[jdaxe]

Totally agree with your sentiment.

I think the inertia of switching is lower for ML because libraries like pytorch support more backends besides just CUDA, such as ROCm (AMD) and MPS (Apple Metal). I used the pytorch ROCm backend recently and there's certainly some more work that needs to be done in this space to get it to the same level as the CUDA backend (performance and compatibility wise), but at least it makes other GPU vendors an option.