They're already running on VC-subsidized fiat money, which will crash soon.
They need massive datasets to train on to be even marginally useful -- data regular people don't have, which immediately forces centralization.
And people like Le Cun have made plausible arguments that they don't "plan" in any reasonable way -- just "react" to prompts
https://www.nytimes.com/2026/01/26/technology/an-ai-pioneer-warns-the-tech-herd-is-marching-into-a-dead-end.html
They've failed just about every time they've been asked to run anything as simple as a vending machine.
