

This Is How the AI Bubble Could Burst
434 snips Sep 23, 2025
In this discussion, Paul Kedrosky, an investor and venture partner, explores the staggering $300-$400 billion AI infrastructure spending by American tech firms. He warns of a potential AI bubble, highlighting economic risks, from capital outflow in manufacturing to energy demands driving inflation. Kedrosky explains the complexities of financing data centers and the vulnerabilities they pose across various sectors. With intriguing insights into the political implications and real-world opportunities for AI, he underscores the urgency of addressing these looming risks.
AI Snips
Chapters
Transcript
Episode notes
AI CapEx Is Economically Massive
- U.S. tech firms will spend an unprecedented $300–$400B this year on AI infrastructure.
- That spending is historically large enough to shape GDP and markets despite limited near-term AI revenue.
GPUs Depreciate Fast
- GPUs and AI hardware depreciate quickly, with useful life around 2.5–3.5 years.
- That rapid decay makes AI CapEx more like perishable goods than long-lived infrastructure.
AI Sucks Capital From Other Industries
- Capital is being drawn into a narrow AI infrastructure cluster, crowding out other sectors.
- This creates higher hurdle rates for small manufacturers seeking funding.