My laptop was capable of upgrading to Win11 and WSLg was my main motivation. I like the mix of Windows and Linux and having Linux graphics and pulseaudio mixed in with native Windows. This most likely wouldn't be described as a "consumer feature". More important for me are some opencv projects I have: You get access to your GPU compute. I can install CUDA+DNN on my WSL2 Debian. This is compelling IMO.

An option you can consider by sticking with Win10 is to either wait for Win10 21H2 or join the Insider Program Preview to get it now. Win10 21H2 gives you GPU compute access (no graphics or pulseaudio). On an older machine, I was able to build opencv-4.5 and leverage CUDA+DNN just like on Win11.

EDIT: Win10 21H2 should be released this month. The fact that it is in the Preview Channel means it's most likely a final build.

Why can't you just do ML in Windows? All common Python packages are supported. WSL2 CUDA probably gonna come with some overhead.

I'm sure it's possible, but I was trying to use Stylegan2, which I spent way too many hours fiddling with trying to get it to work in Windows.

I'm sure it was something simple I was doing wrong, but everyone I asked just went "just use Linux, Windows isn't recommended".