Eisman: The real AI risk is OpenAI, not chip depreciation
Key points
Key takeaways from a New Money interview with Steve Eisman, former portfolio manager, Q4 2026.
Concentration risk. Eisman says seventy percent of hyperscaler AI revenue comes from just OpenAI plus Anthropic. That equals twenty-five to thirty percent of total cloud revenue. If either company gets into trouble, everyone downstream is affected.
Nvidia receivables. Seventy percent of Nvidia accounts receivable at the end of July came from just five customers. Revenue still grew more than one hundred percent.
Oracle exposure. Oracle's remaining performance obligations jumped from roughly $150bn to roughly $400bn within three months. Roughly half was attributable to OpenAI alone. Its stock has since fallen from around $330 to around $150.
OpenAI is the weaker of the two. Per Wall Street Journal reporting, OpenAI booked roughly $6.5bn revenue for Q2 2026 against roughly $12bn in costs. Revenue rose $1bn while costs climbed $3bn. Revenue grew eighteen percent over three months, while Anthropic grew more than one hundred percent over the same period.
Burry depreciation thesis rated too academic. Eisman pushes back on Michael Burry, arguing that stretched AI-server useful lives barely matter if OpenAI plus Anthropic keep growing. The far bigger risk is OpenAI failing. That would reverse the whole spending chain into a massive correction unrelated to depreciation.
Circular financing. Most funding for the pair comes from equity stakes taken by Nvidia, Amazon, Google, Microsoft, SoftBank. Money flows one way, commitments come back the other. The chain holds only while the two keep growing, with investors continuing to fund them.
No durable moats among LLM providers. Enterprises have grown cost-conscious after blowing AI budgets. Open-weight models such as DeepSeek plus Kimi K3 took significant share of usage. LLM providers rotate atop the rankings without switching costs. Hyperscalers have scale moats yet depend on the very LLM providers they host. Google raised $85bn equity, marking its first capital raise since going public.
Doomsday talk is a complete subterfuge. Eisman argues the existential-risk narrative is nervous fear-mongering. Companies hope regulation will ban Chinese models, handing a legal duopoly-moat to incumbents. If Anthropic CEO Dario Amodei genuinely believed his product dangerous, he would postpone his IPO. Eisman takes people seriously when they put their money at risk.
Read more: New Money
