These companies will not be able to keep making large improvements to their models as time goes on. You gotta remember to not be holding the bag when the music stops.

Gustafson would hard disagree with you, while Amdahl shares your pessimism.

I think Gustafson has better receipts... His main argument is that software will update to take advantage of new parallelism in hardware. And that's definitely been the case.

AI has ridiculous levels exploitable parallelism. It'll scale up and out.