I just remembered there's wetware being developed. Local setup SoTa model will probably require 1TB VRAM by 30s, right? And that's expensive for fabs etc to shit out… but what about a garloid? A wetware has an upfront cost of setting up and running, but it probably scales cheap, right? Meat weaves itself, unlike most of our tech, which has to be painstakingly engraved in a chinaman life force extraction plant.

So, to ramp compute, are we inevitably going to turn to brain tissues? Maybe not human ones, but something people don't give a fuck about, like cuttlefish?