Is There an A.I. Bubble? A Debate. | The Ezra Klein Show

What happened in this episode?

Ezra Klein hosts a debate on whether the current surge in AI investment resembles a financial bubble, with venture capitalist Paul Kadoski arguing the boom is already exhibiting classic overbuild dynamics and analyst Azim contending demand and utilisation are still catching up. Kadoski frames AI as a general purpose technology build-out akin to railroads or fibre, saying credit-fuelled overcapacity tends to end in failures and write-downs; he claims risks are amplified by equity-market concentration and the spread of AI exposure across household portfolios and institutional finance. He further argues hyperscalers are shifting from cash-funded capex towards debt and off-balance-sheet structures, increasing the chance of contagion if growth disappoints.

Azim separates the fate of frontier model labs from the broader build-out of data-centre infrastructure, saying compute is still undersupplied and enterprise demand remains strong, with hyperscalers increasingly serving ‘real customers’ beyond a small set of AI labs. The discussion turns to what would constitute a ‘bubble popping’, the role of Apple as a perceived AI capex hold-out, and whether AI-driven construction and financing are crowding out housing and other investment—Azim likens the risk to ‘Dutch disease’ and suggests policy may be needed to balance resources.

Both consider political risk: Kadoski predicts major regulation is likely after harmful incidents, while Azim argues regulation can stabilise an industry and that data-centre protests reflect a search for social licence, especially given limited local jobs.

What are the key points?

• Paul Kadoski argues AI capex resembles historical overbuild cycles and predicts credit-fuelled spending could reach ‘10 trillion’.
• Kadoski claims prospective IPOs for OpenAI and Anthropic (and possibly SpaceX) would be historically unprecedented in size.
• Azim says compute remains undersupplied and focuses on whether data-centre assets can earn returns from enterprise demand, even if individual model labs struggle.
• The guests disagree on what would count as a bubble bursting versus a correction, including how far investment and asset prices would need to fall.
• Both discuss crowding-out effects in construction and finance, with Azim warning of ‘Dutch disease’-style distortions and calling for policy responses.
• Kadoski argues disruptive labour repricing and backlash or safety incidents could trigger discontinuous regulation; Azim expects regulation and community benefit mechanisms to evolve alongside the build-out.
• Azim claims paid consumer AI adoption is still low, suggesting monetisation is not yet proven at mass scale.

Who is speaking?

Named speakers identified in Egleze's source-linked moments from this episode.

• Dr. Paul Kadoski

Which topics are covered?

How is this page sourced?

Programme

The Ezra Klein Show

Published

16 October 2026

Duration

1h 6m

Original episode

Open the original source ↗

Egleze verifies the speaker, source and timestamp for each linked moment. Unless a moment is explicitly marked independently verified, statements remain attributed claims by the named speaker rather than established facts.