AI Spending Outpaces Results, Not Collapse

I enjoy watching how smart money moves when the narrative starts to crack. Steve Eisman, the investor behind one of the most famous short positions in financial history, has apparently sold a key AI-related tech holding and is publicly voicing doubts about the sector. That is worth noting, not because Eisman is always right, but because he has a specific skill: identifying the gap between a compelling story and the underlying economics. The concern he seems to be raising is one I find completely understandable—AI spending is enormous, the infrastructure build-out is real and expensive, and the returns to the businesses actually paying for it are still, at best, uneven. We are unequivocally in a period where the capital expenditure side of AI is much easier to measure than the productivity gains on the other end.

The friction here is that Eisman made his name on a structural collapse, a fundamental mispricing baked into an entire asset class. AI infrastructure is expensive and some valuations are stretched, but that is a different kind of problem. The companies building out GPU clusters and foundation models are not packaging bad assets into opaque instruments and selling them to unsuspecting buyers. What is actually happening is more mundane: enterprises are still figuring out where AI genuinely derives value inside their operations versus where it just looks impressive in a board deck. That process takes time, and the stock market is historically bad at being patient. When someone like Eisman reduces exposure, it is less likely a "Big Short" moment and more likely a reasonable call that a lot of good news is already in the price.

The distinction between infrastructure cycles and application cycles matters because they almost never move in sync. The infrastructure wave—chips, data centres, cloud capacity—is forward-looking by design and tends to overshoot. The application layer, where businesses and consumers actually get value, is what determines whether the spending was justified. We are still early in that second phase. Eisman selling a position does not mean AI is a bubble in the 2008 sense, but it is definitely a signal that the easy part of the trade, buying anything adjacent to the word "AI" and watching it go up, is probably behind us. What comes next is more interesting: real differentiation between companies that have built something durable and those that have mostly benefited from the tide.