[mhh__]

Processors used to spend most of their time sorting, after a period of wasteful bullshit they may return to hotspot's dominating again in the form of AI inference.

[moffkalast]

Won't we have dedicated analog/optical/whatever inference accelerators by that point as well?