That's not important if the goal is to run existing CUDA code on AMD GPUs. All you have to do is write portable CUDA code in the future regardless of what Nvidia does if you want to keep writing CUDA.

I don't know the economics here, but if the AMD provides a significant cost saving, companies are going to make it work.

> Nvidia can always add things to make it difficult

Sounds like Microsoft embedding the browser in the OS. It's hard to see how doing something like that wouldn't trigger an antitrust case.