Big companies are weird. There are probably 10 teams each at intel, arm, and amd claiming that they can beat the h100 or build a better compute runtime.
Half those teams are full of crap, a quarter will take the funds and try to do something like CUDA (but better) cough ROCm. Then another eighth to a quarter will simply not have the political clout to get the whole thing done.
Add to this that we just funded 10 teams to get 1-2 functional teams… and you see why chasing after an incumbent is hard. Even when you have near infinite money to do so.
Half those teams are full of crap, a quarter will take the funds and try to do something like CUDA (but better) cough ROCm. Then another eighth to a quarter will simply not have the political clout to get the whole thing done.
Add to this that we just funded 10 teams to get 1-2 functional teams… and you see why chasing after an incumbent is hard. Even when you have near infinite money to do so.