Is the AI Trade a Bubble, Really?

What actually makes something a bubble

A bubble isn’t just a high price. It’s a price that has detached from the underlying value of the business and keeps rising mainly because buyers expect to sell to someone else higher up. The technology underneath is often perfectly real — the internet was real in 1999. What turns a boom into a bubble is the moment the price stops being about what a company earns and starts being about who buys next. By that test, the AI trade carries some genuine warning signs, and also some genuine defenses. Both deserve a fair hearing.

Concentration: a handful of names carrying the market

The first red flag is how top-heavy the market has become. A small group of giant technology companies now makes up roughly a third of the entire S&P 500 — close to the same degree of concentration the index carried at the peak of the dot-com bubble. When that few names account for that much of the market, the index stops being a diversified bet and becomes a wager on a single theme. A stumble in one corner — a disappointing chip outlook, say — can drag the whole tape down with it. That fragility is the cost of concentration, and it’s been on display.

Circular financing: when an industry funds its own demand

The second red flag is subtler, and it’s the most important one to understand. A growing share of AI “demand” is the industry financing itself. The pattern goes like this: a chipmaker takes an equity stake in — or extends financing to — an AI startup or a cloud provider; that company turns around and spends the money buying the chipmaker’s chips; the chipmaker books the sale as revenue. Capital travels in a loop among a small cluster of players, and revenue that looks like external, arms-length demand is partly the supplier paying itself.

This isn’t inherently shady — strategic investment is ordinary. The problem is what it does to the signal. Circular flows make demand look stronger and more durable than it is, because the same dollar can appear as an investment, a purchase, and booked revenue as it moves around the loop. And the structure is reflexive: it reinforces itself on the way up and reinforces itself on the way down. The moment outside capital slows, the loop tightens, orders get cut, and the demand that financing manufactured evaporates.

If that sounds familiar, it should. Vendor financing — suppliers lending customers the money to buy their own products — was a hallmark of the late-1990s telecom boom, where equipment makers like Lucent and Nortel propped up sales to customers who ultimately couldn’t pay, then ate the losses when the cycle turned. The same dynamic surfaces wherever a capital-hungry industry grows faster than its customers can fund themselves. It’s worth watching closely in AI for exactly the reason it mattered then: it flatters the numbers most right before it stops working.

The gap between spending and payoff

The third red flag is the distance between what’s being spent and what’s coming back. Companies are pouring hundreds of billions of dollars a year into AI infrastructure, but the measurable return is lagging well behind the outlay. One widely-cited MIT study found that the large majority of enterprises weren’t yet seeing a real return on their generative-AI spending. The spending is a hard fact today; the payoff is still a forecast. A bet can pay off on a delay — but the wider that gap runs, and the longer it stays open, the more the current price is leaning on faith rather than cash flow.