China has made artificial intelligence (AI) a national strategic priority and is building data centre infrastructure at pace to support that goal. The country leverages advantages in cheap power, available land and fast construction to close the hardware gap with the United States — yet limited access to the most advanced AI chips remains a major constraint.
Current capacity and growth trajectory
According to SemiAnalysis, China currently has roughly 24 gigawatts of operating data-centre compute capacity. That exceeds the combined capacity of the rest of Asia but still trails the United States, which has about 56 gigawatts. The gap may shrink quickly: roughly another 50 gigawatts of Chinese capacity are under construction or have been announced.
A striking example is Ulanqab in Inner Mongolia. Once mainly an agricultural city, it now hosts 89 data centres either operating or planned, and the region has commitments to build some 15 gigawatts of compute capacity.
Why these regions?
Ulanqab and similar areas benefit from abundant and cheap electricity. By June, Inner Mongolia had installed 117 gigawatts of wind power capacity, and the region also runs roughly 130 gigawatts of fossil-fuel (primarily coal-fired) power. That creates a substantial local power surplus, which is attractive to data-centre operators.
Electricity prices in Ulanqab are about 0.358 yuan per kWh, compared with industrial rates above 0.60 yuan in many large Chinese cities and around 0.80 yuan in Beijing. Cooler local temperatures reduce cooling costs for large server farms, and some projects connect dedicated power plants directly to data centres to avoid grid transmission constraints and related costs.
Lower construction costs and faster builds
Goldman Sachs estimates that construction costs for data centres around Ulanqab can be roughly 20 percent lower than in China’s larger cities. Projects mobilise thousands of workers, and new facilities typically come online in 12–18 months, compared with 18–24 months in the United States.
The main bottleneck: lack of chips
A crucial remaining hurdle is processors. The United States’ key advantage stems from export restrictions on advanced Nvidia AI chips. Chinese chipmakers currently lack sufficient capacity to meet the domestic surge in demand.
To counter this, Beijing offers incentives to encourage data centres to use domestic processors. Facilities that deploy Chinese chips instead of Nvidia parts can receive favourable tax treatment and cheaper electricity and water.
Market signals and risks ahead
So far, oversupply does not appear to be an acute risk: the rapid rollout of AI applications, particularly autonomous agents, is driving up compute demand quickly, which in turn is increasing prices for rentable chip capacity.
Nevertheless, as long as Washington can maintain export controls and technological advantages in advanced chips, the United States is likely to preserve its leading position. China is accelerating infrastructure deployment, but the shortage of high-end AI chips is the key factor limiting how fully the new data centres can be utilised.
