First class support for Python JIT/DSLs across the whole ecosystem.
Change the way C++ is used and taught, more focused on standard C++ support and libraries, than low level CUDA extensions.
So in a way, I think you're kind of right.
NVIDIA wants Python and C++ people, they want a new thing to moat up on, and they know it has to be legitimately good to defy economic gravity on chips a lot of companies can design and fan now.
And Khronos always expecting the community to do the work for their standards, filling in the missing pieces for actually excellence in developer experience.
Out of a great many outcomes a very topical one today is semiconductors, and the outcomes rival any on anything for corrupt, incompetent, and entirely consistent with our speed run of the road to irrelevance.
By blaming Khronos you're really just reiterating blame on Intel and AMD with the tacit inclusion of Apple. I suppose you could blame Nvidia for not giving away their IP, but they were a staunch OpenCL supporter from the start.
I've got these crazy tuned up CUDA kernels that are relatively straightforward to build in isolation and really where all the magic happens, and there's this new CUTLASS 3 stuff and modern C++ can call it all trivially.
And then there's this increasingly thin film of torch crap that's just this side of unbuildable and drags in this reference counting and broken setup.py and it's a bunch of up and down projections to the real hacker shit.
I'm thinking I'm about one squeeze of the toothpaste tube from just squuezing that junk out and having a nice, clean, well-groomed C++ program that can link anything and link into anything.