SemiAnalysis: China has 24 GW of operational data center computing capacity and another 50 GW announced or under construction, vs. 56 GW operational in the US
Financial Times
Context & Ripple Effects
China's AI-compute strategy has moved from a 2023 target for more than 300 exaflops to a reported 177% year-over-year rise in capacity by mid-2026, alongside a 2030 infrastructure-spending target. The new gigawatt comparison puts that policy drive in physical-buildout terms rather than performance metrics alone.
The United States had already accumulated more than 80 GW across built, underway, planned, or stalled projects in 2025, making project status material to any comparison. China's mid-2026 capacity disclosure and earlier national compute plan show that the planned Chinese capacity is part of a longer state-backed push.
First-order effects
The United States retains the larger operating compute base at 56 GW, while China must convert its 50 GW of planned or under-construction capacity into live facilities to close that operational gap.
China's capacity position is increasingly judged on power-ready data-center delivery, not only on its stated exaflop targets.
Second-order effects
Chinese data-center builders face sharper execution pressure around the logic, memory, and power constraints SemiAnalysis has identified as limits on scaling AI compute.
US developers' larger project pipeline does not automatically translate into operating capacity; the earlier US buildout review had already highlighted limits and return-on-investment concerns.
Third-order effects
If both buildouts proceed, gigawatts of operational, power-backed capacity will become a more central measure of US-China AI competition alongside model performance and chip supply.
The distinction between operating, under-construction, and planned facilities points to an industry in which execution risk can materially reshape apparent compute leadership.
The trend: AI competition is shifting toward the ability to turn semiconductor supply, power, and data-center construction into operating compute capacity at scale.
China is absolutely insane. As if a pork glut weren't enough, now it has an electricity glut too... Just think about how insane this sentence sounds: “Data centers are the solution to electricity oversupply.” [image] [embedded post]
In Inner Mongolia, an abandoned high school is being turned into an AI data centre. Soon, computing racks will occupy the grounds where students once gathered between classes. It's a striking metaphor for China's changing economy. https://www.ft.com/...
In the U.S., the biggest bottleneck to building data centers for AI compute is electricity. We do not have enough power, and there is a significant backlog of power generation equipment, plus an even longer backlog for building power generation facilities, including nuclear plant…
This is the first in a three-part series examining China's breakneck build-out of AI infrastructure as it seeks to become the pre-eminent tech superpower. With @EleanorOlcott and @WillLangley96 from the hinterlands of Ulanqab to Shaoguan Limited free link https://as.ft.com/...
US dominates AI build - roughly equal to the rest of world combined. China has hugely cheap power to build AI at scale and comes a creditable second. But the rest of the world lags badly... https://www.ft.com/...
China races to build data centres in bid for AI supremacy ...push is creating vast new computing hubs such as Ulanqab in Inner Mongolia, where 89 data centres are built/planned. The AI infrastructure build-out is among fastest/biggest anywhere in world. https://www.ft.com/...
Abundance with Chinese characteristics ‘China is building data centres at rapid speed across its energy-rich hinterland, as Beijing seeks to turn abundant electricity and cheap land into an advantage in the global race to develop AI.’ https://www.ft.com/...