The argument over whether artificial intelligence is a stock-market bubble has hardened into two camps that mostly talk past each other. One side, energized by Michael Burry's May warning that conditions resemble the late dot-com era, sees circular financing and dizzy valuations and braces for a crash. The other, including analysts at JPMorgan who applied a five-factor diagnostic and concluded AI fails the classic bubble test, points to real enterprise revenue and genuine demand. Both can be partly right. And that is exactly what should worry investors.
The wrong question is dominating the debate
"Is it a bubble?" is a binary that flatters our desire for certainty. Markets rarely cooperate. The more useful question is about fragility: how much of the economy's growth, and the index's gains, now rests on a single thesis, and what happens if that thesis merely disappoints rather than collapses?
The numbers behind the unease are real. A National Bureau of Economic Research study this February found that roughly 90% of firms reported no measurable impact of AI on productivity, even as executives projected output gains. That gap between spending and demonstrated returns is the productivity paradox in modern dress. It does not prove a bubble. It proves that an enormous bet is being placed on results that have not yet shown up in the data.
Concentration is the real risk
Estimates suggest an equity crash on the scale of the early 2000s could erase something like $33 trillion in value, more than U.S. GDP. The figure is less important than what it reveals: a market whose fate is bound to a handful of names and a handful of capital-spending decisions. When a few firms fund each other's growth and the same chips, models, and data centers underpin everyone's projections, correlation quietly replaces diversification. That is the condition in which a disappointment becomes a cascade.
- Real revenue is not the same as durable returns. Railroads had real revenue in 1873, too.
- Circular financing inflates demand. When suppliers invest in their own customers, growth can look stronger than the end market supports.
- The exit is narrow. Concentrated ownership means everyone reaches for the door at once.
Why the skeptics and the believers are both useful
The bulls are correct that AI is not the pets.com era. The technology is being deployed, the data centers are being built, and the spending is contributing measurably to economic growth. Federal Reserve officials have noted that distinction repeatedly. Dismissing all of it as mania is lazy.
But the bears are correct that valuation leaves no room for error. When a sector is priced for transformation and delivers only improvement, the repricing is brutal even without fraud or collapse. The dot-com crash did not happen because the internet was fake; it happened because prices assumed a timeline reality refused to honor. The internet still changed the world, a few years and a 78% Nasdaq drawdown later.
How I would position for ambiguity
If you cannot know whether it is a bubble, manage as though it might be, without abandoning the upside. That means trimming concentration, favoring the companies whose AI revenue is already cash-generative over those selling a promise, and stress-testing portfolios against a scenario where capital spending slows for a year. It also means resisting the urge to short a trend powered by genuine demand simply because it is expensive.
The honest conclusion is unsatisfying: this may be a golden era and a dangerous overshoot at the same time. History suggests transformative technologies usually are. The investors who survive the next few years will not be the ones who guessed the label correctly. They will be the ones who refused to bet everything on either answer.
This article is opinion and analysis and does not constitute investment advice. Figures cited reflect publicly reported research and analyst commentary.