon Valley absorbed sci-fi tropes in their youth, but “the difference is, Musk and the SpaceX crew take it seriously.” In the same call, he highlights a specific element from SpaceX’s S-1 filing that leans into future optionality, including a line about asteroid mining: he points to “page 71 of the SpaceX S-1,” which states, “We plan to pursue asteroid mining operations to extract metals and other critical resources from near-Earth and main-belt asteroids, providing abundant raw materials for space-based industries.”

Eisman then pivots from the sci-fi premise to the market premise. He’s puzzled, he says, that Musk is building a conglomerate-like platform while the broader corporate world is “de-conglomerate-izing.” The reasoning is straightforward: investors often prefer “pure play” exposures to themes they can underwrite, and Eisman argues people want clean exposure in areas like “rockets, Starlink and AI,” not a wide bundle that makes it harder to model which part will win. This is the kind of complaint that hits boards and investors in the gut, because it challenges not just a valuation number, but the structure of what the market is rewarding.

Then Eisman drags the conversation into Tesla, which is where he turns from valuation math to what he calls track record. The source says the prospect that SpaceX will buy Tesla, Musk's second largest holding, is “especially appalling” to Eisman. He claims Tesla has been “a horrible failure for the past several years,” pointing to repeated promises that self-driving cars and robotaxis are coming “which he doesn't do,” and adds that “earnings go down year after year.” He also frames the behavior as market psychology: “Musk is a cult, so people keep saying 'wait till next year.'”

After that, the real crux of Eisman’s AI thesis shows up: he doesn’t just dislike the headline stock. He worries about the economics of the entire AI stack. The source says he highlights a part of SpaceX’s S-1 that functions like a manifesto betting the company’s future on AI, including the fact that its chief AI product is named Grok. Eisman connects the term to Robert Heinlein’s 1961 sci-fi novel “Stranger in a Strange Land,” where “grok” means “deep understanding,” then says the product lives up to that name poorly. According to Eisman, “Grok the product's a lightweight,” and he notes “Grok is a third-tier product.” He even adds a reported detail from the source’s account: “I've heard reports that even the engineers in Musk's own space division won't use it because it sucks.”

But the bigger point is the incentive structure behind AI companies. Eisman says the hyperscalers face a “darkening” outlook, driven by “two vectors.” The first is that for hyperscalers, AI is becoming “increasingly capital intensive.” He cites a specific budget shift at Alphabet: “Last year, Alphabet spent $80 billion on AI and funded it from cash flow. This year, it will spend $180 to $190 billion and raised $85 billion in stock.” Eisman’s framing is that when capital intensity rises, companies start leaning harder on equity issuance because the internal cash generation does not scale fast enough for the “table stakes.” For decision-makers, this matters because it affects dilution math, refinancing risk, and the willingness to keep pouring money into bets that do not yet have durable profits.

The second vector is his blunt claim that there are “no 'moats' in AI.” The source describes his logic: “Someone moves to ChatGPT then to Gemini then to Claude.” Even if AI is as transformative as he implies, providers of AI-facing software can get swapped quickly. So Eisman argues that the better economics come from being the infrastructure and equipment beneficiaries, the “suppliers selling them the picks and shovels, the chips and networking gear.” In his view, those products are “highly customized and protected,” unlike competing AI applications that can be easily replaced.

That is why he says it’s more attractive to ride the capex boom with picks-and-shovels names such as Nvidia, Arista, or Cisco rather than be exposed to AI application battles among hyperscalers and enterprise software giants like Meta, Oracle, Microsoft, or Alphabet. Importantly, he stresses he is not issuing a direct trading signal: “Eisman stresses that he's not advising anyone to short SpaceX. 'I have no opinion on what will happen to SpaceX,' he says. 'From a fundamental perspective, it's ridiculous. But a lot of things can be ridiculous for a long time.'”

So what’s the strategic stake for executives and boards? Eisman’s message is essentially about misaligned pricing versus uncertain fundamentals. When capital intensity ramps and switching costs are low, the market often overpays for the story and underprices the bottlenecks. If investors are indeed valuing high-revenue multiples in a way that ignores dilution pressure and commoditization risk, then the board-level question is not just “Is this company exciting?” It is “Where does the economic moat actually sit once the AI arms race turns into a hardware and infrastructure supply chain sprint?”