Travel two hours west of Beijing by train and you reach the rolling grasslands and ancient cinder cones of Inner Mongolia. Long China's capital of sheep farming and coal mining, this vast arid land has over the past few years become the hottest place in the country to build an AI data center.

In Ulanqab, a city of about 1.5 million people, nearly 100 data centers have opened or begun construction since 2016. According to a Goldman Sachs research note, Chinese companies have pledged projects with a combined estimated capacity of 12.5 gigawatts in the city, and more than 70 percent of those commitments were announced within the past year. For comparison: OpenAI's $500 billion Stargate Project is set to reach 10 gigawatts of total capacity when complete.

Why here

Several factors pull companies to Ulanqab:

  • Cold climate — the city sits at high elevation on the Inner Mongolian Plateau with long, cold winters, so data centers spend less energy on cooling.
  • Proximity to Beijing — data reaches China's populous regions with minimal latency.
  • Cheap electricity — driven by fast growth in wind and solar plus an abundant coal supply, power costs less than almost anywhere else in China.

Who is building them is also notable. For the first time, Chinese AI companies are investing heavily in their own infrastructure rather than renting compute from cloud providers. DeepSeek is reportedly building a massive facility in the city, as are ByteDance, Alibaba and Xiaohongshu. For years Chinese firms spent far less on physical infrastructure than their American peers. The Ulanqab boom signals that gap beginning to close.

One problem: water

Ulanqab is about as dry as Denver, getting roughly 14 inches of rain each year. The local government is already struggling to meet residents' water demand, before most of the planned projects are even running. The city's water company was recently forced to shut off several waterworks for seven hours each night to blunt peak demand.

The facilities under construction will need less water in winter — weather data from the local government shows they require additional cooling water during only two months of the year. Even so, the new infrastructure as a whole could pose a significant environmental problem for the region.

From backup storage to training hub

Inner Mongolia has been a data center region for at least a decade, long before the current AI wave. Huawei built its first facility in Ulanqab in 2016 and Apple followed three years later. In 2021 the area was designated one of the main hubs of a national government project called "Eastern Data, Western Compute," which aims to move data centers into China's western hinterlands.

There was a drawback. Sitting far from the populous eastern coast, these facilities initially faced high latency when moving data to most users, so they were largely relegated to backup storage — until AI gave them a new purpose. As Andrew Stokols, a professor at Singapore Management University who studies China's compute infrastructure, puts it, the rise of AI in 2022 brought the realization that those remote data centers could be well used for model training. A training run can take months and requires little real-time tinkering, so latency stops mattering.

Two dedicated fiber optic cables built in 2017 and 2019 cut average latency below five milliseconds — fast enough to support real-time exchanges like inference. Stokols says Ulanqab's growth, compared with other Chinese hubs, has been driven more by commercial demand than state investment: as startups like DeepSeek, Moonshot AI and Zhipu AI attracted paying users at home, a business case emerged for inference facilities close enough to serve them without excessive latency.

Soaking up surplus renewables

The Chinese government has another reason to favor the region: it can absorb the country's excess renewable energy capacity. Damien Ma, director of the Singapore-based research center Carnegie China, says there is a positive correlation between the regions of China with the most unused renewable energy and those that have built the most data centers.