It sounds like they did significant work getting the whole stack to work well, from Wayland to the RDP back-end to improvements to the RDP client on the Win32 side.
GPU passthrough on WSL2 is a thing since around December 2020 -- but I'm not sure if it's in stable yet, I had to install an insider build for it. It works surprisingly well and I was able to develop my ML project using it (with the help of VSCode devcontainers)
I presume getting GUI working on top of those GPU APIs was a trivial task (and maybe done by one of their interns or during a hackathon).