Summary
Altimeter’s Brad Gerstner delivers a rapid-fire market and tech check on the All-In Podcast stage, walking through why he believes claims of an AI bubble are overstated. He cites Nvidia trading at 14 times next year’s earnings, Anthropic’s revenue climbing from $2 billion in January to $11 billion by March, and Dell’s stock rising fivefold in 18 months as evidence that this cycle is earnings-driven rather than multiple-expansion-driven like the dot-com era.
Gerstner digs into the capex-versus-revenue question that has dogged the industry, referencing his October conversation with Sam Altman about OpenAI’s trillion-dollar compute commitments against modest gap revenue at the time. He presents data from Semi Analysis’ Dylan Patel showing the US is set to add 43 gigawatts of compute next year, with over half of total US compute eventually controlled by just two labs by 2028.
He also addresses the total addressable market question, arguing the knowledge-work TAM is large enough that AI companies only need to capture about 4% of it to justify current infrastructure spending, pointing to 47 quadrillion tokens produced this year and 40x growth in Codex users over eight months. The talk offers a numbers-heavy framework for evaluating whether the AI infrastructure buildout is sustainable.
📺 Source: All-In Podcast · Published September 17, 2026
🏷️ Format: Podcast
