A nagging question is undermining the AI narrative: are we building too much? Sebastian Siemiatkowski of Klarna has voiced what many are thinking but afraid to say. He questions the “thoughtful thinking” behind the massive capital expenditure on data centers. If the demand for AI turns out to be less than the infinite growth priced into stocks, the industry is facing an overcapacity crisis.
This “infrastructure trap” is a classic feature of bubbles. In the 1800s, it was railroads; in the 1990s, it was fiber optics; in 2025, it is GPUs and data centers. The result is always the same: a boom in construction followed by a bust in profitability.
The market is beginning to price in this bust. The slide in tech stocks and the crash in crypto (which relies on similar hardware) reflects a fear that the hardware cycle has peaked. If Nvidia’s chips are not utilized to their full potential, the $4 trillion valuation collapses.
The energy constraints mentioned by Google’s CEO add another layer of risk. You can build the data centers, but if you can’t power them, they are useless assets.
If the infrastructure boom turns into a glut, the capital destruction will be immense, validating the nervousness of fintech leaders and bankers alike.