Fascinating cover story from Caixin on the U.S.-China AI chip race.
The conventional narrative: China is playing catch-up.
The reality: two fundamentally different strategies are emerging.
U.S. approach: brute-force capital. OpenAI’s latest chip orders require 26GW of power.
时政
(
twitter.com
)
Fascinating cover story from Caixin on the U.S.-China AI chip race.
The conventional narrative: China is playing catch-up.
The reality: two fundamentally different strategies are emerging.
U.S. approach: brute-force capital. OpenAI’s latest chip orders require 26GW of power. That’s three New York Cities at peak demand.
China’s approach: system-level optimization. Huawei’s Atlas 900 links 384 lower-powered chips to achieve 2x the performance of comparable systems. Higher energy cost, yes. But it works.
The counterintuitive finding: China’s real bottleneck isn’t supply anymore. It’s demand. Utilization rates at AI computing centers are below 30%.
Meanwhile, Bain estimates the U.S. is building toward an $800 billion revenue gap by 2030. The infrastructure spend may be running ahead of actual applications.
Worth reading for anyone tracking this space. Link in comments.
点击图片查看原图
