Everyone is designing their own bespoke AI silicon now. CUDA is just a backend for things to run on top of, like how if a whole industry uses a certain power to

Everyone is designing their own bespoke AI silicon now. CUDA is just a backend for things to run on top of, like how if a whole industry uses a certain power tool brand so everyone in that industry has a backup battery for it and inertia goes into effect. However the vast majority of people working in AI couldn't write 'hello world' in CUDA. The language that makes AI and ML run is Python. If some other hardware came along that could seamlessly take over and let all the Python engines work just as fast no one would notice. That is my perspective anyway.