China's AI data-centre race moves to Inner Mongolia, where cheap power and chip-friendly subsidies draw Huawei, Alibaba and ByteDance's compute
What's the deal? The Financial Times, in the first part of a series on China's data-centre push reported from Ulanqab in Inner Mongolia, says the city has become Asia's fastest-growing data-centre hub according to an operator executive. Ulanqab has 89 data centres built or planned and about 15GW of committed capacity. The FT saw at least a dozen companies across four campuses, including Alibaba, Huawei (three data centres) and VNET. ZData is converting an abandoned high school into one.
Scale: Citing SemiAnalysis, the FT puts China's operational data-centre capacity at 24GW, more than the rest of Asia combined but less than half the US's 56GW, with a further 50GW under construction or announced.
Why Inner Mongolia: Power is the draw. The region had 117GW of wind capacity by June plus about 130GW of mostly coal-fired capacity, which analyst David Fishman of The Lantau Group says may be the largest local electricity oversupply of any Chinese region. Data centres pay about US$0.05/kWh, versus more than US$0.08 in many cities and about US$0.11 in Beijing. Goldman Sachs puts construction costs about 20% below larger cities, and prefabricated builds take 12 to 18 months versus 18 to 24 in the US. The FT adds that, according to people familiar with the policies, data centres running on domestic processors rather than Nvidia chips get better tax benefits and power and water discounts.
Who is there: ByteDance is China's largest renter of compute, according to the FT. People briefed on the matter say DeepSeek and Z.ai are building in Ulanqab; Z.ai did not respond to the FT. Neither company has confirmed this publicly.
What's still open: Chips remain the binding constraint given US and Chinese restrictions on Nvidia purchases and limited Chinese fab capacity. Jefferies' Edison Lee says the question is whether the data centres can get enough chips, though he sees no risk of overbuilding. Water is scarce in the region, and the FT gives no total capex figure.
The signal: China is competing on the AI-compute variables it controls (cheap power, fast build times, subsidies tilted towards domestic chips) while the chip question stays unresolved. For Europe, which faces expensive power and slow permitting, it is a useful benchmark for sovereign-compute ambitions. That comparison is our analysis, not the FT's.
Source: Financial Times: China races to build data centres in bid for AI supremacy (8 October 2026).
Image credit: Financial Times
