I wonder how many of those chips were acquired to run metaverse stuff. Should be lots of overlap between rendering graphics and running cuda based models. I'm interested in seeing how the behemoths that are Meta and Google catch openai. I think it's a question of when, not if. Both companies just have a ridiculous amount of resources to throw behind these efforts. At least meta is releasing their stuff as "open sourc…
Almost no overlap: this metaverse thing just needs classical CPU servers (and not a lot of them considering the minimal user activity there).
For now Google is still late to the party (full proprietary, and nobody has seen the supposedly good model called Ultra, only an average one called Pro), and Meta is actually the company that has pushed the field forward for all companies (with LLaMA).