Bubbles popping never cause the knowledge to be lost. The railway bubble popping didn't cause railways to vanish.

What did stop is investment, and they stopped running some lines, and some stuff was nationalized.

I doubt we'll see nationalizations because that's not America's style (thank goodness!) but if the tide goes out on AI funding we should expect to see much longer gaps between model releases, people complaining that model knowledge is getting stale because there wasn't a pretraining run for long enough, price rises, capacity bottlenecks that go unaddressed, few if any new product launches, maybe some mergers and layoffs. Some labs will quit the market. The market might consolidate down to just two models.

Conceivably the industry might try and get a waiver from anti-trust rules so firms can pool resources to train base models, sharing them with each other and differentiating through RL or product level differences.

A lot of this depends on what happens in China. It's easy to think China can just subsidize everything forever but their own economy is showing signs of serious weakness caused by decades of resource misallocation, like the huge real estate bubble, the pullback on the funding of humanoid robots and so on.

>I doubt we'll see nationalizations because that's not America's style (thank goodness!)

Laughing in a TGV