Why don't they just release a basic GPU with 128GB RAM and eat NVidia's local generative AI lunch? The networking effect of all devs porting their LLMs etc. to that card would instantly put them as a major CUDA threat. But beancounters running the company would never get such an idea...
besides the clear limitation of the memory technology they are using compared to the nvidia's enterprise solution, for such large GPU chips that could really make use of such memory, they need to make binning possible by selling cut-down versions of them as well.
nvidia can pull this off because they can sell lower-end chips at the same time. intel is barely making a dent in sales, and making a high-end chip will only be very risky, at the cost of potentially benefiting a niche crowd.
> put them as a major CUDA threat. that is a software/ecosystem problem, which hardware alone cannot solve. for all the devs that use Macs, even in AI it is only about inference at the moment. nobody is coming at CUDA for training for the near future. amd tried and failed plenty already.