Rust for GPU kernels

I have worked with both rust and c++ in prod and rust honestly feels like a breath of fresh air , never had issues with the ownership and borrow model and its been mostly struggling with std library like anyone starting in a new language.

Lately i had a discussion with couple of colleagues of mine about rust and c++ and how they fit in working with gpus and mainly ML inference (thats our domain , i know there's also graphics...etc) , and discussing the "known" stack mentioned in every Job description or blog which is python and c++ (cuda and cutlass) but in reality and from experience and other colleagues and friends in the domain - mostly working at nvidia and togetherai - they are moving away from c++ for kernellling and going more towards python in cutedsl and cutlass 4 and as of lately CuTile-rs in rust.

I have been pushing for a change in the stack away from c++ , mainly replacing it with rust and others.
Even nvidia is saying on the main page that they suggest using cutedsl or starting with it first , FlashAttention 4 authors sing praise to the cutedsl away from struggling with c++ and cutlass 3.

I was thinking it was about time to move from c++, python --> rust , python ,(CuTeDSL ,Triton,Mojo) for kernels and we can easily bind them and call them from code.

Wanted to know what you guys think about rust replacing c++ in this domain ?

Anyone having the same issues or noticing the patterns?