China Moves Majority of Its Data Center Capacity to Rural North, Backs Buildout With $295 Billion in State Money

China is relocating the physical backbone of its AI ambitions, and it's doing it by government fiat, not market signals.

More than half of China's data center capacity now sits in the country's northern and northwestern regions, and those areas are projected to overtake Beijing's computing dominance by 2028, according to Crypto Briefing. The national program driving this is called Eastern Data, Western Computing (EDWC), which launched in February 2022, though some sources date the formal policy to 2021. Either way, the goal is the same: pull AI training workloads out of crowded eastern megacities and into places with cheap land, cheap power, and cold weather that cuts cooling costs.

Ulanqab, a city in Inner Mongolia most Americans could not find on a map, has become the centerpiece. It has racked up commitments for 12.5 gigawatts of data center capacity, with roughly 70% of those announcements landing in just the past year. Envision Group's Galaxy Campus there is planned for 2 GW of capacity and up to 1 million GPUs, with over 80% of its power coming from renewables. DeepSeek, the AI lab that rattled Silicon Valley earlier this year, is building its own 1 GW compute campus in the same city, with a partial launch expected in late 2027 or early 2028.

The Numbers China Is Bragging About

By mid-2024, the eight national computing hubs under this strategy housed 1.95 million server racks running at 63% utilization. Electricity in China's western regions runs about 50% cheaper than in the Beijing-Tianjin-Hebei corridor, and total computing costs are more than 60% lower. Western data centers are hitting Power Usage Effectiveness ratios of 1.04 to 1.15, versus 1.32 to 1.38 in the east, meaning far less energy wasted on overhead like cooling.

Beijing is backing this with a draft national AI data center grid plan targeting full interconnection by 2028, supported by government investment approaching 2 trillion yuan, roughly $295 billion, over five years. National capacity is projected to climb from about 32 GW at the end of 2025 to roughly 40 GW by the end of 2026.

Poor Provinces, Big Server Farms

Guizhou is one of the biggest beneficiaries, and it's also one of China's most indebted provinces. It offers cooler weather, cheap land, and abundant renewable power. Beijing has mandated that all new data centers draw 80% of their power from renewable sources before the decade ends. Simeng Deng, an analyst at Rystad Energy, said that provinces with expanded solar and wind capacity make it easier for operators to hit those renewable targets, and large server farms can soak up curtailed wind and solar output that would otherwise go to waste.

The obvious tension remains: a heavily indebted provincial government hosting capital-intensive infrastructure that may or may not generate enough local tax revenue or jobs to justify the subsidies and land deals that got it built. No source quantifies Guizhou's debt load against the economic return from hosting these campuses, and that gap is worth watching.

The Power Bill Is Coming Due

China's data centers consumed 170 billion kilowatt-hours of electricity in 2025, or 1.6% of the country's total power use, with demand at the eight national hub nodes growing at an average annual rate of 39.5% over the past three years, according to People's Daily reporting carried by Alwihda Info. Tian Lei, director of the energy strategy center at the Energy Research Institute under China's National Development and Reform Commission, noted that a single 700-watt AI chip uses more than 10 times the power of a conventional CPU, and a 10,000-chip cluster needs roughly 7 megawatts just to run the chips. Cooling alone eats up more than a fifth of a computing center's total electricity draw.

China Daily, citing a Wood Mackenzie report titled "Rewiring China's Grid for the AI Era," put a bigger number on the table: data center electricity consumption is projected to nearly quadruple by 2030, hitting 774 terawatt-hours and consuming 6% of China's total electricity, triple today's share, and climbing to 17% by 2060. Electricity used by internet data services alone jumped 44% year-on-year to 49.4 billion kilowatt-hours in the first half of 2026, according to China's National Energy Administration. Wu Liqiang of the China Electricity Council said that first-half total was already nearing all of 2024's usage.

Wood Mackenzie's report expects the eight national hubs to keep hosting over 70% of China's total computing capacity through 2060, even as grid constraints push some new installations into renewable-rich areas outside those hubs.

What remains unanswered: whether a state-directed capital allocation of this size, running through provinces already carrying heavy debt loads, produces a return that justifies the subsidy, or whether Beijing is simply racing to out-build the United States on AI infrastructure regardless of the balance sheet. That question matters for anyone trying to figure out who actually wins the AI compute race.