> And then another one?

But what when salaries would increase by factor 100-1000? This would make AI use even cheaper. (About as realistic...)

> And then the cloud stuff goes away because all the vendors go broke, despite the price rises, and you have to switch to using local models?

In this case there will be cheap inference hardware on the market as someone will sell all those assets. They don't disappear when a company goes broke (object permanence).

We could also speculate what would happen if all hyperscalers (AWS, GCP, Azure, Alibaba, ...) would suddenly be broke. Again, about as realistic a scenario.