A city in Inner Mongolia has emerged as one of China’s fastest-growing AI compute hubs, with companies committing to data-center projects representing an estimated 12.5 gigawatts of capacity — a sign that the competition over artificial intelligence is moving deeper into power, land, cooling and physical infrastructure.
WHAT’S HAPPENING
Ulanqab, a city in Inner Mongolia roughly two hours west of Beijing by train, is rapidly becoming one of China’s most important AI infrastructure centers.
Nearly 100 data centers have opened or are under construction there since 2016, according to reporting based partly on a recent Goldman Sachs analysis.
More strikingly, companies have committed to projects representing an estimated 12.5 gigawatts of data-center capacity, with more than 70% of those commitments announced within the past year.
That does not mean 12.5 gigawatts are already operating.
It means the planned buildout has reached a scale that places Ulanqab among the fastest-growing compute clusters in Asia.
Chinese technology companies including DeepSeek, ByteDance, Alibaba and Xiaohongshu are reportedly among those expanding infrastructure in the region.
WHY IT MATTERS
For years, Chinese AI companies often relied more heavily on rented cloud capacity while American technology companies poured enormous sums into their own physical infrastructure.
That may be changing.
The AI race is increasingly becoming a competition over more than models.
It is becoming a competition over:
electricity, land, cooling, fiber connections and control of compute capacity.
Ulanqab offers several advantages.
Its cold winters can reduce cooling requirements. Electricity is relatively inexpensive. The region has large supplies of wind and solar power, alongside a continuing coal base. And dedicated fiber connections reportedly allow data to reach Beijing with latency below five milliseconds.
That combination makes Ulanqab particularly attractive for both AI training and inference.
WHO BENEFITS
Chinese AI companies that control their own infrastructure could gain more predictable access to compute while reducing dependence on outside cloud providers.
Energy producers may also benefit.
China has built enormous amounts of renewable generation capacity, and large AI data centers could help absorb electricity that might otherwise be difficult to use efficiently.
Local governments may gain investment, construction activity and technology jobs as compute clusters expand.
And China itself could strengthen its ability to compete with the United States by building a larger domestic foundation for advanced AI.
WHO LOSES
The biggest pressure may fall on local resources.
Ulanqab receives only about 14 inches of rainfall annually, and the region was already facing water-management challenges before much of the planned data-center capacity became operational.
The local water authority has reportedly used overnight shutdowns at some facilities to manage peak demand.
Electricity presents another complication.
Although wind and solar are expanding rapidly, coal still supplies a substantial share of the region’s power.
That means the environmental impact of Ulanqab’s AI expansion will depend heavily on how quickly cleaner generation can replace fossil-fuel power while still providing the constant electricity data centers require.
WHAT HAPPENS NEXT
Watch whether the 12.5 gigawatts of committed capacity actually gets built — and how quickly.
Also watch whether other Chinese AI companies begin shifting from renting compute toward owning more of their infrastructure directly.
The larger question is becoming difficult to ignore.
The global AI competition may increasingly be decided not only by who builds the smartest model.
It may also depend on who can build the physical system underneath it fastest, cheapest and at the greatest scale.
