From article:

Jim Chanos says the money will come from what the chips produce, not where they reside

Data centers could spell doom for the AI investment story, warns Jim Chanos.

Chanos said he's even less convinced on the newest version of data centers - those that host GPUs, or rent out computing power to AI companies. "It's basically a commodity business, particularly as everybody's building them. And our view to clients has been the magic and the money is going to come from what the chips produce ultimately, not where they reside," he said.

Investors wanting to make AI bets should stick to pure AI titans like OpenAI, xAI, Anthropic, or hyperscalers, he said. "But investing in bitcoin miners that are converting to data center companies or me-too companies that are jumping on the so-called neoclouds, which are basically just landlords, to me seems problematic for two reasons," he said.

The first is that hosting GPUs is a low -margin, low-return on capital business, and the second is any data center buying the GPUs themselves must be making a bet on depreciation, he said. Chanos' own estimate is that those GPUs could end up obsolete or not fast enough for clients within five years. Hedge fund founder Harris Kupperman and "Big Short" investor Michael Burry have also delivered strong warnings on GPU depreciation.

Chanos said that among the big hyperscalers building out their data centers - such as Microsoft (MSFT), Meta Platforms (META), Oracle (ORCL) and Amazon (AMZN) - the first two will be able to finance that from their cash flows, but the rest "will need external financing."

The overriding AI bet is the technology will hit an inflection point in 2027 or 2028 and profits will flow. Oracle and Amazon have shown an inability thus far to monetize those big investments in AI, equipment, hardware and locations, and that's one reason he's short on Oracle.