Tootfinder

Hot take: we need FAR fewer AI datacenters or #AI #gigafactories than the markets are predicting right now.
We are in a weird moment in time where tokens in the cloud are heavily subsidized, which ramps up prices for GPUs. But without that subsidy (or after the bubble bursted) for the majority of cases, the sensible way of serving AI inference is to have it run locally.
#Nvivia seems to be aware of this, which is why they severely limited VRAM on newer consumer cards - all to keep the circular financing between 4 companies running.