Comment by ImHereToVote

The main motivation is to make open models only work using CUDA. Once models work on AMD, Intel, or Tenstorrent, it might genuinely be over.