'Big Short' Investor Explains How the AI Bubble Will Burst

Steve Eisman analyzes the AI bubble through the lens of his experience identifying the 2008 housing crisis. He identifies critical structural problems in the AI ecosystem: (1) Extreme concentration risk—70% of Nvidia's accounts receivable comes from five customers, and 70% of hyperscaler AI revenue comes exclusively from OpenAI and Anthropic, meaning the entire chain from chip manufacturers to cloud providers depends on the viability of two unprofitable companies. (2) OpenAI and Anthropic's precarious financial position—OpenAI had $6.5 billion in revenue but $12 billion in costs in Q2 2024, with costs growing three times faster than revenue, while Anthropic shows better unit economics but both companies rely entirely on continued capital raises from tech giants. (3) Absence of competitive moats in LLM providers—unlike hyperscalers' infrastructure advantages, LLM providers face constant customer switching as users migrate between Claude, GPT, and emerging models like DeepSeek without loyalty. (4) Token pricing dynamics shifting unfavorably—token maxing (forced AI usage regardless of need) ended when companies realized the true costs, leading to price increases that reduced usage and increased reliance on cheaper open-weight models. (5) Hyperscaler cash flow deterioration—Microsoft, Google, and Amazon's traditionally strong cash generation has evaporated due to massive data center capex spending ($700 billion across hyperscalers this year), forcing Google to raise $85 billion in equity capital for the first time since going public. Eisman argues that the AI existential risk narrative is manufactured—companies are using AI safety concerns and regulation advocacy to create legal moats that don't exist naturally. He contrasts this with actual investments: management selling stock is a major red flag, cyclical businesses are unattractive, and he prefers story-driven companies with real growth tailwinds and embedded switching costs. For AI investing, he recommends picks-and-shovels plays (Nvidia, Micron, GE Vernova, Arista Networks) over LLM providers, given power infrastructure becoming the limiting constraint. He remains skeptical of predicting outcomes but notes that if OpenAI or Anthropic fail, the entire chain collapses and the U.S. would face recession-level impacts. He believes the tech debt and U.S. government deficit are concerning but not catastrophic due to Treasury's role as the global financial system backbone.