Beijing is losing its data center crown to windswept steppe cities running on cheap renewable power and massive GPU farms
China is executing one of the most ambitious infrastructure reshuffles in modern tech history, and the winners are cities most people outside the country have never heard of. More than half of China’s data center capacity now sits in its northern and northwestern regions, and those areas are on track to surpass Beijing’s computing dominance by 2028.
The engine behind this migration has a name: “Eastern Data, Western Computing,” or EDWC. Launched in February 2022, the national strategy is designed to push large-scale computing workloads, particularly AI training, away from congested eastern megacities and into regions where land is plentiful, electricity is cheap, and renewable energy is abundant.
The new capitals of compute #
Ulanqab, a city in Inner Mongolia, has emerged as perhaps the most consequential node in this network. The city has racked up commitments for 12.5 GW of data center capacity, with roughly 70% of those announcements coming in the past year alone.
The Envision Group’s Galaxy Campus in Ulanqab is planned with a capacity of 2 GW and up to 1 million GPUs, with over 80% of its power sourced from renewables. DeepSeek, the AI lab that grabbed global attention earlier this year, is also building a 1 GW AI compute campus in the same city, with a partial launch expected by late 2027 or early 2028.
Zhongwei, in the Ningxia autonomous region, is another rising hub. These locations share a few critical advantages: cold, dry climates that reduce cooling costs, vast open land, and proximity to some of China’s largest wind and solar installations.
By mid-2024, the eight national computing hubs designated under the strategy collectively housed 1.95 million server racks, running at a 63% utilization rate.
Why the west wins on economics and efficiency #
Electricity costs in China’s western regions run approximately 50% lower than in the Beijing-Tianjin-Hebei corridor. Computing costs are over 60% cheaper.
Western data center hubs are achieving Power Usage Effectiveness ratios of 1.04 to 1.15. Their eastern counterparts hover between 1.32 and 1.38. PUE measures how much total energy a data center uses relative to the energy that actually powers its servers. A PUE of 1.0 would mean every watt goes to computing with zero overhead.
The $295 billion bet #
Beijing is backing this transformation with serious capital. A draft plan for a national AI data center grid targets full interconnection by 2028, supported by governmental investments approaching 2 trillion yuan, roughly $295 billion, over five years.
The investment is expected to push China’s total data center capacity from approximately 32 GW at the end of 2025 to around 40 GW by the end of 2026. By 2030, capacity could nearly double from current levels. AI and high-performance computing workloads are projected to represent 48% of that capacity by the end of the decade.
China’s eastern seaboard faces growing constraints: expensive real estate, strained power grids, water scarcity for cooling, and tightening environmental regulations. Moving data centers west alleviates those pressures while simultaneously developing economically lagging regions.
Connecting remote data centers in Inner Mongolia and Ningxia to users and businesses concentrated along the eastern coast requires ultra-low-latency networking infrastructure that doesn’t fully exist yet. The 2028 interconnection target will require not just building server farms but laying the fiber and deploying the networking technology to make distributed computing feel seamless.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our