The sad part is that most of the people are using a higher level API like PyTorch. So AMD or Intel just need a simple low level API to access their HW, and then

The sad part is that most of the people are using a higher level API like PyTorch. So AMD or Intel just need a simple low level API to access their HW, and then write and tune some kernels for PyTorch (and I suspect the community or AI will gladly help with the tuning). So basically there does not seem to be a real CUDA lock-in, it's just the fact that the competition still seems unable to do even this.