AI Is the Bubble to Burst Them All (www.wired.com)

The piece argues that AI may be more than a conventional tech bubble — it could be the “ultimate bubble” — and evaluates that claim using Brent Goldfarb and David Kirsch’s empirically grounded framework from Bubbles and Crashes. That framework scores innovations on four predictors of bubbledom: uncertainty, prevalence of pure-play bets, entry by novice retail investors, and coordinating narratives. Applied to generative AI, the article finds alarm bells: core business models remain unclear (big players still burn billions, inference costs are high, and many companies reportedly lose money on most user queries), early promises have proven uneven, and a striking MIT finding showed 95% of firms adopting generative AI didn’t profit. Historical analogies — radio and the dotcom era — underscore how powerful narratives can inflate valuations before real revenue streams materialize. The market structure amplifies systemic risk: heavy concentration in Nvidia (about 8% of market value as of late summer 2025), massive VC/private-market flows into pure-play AI startups, and retail mania via Robinhood-style trading create the novice-investor and liquidity conditions that historically precede crashes. Cross‑dependencies (Nvidia chips, Microsoft cloud, OpenAI models) and huge private stakes (SoftBank, potential IPOs) raise the prospect that a correction would be more than a sector event. For AI/ML practitioners and investors, the takeaway is pragmatic: separate hype‑driven narratives from hard product‑market fit, account for persistent compute and data licensing costs, and prepare for material downside risk even as the technology’s long‑term value remains real but uncertain.