Nvidia too, but the bad blood between Apple and Nvidia makes it less likely. Ultimately internal adoption of MoltenVK and CUDA code access by whatever means should be priority #1.
Since Nvidia doesn't work on modern Mac hardware anymore, there's no need to run Metal on Nvidia GPUs. Nvidia isn't coming back and neither is AMD to stay, they're the past for Apple.
There has to be a reason Apple is even holding on to the modular Mac Pro concept as a working blueprint for a product they continue to spend money in R&D on.
Keeping the Mac Pro up and alive isn't that expensive anymore. There's no research and development for it anymore, only maintenance. The reason is simple, it's by far the most powerful machine Apple has right now. They don't have a M-series chip that can keep up with it in every task, only in selected benchmarks for marketing.
If they found a way to move the "UltraFusion" interconnect to a nearby "discreet" slot that would make sense and if they found a way to Rosetta-ize CUDA code to their GPUs I would be fine with that too.
Too expensive and physically too difficult to do to become feasible. Sure it could be done, but not in this "low cost" price segment. More thinking about full blown Nvidia DGX. So, maybe $200k for such a Mac Pro?
Personally, I don't care how they ingest the CUDA code, Streamcore processor code and get Metal working with Vulkan it's just critical that they find a way to make it happen.
That's not going to happen. CUDA is proprietary and only works on Nvidia hardware by definition. CUDA is not open source, so not anyone can use it, you need to buy an Nvidia card to do it. Nvidia isn't going to license it to Apple either, they want to sell hardware.
There have been a number projects trying to run CUDA code on non Nvidia hardware. They all failed, because it's a translation layer or emulation which comes with a massive performance hit. And neither is Vulkan going to happen, Apple moved away from CUDA, from OpenGL and Vulkan/MoltenVK. Technically we still have MoltenVK, but it's 3rd party and Apple is not going to utilize it. They have their own API which is tailored to their own hardware. The rest is just wishful thinking.
And for the record, I too would massively benefit from Nvidia cards in Macs, which would make porting my stuff so much easier. But it's just not going to happen, which is the reason I'm running Mac Pro class Dell machines now and the Mac Pros are collecting dust.