Clement warns the AI data-center boom could break by 2028

The spending depends on a few buyers

Clement had already warned about the market in a May article for Reuters, arguing that institutional investors should focus not on whether the boom ends, but on what happens to markets when it does. In his Bloomberg interview, he said the US bubble is being fueled by investment in the wrong technologies, especially advanced models that require substantial resources.

“We are putting money into the wrong version of the AI future,” he said. Big language models running in data centers are drawing investment, while he sees the future in smaller open-weight models that can run on ordinary desktop computers. As more applications shift to smaller, cheaper models running locally, he said, data-center investment has already exceeded what is needed.

Clement also questioned whether future company earnings will justify current stock prices. Bloomberg host Dani Burger asked about the trillions of dollars planned for data-center infrastructure in the coming years. Clement said those plans depend on a handful of technology companies whose ambitions may not materialize.

“If you take away demand from OpenAI and Anthropic, there is almost no demand for AI compute that would justify trillion-dollar investments. Only hyperscale cloud companies are planning to spend those sums next year,” he said.

A shorter runway for the boom

Two months ago, Clement thought the bubble could easily last another two years. Now, he says, a downward revision to growth plans and a slowdown in growth itself could be enough to alter earnings forecasts across almost the entire supply chain.

That shift matters because the exposure is not confined to technology companies. The Russian report notes that the US economy is now closely tied to the data-center boom, so a slowdown in AI spending could affect a wide range of industries.

I think the most revealing part of Clement’s case is its dependence on a small number of buyers: if OpenAI and Anthropic are central to the demand story, the trillion-dollar buildout is less a broad measure of AI adoption than a wager on a few companies’ plans. The announcement is quiet about how much of that planned capacity has other uses if those plans change.