3 Comments
User's avatar
Konrad Seifert's avatar

Another case that might convince more people: if we want to deploy AI in logistics and manufacturing at scale, we'll need exceptional latency for embodied agents to work. They will make huge amounts of inference calls to function, most of our future compute needs original there, not humans calling AIs, but AIs calling tools and other AIs.

If we can't supply the compute in Europe, China alone will remain the manufacturing backend, bc they have the energy, space and engineering prowess.

Unfortunately, the investors I've talked to seem to think that even if you wanna buy compute, there isn't currently enough to buy up. So we'll have to also ramp up domestic fabs, potentially?

Simon Grimm's avatar

Great point. From my initial discussions I think the big constraint right now is energy? Only starting to read through the Seminalaysis backlog, I might learn more along the way.

Konrad Seifert's avatar

Talked to some investors who buy compute and one who's selling it, it seems that there isn't enough hardware to go around (even old hardware is sold at an absurd premium). But yeah I think energy is main bottleneck, but iiuc even if we solved that, we'd quickly run into compute hardware supply bottlenecks lol