The AI bubble debate keeps asking the wrong question.

Every few weeks, someone important says the word “bubble” out loud. Sam Altman said it. Jeff Bezos said it. Jensen Huang says the opposite, on a schedule. Economists now grade the AI boom against a checklist — overinvestment, overvaluation, over-ownership, over-leverage — like a patient in triage. The financial press turns this into a ritual: will 2026 repeat 2000, or does real revenue make this time different?

That framing feels safe. It also misses the point.

The dot-com crash was a story about companies that ran out of money. The AI story in 2026 looks stranger. These companies post real, enormous, growing revenue. Yet they still run on borrowed money, at a scale that dwarfs anything Pets.com or Webvan touched. And they lend to each other in ways that are getting hard to draw on a whiteboard.

The Circle, Drawn Out

Analysts have started calling it circular financing. A chipmaker invests in a startup. The startup buys the chipmaker’s chips with that money. The chipmaker books it as revenue. The revenue lifts the chipmaker’s valuation. The higher valuation raises more money. Some of that money flows back into more startups.

Multiply that loop across a handful of hyperscalers and a cluster of AI infrastructure startups. Add enough debt issuance that Meta, Amazon, and Microsoft now rank among the biggest corporate borrowers on the planet, with investment-grade bond issuance from AI-linked companies surging well past historical norms. You get a financial structure expanding faster than any believable adoption curve can justify.

This is the part that actually resembles 2000 — not the revenue numbers. Telecom equipment makers once financed their own customers’ gear purchases and called it growth, until the day they couldn’t. Circular financing doesn’t create fraud by itself. It creates fragility. Confidence, not cash flow, ends up doing more of the load-bearing work than anyone wants to admit.

Alibaba’s own earnings this month showed the strain even a cash-rich giant feels when it commits billions to AI infrastructure before the payoff fully lands, a pattern Ai Insights News traced in its breakdown of the company’s cloud spending.

Where the Risk Actually Sits

Here’s the genuinely new part. It should worry people who don’t own a single AI stock.

Data-center debt tied to the AI buildout increasingly gets packaged and sold to insurance companies and other institutional investors. That’s the same basic move that turned a housing slowdown into a global financial crisis in 2008. Private equity firms have loaded life insurers with this kind of risk. None of that happened during the dot-com bust, which stayed mostly contained to portfolios holding tech stocks.

Oliver Wyman put a number on a dot-com-style equity reset today, given how much bigger and more concentrated the market has become: roughly $33 trillion. That’s more than the entire U.S. economy produces in a year.

That figure isn’t a prediction. It measures how much bigger the room has gotten since the last time the music stopped.

The Bull Case Still Holds Up

Skeptics aren’t simply right, and believers aren’t simply in denial. Microsoft, Alphabet, Meta, Amazon, and Nvidia aren’t vaporware companies propped up by ticker symbols. They post real, enormous free cash flow. Fed Chair Jerome Powell has drawn this distinction publicly: these are companies with actual earnings, not eyeballs.

Anthropic’s revenue trajectory this year makes the same case. Quarterly revenue reportedly rocketed past $4 billion and then more than doubled the following quarter. Speculative manias don’t usually produce growth curves like that. Something real is happening — even inside individual companies, where AI agents now handle parts of daily workplace communication that used to sit with a person.

But “not a fiction” and “correctly priced” are different claims. The market has treated them as one for two years.

The Trigger Nobody Controls

Strip away the noise and one point holds up: nobody thinks AI stops being useful. The disagreement is about financing. Can it survive a shock — a weak earnings quarter from a hyperscaler, a surprise rate hike, a credit spread that widens and keeps widening?

Unlike 2000, when the trigger came from inside (companies simply ran out of runway), the AI story’s most likely trigger sits outside it: interest rates. Cheap capital is the oxygen in this room. The Fed controls the vent, not Silicon Valley.

That’s the uncomfortable asymmetry at the center of this debate. AI’s champions keep defending a story about innovation. The more urgent question is a story about plumbing: who holds the debt, how it gets packaged, and what happens to a pension fund’s balance sheet three steps removed from a chip order, if the loop ever stops closing.

Nobody asks that question at a keynote. Someone probably should.