Still need big hardware for training LLMs and other models (eg FSD). And a piece of it is betting on further break throughs.
But the edge will certainly eat away at demand.
Of course. I’m in the Ed Zitron camp regarding AI.