AI Datacenter Costs Exceed Manhattan Project While China Race Persists
(Modulo a Taiwan invasion, even dumber skilled immigration policies, massive changes in export controls, etc. — and as I and others have said, losing on robotics is a much realer risk now)
Miles Brundage@Miles_BrundageThe Manhattan Project stopped being a good analogy for AI development’s ambition when individual datacenters became more expensive than the whole thing. But there is 1 parallel left: the feverish pace is outlasting the original goal (beat Hitler / beat China) being a ~done deal.
#20Miles Brundage@MILES_BRUNDAGETimely
Lennart Heim@ohlennartIf Chinese domestic AI chip production were booming, why isn't it showing up in SMIC's finances—China's advanced-node fab? SMIC's closest AI/datacenter category has been flat at $300M/quarter for 3 years. Over the same period, TSMC's equivalent 3x'd.
Also I'm being somewhat loose re: "original goal" - obviously many people involved in AI development, if not most people involved, have goals other than beating China. But beating China is the most commonly cited reason why we need to prioritize speed over caution
Miles Brundage@Miles_BrundageTimely
This is a reasonable position to have though I also think that there are many people who genuinely think China's about to leapfrog us even though that seems much, much less likely than it used to to experts
Miles Brundage@Miles_BrundageAlso I'm being somewhat loose re: "original goal" - obviously many people involved in AI development, if not most people involved, have goals other than beating China. But beating China is the most commonly cited reason why we need to prioritize speed over caution