Why is NVidia seemingly going along with the MSFT lock-in push?

This is in regards to the big announcement a few days ago. I wanted to see if it could achieve the same performance I can with 90+ it/s on Ubuntu with a 4090 using my own TRT compile pipeline. However, the directml packages they said to install for Olive don’t exist on Linux. And even if they did I have since discovered they need MSFT DirectX. The demo just doesn’t work on Linux. Then I hear perhaps it was intended to be that way. Thus, …

Why does the new announced driver which is supposed to squeeze more usage from the GPU not have a Linux version?
Why does this new announcement require directml which requires MSFT’s directX?
Why does the Olive Stable Diffusion demo not work on Linux?
AI books website tutorials are 98%(?) pytorch/cuda based. How does directml fit in with that?
Does MSFT really think it can renege on their commitment to embrace Linux?
Doesn’t NVidia have the expertise to produce drivers and the lowest level tools to achieve the highest possible performance without needing MSFT’s DirectX? After all NVidia owns the hardware.
Can’t NVidia do in PYTorch what MSFT is doing without vendor lock-in?
Should open software require it be run on Windows?

Am I wrong about a few of these? Perhaps but something smells suspect.