China's AI Data Centers Accelerate Expansion into Energy Heartlands, with Ulanqab in Inner Mongolia Emerging as Asia's Fastest-Growing Hub
nashnova research
China has 24 GW of data-center capacity online — more than the rest of Asia combined — and the Inner Mongolian city of Ulanqab alone has committed roughly 15 GW, turning cheap power and freezing winters into a strategic rear base for the AI race.
How big is 24 GW, and how much has China actually built?
Chip-research firm SemiAnalysis puts China's operational data-center capacity at 24 GW, exceeding the rest of Asia combined but still less than half of the U.S.'s 56 GW.
Another 50 GW is under construction or announced. This means → if all of it lands, China's total reaches roughly 74 GW — on par with America's current footprint.
In plain terms = China is already Asia's largest data-center market, and it is still doubling down.
Why Ulanqab, of all places?
This remote farming city — once known for potatoes and yogurt — now has 89 data centers built or planned, with committed capacity of about 15 GW, making it Asia's fastest-growing data-center hub.
The core edge boils down to cheap power and cold weather. Local electricity costs roughly ¥0.358/kWh; in Beijing the rate is about ¥0.80 — more than double. The annual average temperature is just 4.3 °C, slashing cooling costs naturally.
Inner Mongolia's installed wind capacity — 117 GW, the highest in China — is nearly four times the UK's entire generation capacity. Fossil-fuel capacity adds another 130 GW, mostly coal. This reflects a city sitting on a massive surplus-power pool, and data centers are the sponge.
David Fishman, energy analyst at The Lantau Group, says Inner Mongolia likely has "the largest local power surplus of any jurisdiction in the world."
Which heavyweights are already in?
Alibaba and Huawei operate data centers in Ulanqab; Huawei already has three. ByteDance has become China's single largest lessee of compute infrastructure.
AI labs DeepSeek and Z.ai are also building facilities there. Operators ZData and VNET are constructing in parallel.
Huawei's rotating chairman Eric Xu visited Ulanqab for the first time in August: "I thought it was a small city — the scale was beyond anything I expected." This means → even Huawei's own leadership underestimated the build-out density on the ground.
Compared with the U.S., why is China building so fast?
The American bottleneck is grid interconnection — projects queue for years waiting for power hookups.
China's model is different: land, electricity, and construction capacity can be marshaled centrally. In plain terms = the U.S. is stuck on the "last mile of power," while China moves the data center to where the power already is.
Creative reuse is part of the story — operator ZData converted an abandoned high school, shuttered after enrollment fell, into a data center. The playground where students once took breaks will soon hold rows of server racks.
Can this pace last?
Ulanqab sits just two hours from Beijing by high-speed rail, giving engineering teams quick access — a rare logistical advantage for a remote site.
The central question remains: whether this scale and speed can keep converting into real AI competitiveness over time.
This reflects a pattern where China's AI infrastructure — the hardware layer — is running ahead; the real test is whether the models and applications built on top can keep up.
市场有风险，内容仅供研究参考，不构成投资建议。