I think this is a bit overblown.

In the event of a crash, the current generation of cards will still be just fine for a wide variety of ai/ml tasks. The main problem is that we'll have more than we know what to do with if someone has to sell of their million card mega cluster...

The problem is the failure rate of GPUs is extremely high