Ed Zitron

The Stock Market Has An OpenAI & Anthropic Problem

• Top takeaway The guest argues future hyperscaler revenue growth is contingent on OpenAI and Anthropic's ability to spend massively on compute - over $200 billion in 2027 alone, and possibly $300-400 billion beyond that
• The guest believes OpenAI and Anthropic would need to be the size of Microsoft and Google by 2030 and become profitable to sustain the current setup - and even then, capacity is overbuilt by roughly 10x
• The guest estimates markets could be cut by at least 50% if the AI compute thesis breaks
• The guest sees a cataclysmic chain: less GPU spending hurts Nvidia, which in turn hurts ODMs like Foxconn and the Taiwanese stock market, with knock-on effects on US markets
• The guest views current AI spending as essentially a 'kick the can' exercise to keep valuations aloft

Ed Zitron argues Microsoft, Google, Amazon revenue growth increasingly depends on OpenAI and Anthropic spending $200B+ on compute. Analyst estimates cited point to 76% of Microsoft's revenue growth through 2028 tied to OpenAI compute spend. UBS estimates 48% of Google Cloud revenue by 2027 could come from OpenAI and Anthropic. Slowed AI capex, Ed warns, cascades from hyperscalers through Nvidia to Taiwanese ODMs and exposed Japanese banks. Ed argues AI need not fail technologically — overbuilt capacity of roughly 10x could trigger a 50% market drawdown.

Is A.I. A Massive Lie? | Ed Zitron

• Top takeaway Ed Zitron argues LLMs are not actually intelligent, pointing to instances like Bloomberg's AskB platform confidently returning an obviously wrong Microsoft share price as proof that subtle errors make them unfit for financial work.
• He maintains the AI investment cycle is sustained by circular financing, with hyperscalers, Nvidia, SoftBank, and venture capital repeatedly recycling money into the same two loss-making compute buyers (Anthropic and OpenAI).
• Zitron predicts a 50% or larger revaluation of AI-exposed equities and warns the aftermath could be worse than the dot-com bust because unfinished data centers still carry ongoing electricity and completion costs.
• He argues that any productivity benefit from AI is hard to verify, that coding metrics like commits are essentially meaningless, and that engineer-hours spent babysitting LLMs often exceed the value produced.
• Zitron frames the cultural pressure to adopt AI as a kind of corporate religion that CEOs embrace to appear forward-looking, drawing a sharp contrast with the begrudging adoption of prior mandates like DEI.

Ed Zitron frames the AI boom as a speculative bubble — roughly 70%+ of hyperscaler AI revenue is tied to OpenAI and Anthropic's circular compute spending. He argues LLMs are fundamentally probabilistic and can't reliably catch their own errors, so trillion-dollar productivity gains won't materialize. A reckoning could cut AI-exposed names 50%+ with cascading damage to venture capital, private credit, and Japanese lenders exposed to OpenAI bets. The dot-com analogy misleads: AI data centers are unfinished assets needing continuous capital, not already-paid 1990s-style fiber. Job-displacement claims are overstated; AI mostly erodes work that was already being automated, like transcription and contract roles.