Nothing per se. It's just that it lags far enough behind CUDA that unless you have a specific reason to avoid NVIDIA hardware there are very few pragmatic reasons to use it.