The Intangible Economy with Kai Wu

What Past Capital Cycles Can Teach Us About AI with Edward Chancellor

May 12, 2026
Guest
Edward Chancellor
Edward Chancellor, financial historian and author who studies capital cycles and bubbles. He explores how past manias like railways and dot-coms mirror today’s AI investment rush. Short, punchy takes cover why markets fund tech shifts but miss winners, the prisoner’s-dilemma of hyperscaler spending, GPU depreciation impacts, limits of large language models, and where anti-bubbles may hide value.
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
INSIGHT

How CapEx Booms Temporarily Inflate Profits

  • CapEx booms can temporarily inflate aggregate corporate profits because buyers delay depreciating newly acquired assets.
  • Chancellor explains late-1990s telecom spending raised reported profits until misallocated capital forced abrupt depreciation and profit collapse.
INSIGHT

LLM Hallucinations Could Limit Market Size

  • Large language models (LLMs) have inherent reliability limits—hallucinations from probabilistic token prediction—which may cap demand in high-stakes use cases.
  • Chancellor warns LLM error rates (best ~2% in one test) make them unsuitable where near-zero error is required.
ANECDOTE

Car Rental Vendor Wiped Customer Data With Claude

  • Real-world mishaps highlight LLM risks: a car-rental software vendor used Claude to update code and wiped its entire customer database.
  • Chancellor uses this incident to illustrate operational fragility and downstream business disruption.
Get the Snipd Podcast app to discover more snips from this episode
Get the app