NVIDIA should hire the guy, then hire whoever he says he wants on the team and let them rip. That's what I'd do if I were in their shoes. Viable paths off C/C++ are badly needed, and currently the only real viable path with an ecosystem and community is Rust.
NVidia decided for Ada against Rust for their automotive firmware, their cards are designed based on C++'s memory model, and they have a big ISO C++ presence.
NVIDIA has so much cash these days that they don't have to do just one thing. They can do _all_ the things, at the same time. This realistically would only take a 5-7 person team, including the manager. That's $2M/yr tops, all in, which is peanuts for NVIDIA. Much less if the author is not based in the US.
Enable more people to build more GPU-accelerated software, which in turn drives sales of their GPUs. A cuda-specific library like this also strengthens their monopoly on GPU compute.
It's hard to put in numbers because it depends on the software that will be built. A single marginally popular software adding or improving GPU-acceleration could easily increase GPU sales by $2M/year
Precisely. Their lead over AMD is insurmountable only because of CUDA. They should feed the goose that lays such awesome golden eggs, and feed it well.
