Yes, (almost) everybody doing machine learning is on CUDA. There are multiple pieces of this work, and the DirectX API's are only part of it. Other pieces get C

Yes, (almost) everybody doing machine learning is on CUDA. There are multiple pieces of this work, and the DirectX API's are only part of it. Other pieces get CUDA running as well. And yet another piece is a layer to get OpenGL and OpenCL workloads running on DX12 as well, rather similar in scope to how MoltenVK and the gfx-hal Vulkan Portability work are a layer to get Vulkan workloads running on Metal. This is a big effort, and it seems to me their goal is to get things to the point where stuff Just Works and you don't have to think too hard about the various bits of (technically difficult!) infrastructure to get you there.