I have a feeling they might not actually need the latest fab tech. What matters is FLOPS/dollar (production cost) and FLOPS/kilowatt (energy use of resulting c

I have a feeling they might not actually need the latest fab tech. What matters is FLOPS/dollar (production cost) and FLOPS/kilowatt (energy use of resulting chips). If you can use an older tech fab to get production cost down enough, it might be worth living with bigger transistors that use a little more energy. AI stuff is massively parallel, unlike most other tech stuff, so they don't actually need super low latency and high clock speeds.