Plain English with Derek Thompson "Yes, AI Is a Bubble. There Is No Question."
470 snips
Mar 17, 2026 
Paul Kedrosky, investor and veteran tech-market commentator, makes the case that AI is unmistakably a bubble. He digs into overbuilt chips and data centers, why hyperscalers keep spending, and why markets stopped cheering AI capex. The conversation also explores AI agents, token-hungry coding tools, pressure on software moats, Nvidia’s inference challenge, and the surprising energy boom behind data centers.
AI Snips
Chapters
Books
Transcript
Episode notes
The Railroad Analogy Separates Usefulness From Valuation
- Railroads show that a technology can be transformative and still be massively overbuilt, leaving half the peak-era track later abandoned.
- Paul Kedrosky says the buildout also triggered repeated financial crashes, proving useful infrastructure can still produce years of carnage.
Why Hyperscalers Keep Spending After The Market Turns
- Paul Kedrosky says markets have flipped from rewarding AI capex to penalizing it, turning hyperscaler spending from a valuation tailwind into a drag.
- He argues companies keep spending anyway because quitting first would signal surrender in the race to become one of the eventual oligopolists.
AI Spending Is Starting To Crowd Out Financial Cushion
- AI capex now consumes so much free cash flow that hyperscalers cannot fund buybacks the way they once did, exposing investors to more dilution.
- Paul Kedrosky says debt and private credit increasingly back this system while SaaS weakness and data-center exposure pressure firms like Blue Owl and Blackstone.
