it's totally possible right now and less dumb than VR or something but people aren't doing it.
to get around the locally running a AI model to generate all that shit problem you put a TPU in your 2nd PciE slot thats dedicated to the ai work
[You must be logged in to view images. Log in or Register.]
was listening to a podcast last night talk about these things are apparently it's google abandonware but people on github are working on it and writing drivers n shit to keep it going because its insanely more efficient at AI workloads than a GPU apparently