In 2022, China announced an ambitious undertaking: to leverage the country’s geography for its AI ambitions.
Four years later, the story of “Eastern Data, Western Compute” (东数西算) is buried in government reports, spreadsheets, and anecdotes about racks sitting idle.
We put the data in one place, and the black box opened.
Guangdong, the southern industrial hub that heralded China’s economic miracle, tops the chart at 127 million. Multimillion-strong cities cluster along the eastern seaboard and stretch into the heartland. The west, however, is arid, mountainous, and far less inhabited.
Tibet, Asia’s great water tower, cradles alpine lakes and precious glaciers. The southwest’s rivers flow downstream from the Himalayan range. Meanwhile, China’s crowded northern and eastern provinces are thirsty — and so are data centers.
Inner Mongolia leads the country in electricity generation, propelled by traditional coal reserves as well as renewables like wind and solar. The rivers that made Sichuan and Yunnan rich in water also provide hydropower. In Xinjiang and Gansu, renewables rule the day. Much of this energy is sent eastwards; they are the backbone of China’s gleaming skyscrapers and world-leading industry.
In May 2021, four government agencies jointly published instructions on “Accelerating the Construction of a Coordination and Innovation System for the Nationwide Integrated Big Data Center.” The plan takes shape: China would coordinate its resource-abundant west to serve the computing demands of its eastern innovation hubs. Eight national computing hubs and 10 clusters were designated. Notice where they are: while some are clearly in remote western provinces, others remain close to population centers.
By the end of 2023, Zhangjiakou (near Beijing) hits 330k racks; Shaoguan (~300km north of Shenzhen) nears its target; Qingyang, in remote Gansu, reaches 5.3 EFLOPS. Growth seems even between the east and west.
Qingyang leaps from 5.3 to 50+ EFLOPS in a single year; Horinger’s intelligent compute goes from 400 PFLOPS to nearly 48,000. Maybe real change is coming?
Gansu province attracts national attention: a desert prefecture powers a national AI base with local green energy. State media claims that “half of the country’s top AI firms” set up shop there.
Across all ten national clusters, the average rack-up rate (上架率) was about 62% in 2024 (Caijing). Zhangjiakou and Horinger only reached around 60%. The good news: Zhongwei and Qingyang, two small western ones built to demand, are nearly fully utilized. The other four (dashed) don’t disclose a figure at all.
Actual utilization rates of intelligent computing centers in China average to merely 32%. (China Academy of Information and Communications Technology / iFeng)
The buildout continues to wrap around four megacity regions — and their enormous populations. Beijing spills into Hebei, where Zhangjiakou is; the Yangtze Delta into Jiangsu and Zhejiang; the Greater Bay Area into inland Guangdong’s Shaoguan; and the dense Chengdu-Chongqing corridor, rather than new data center towns, remain the nexus of the west. The map didn’t move west so much as 45 minutes out of town. (Bubble = region population.)
By end-2025, intelligent compute concentrates fast: Zhangjiakou tops 200 EFLOPS, Zhongwei 130, Qingyang 114, Gui’an 100. These hubs together make up ~80% of China’s smart compute capacity.
But no one knows if they are getting used. Anecdotal media reports piece together a pessimistic picture: AI companies can’t tolerate the latency cost, the chip chokehold leaves overbuilt racks empty, and some western hubs continue to build general-purpose capacity that can’t service AI.
Poorer provinces were sold a Field-of-Dreams theory of development. A few — Guizhou, Inner Mongolia — may make it. For many, these data centers risk becoming the next ghost buildings rather than tickets into the AI boom.