Big Short legend Steve Eisman says everyone is buying the wrong AI stocks

Steve Eisman argues SpaceX is valued at over 100x revenue despite revenue of roughly $19 billion, comparable to Kellogg’s/Ferrero’s cereal business at about $21 billion. He criticizes SpaceX’s AI-linked ambitions, including a $28.5 trillion TAM and asteroid mining plans, as unrealistic and says the company may need ongoing capital raises to fund capex. He is also negative on the broader AI hyperscaler trade, preferring chip and networking suppliers over the large platform companies.

Analysis

The key market takeaway is not the rhetorical shot at one stock, but the signal that the AI trade is splitting into two separate ecosystems: monetizers and suppliers. If capex intensity keeps rising while end-user applications remain commoditized, incremental economics migrate away from software names with weak pricing power and toward the bottleneck layer—GPU, networking, optical, power, and rack infrastructure. That argues for continued relative strength in the picks-and-shovels complex even if headline AI enthusiasm cools.

The vulnerable names are the ones relying on narrative-driven multiple expansion without clear evidence of durable monetization. Hyperscaler AI spend can support revenue growth for a while, but if the market begins to focus on cash conversion and depreciation drag, multiple compression can happen faster than estimates come down. The second-order loser is not just the large-platform software stack; it is also the supply chain tied to hyperscaler budgets if the market starts discounting a longer payback period and lower ROI on AI inference and consumer AI products.

The contrarian angle is that the market may already be partially positioned for “AI winners vs AI losers,” but not fully for a capex plateau scenario. If enterprise customers resist AI price increases and model differentiation remains weak, spending growth could decelerate sharply in 6-12 months, hitting software names first and then flowing back into semis and network hardware with a lag. The most crowded risk is that shorts in the obvious narrative names get squeezed before fundamentals matter, so timing matters more than thesis validity.

Near term, this is more a relative-value setup than a clean directional short basket. The cleanest expression is long infrastructure beneficiaries versus short the most expensive AI-platform proxies, with options preferred where borrow or squeeze risk is high. The biggest catalyst to watch is capex guidance from hyperscalers over the next two earnings cycles; if budgets are reaffirmed, the trade works, but if management commentary shifts to optimization and ROI discipline, the multiple reset can be swift.

Key Decisions for Investors

• Long NVDA / short PLTR on a 3-6 month horizon: best expression of 'picks-and-shovels vs narrative software'; target 15-20% relative upside if AI capex remains elevated, stop if PLTR’s monetization metrics inflect materially.
• Long CSCO or a basket of NVDA+CSCO vs short MSFT or GOOGL, 6-12 months: play the widening gap between infrastructure demand and AI application pricing pressure; expect 8-12% relative outperformance if cloud capex growth stays above 20%.
• Buy call spreads on NVDA into next hyperscaler earnings season: use upside convexity to capture any reaffirmation of AI spending, with defined risk if capex commentary softens.
• Avoid initiating fresh shorts in TSLA, META, or ORCL until after the next two major capex prints: these names can stay expensive through multiple quarters, but they are the most exposed if the market shifts from growth-at-any-price to ROI scrutiny.
• If you want a hedged implementation, pair long NVDA with short MSFT: this isolates the supplier advantage while expressing concern that platform-layer economics will compress as AI becomes more competitive and capital intensive.