@is_OwenLewis
Post
A few reasons that I can see. Chips are less efficient in China, therefore China draws more power for every unit of effective compute, shrinking the electricity gap relative to the raw hardware disparity.
The U.S. has the opposite problem, with not enough electricity available where needed to run every chip at high utilization.
Effective “energized compute” therefore lags nameplate hardware capacity in the U.S., while China’s far larger and faster growing generation allows fuller utilization of whatever chips it can obtain or produce.
Basically, the US is relatively hardware rich but power constrained. China is relatively power rich but hardware constrained and less efficient. These inverse bottlenecks compress the observed electricity ratio. The gap could widen or narrow further depending on how quickly the US resolves grid and equipment bottlenecks versus how fast China improves domestic chip efficiency and HBM/packaging yields.
Image from X post
Quoted post by Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) (@teortaxesTex) That's roughly 217 TWh vs 115 TWh annually, which is a solid lead but not remotely as large as the compute disparity should predict. Why?
Explanation
What it says The post explains why U.S. data centers apparently use only ~217 TWh/year versus China’s ~115 TWh/year even though the U.S. has a much larger stock of advanced AI compute. Its answer: the two countries have opposite bottlenecks. China’s chips are less energy-efficient, so each unit of effective compute consumes more electricity. The U.S. has more capable hardware but insufficient electricity/grid capacity in the right places, so some installed compute cannot run at high utilization.
Context The claimed picture is:
US: lots of chips → constrained power → lower utilization China: fewer/weaker chips → abundant power → higher utilization + more watts/compute
So electricity consumption is a poor proxy for installed AI capability. China’s enormous and rapidly growing generation fleet partially compensates for its hardware disadvantage.
Why it matters The strategically relevant quantity is not GPUs owned, or electricity generated, but energized effective compute: useful compute actually running. If the post is right, U.S. AI infrastructure advantage is being partially stranded by power/grid bottlenecks, while China can extract unusually high utilization from scarcer hardware. The future gap therefore depends both on U.S. power build-out and on China improving chips, HBM, and packaging—not merely chip export controls.
Images The charts visually reinforce the asymmetry: China generates roughly twice as much electricity overall, while Chinese data centers consume a much smaller fraction of national generation than U.S. data centers.