Latest Updates

Data centers in Ulanqab, Inner Mongolia, China. [Photo/Wang Kai]

On the grasslands of northern China, about 240 kilometers from Beijing, a city of fewer than 2 million people has quietly become an important player in the country's AI drive.

Ulanqab, whose name means "red mountain pass" in Mongolian, has attracted more than 100 data centre projects to date. Last month, Envision Group put its AI computing mega-cluster into operation here, with a computing capacity of one million PFLOPS. Covering an area roughly the size of 20 football pitches, it is among the largest facilities of its kind in the world.

It is only the most visible sign of a much larger build-out. With a total of 460,000 server racks built, the city's operational computing capacity has reached 172,000 P, up 112 percent year on year, with intelligent computing accounting for more than 95 percent. That represents roughly 8.7 percent of China's national computing capacity.

Twelve years after Ulanqab began its journey to build a "cloud valley" with an agreement signed with Huawei to construct its cloud data center, the city set itself a more ambitious target in late 2025. Its three-year action plan calls for the city to become a "Token Capital", linking wind and solar power to green computing and, ultimately, the production of tokens.

This ambition is underpinned by a decisive cost advantage. Ulanqab sits within the West Inner Mongolia Power Grid, an independent one outside China's two main state grids. That independence gives the city industrial electricity prices of roughly 0.3 yuan per kilowatt hour, among the lowest in China.

Its climate helps, too. With an average annual temperature of just 4.3°C, Ulanqab can make use of natural cooling for nearly 10 months of the year, helping keep its designed power usage effectiveness, or PUE, below 1.2, a key efficiency benchmark for modern data centers.

And then there is the network. Two dedicated 144-core fiber-optic cables connect Ulanqab directly to Beijing, with measured one-way latency as low as 2.1 milliseconds and failover switching in under 20 milliseconds. That means AI companies in Beijing can train models on servers in Ulanqab without meaningful delays.

Geographically, Ulanqab is well placed to bring electricity generation and computing demand into closer alignment.

Located in the middle of Inner Mongolia, China's largest renewable-energy-generating region, driven by wind and solar, the city had reached 20.28 million kilowatts of grid-connected renewable-energy capacity by early 2026. Equipment manufacturing is increasingly being supplied locally, too. Investment in wind, solar, hydrogen and energy storage has reached 3.15 billion yuan. A 1.05-million-kilowatt/6.3-million-kilowatt-hour compressed-air energy storage project is scheduled to be completed by the end of 2026.

Leveraging the proximity to power sources, Centrin Data, a company, has set up the first data center project that integrates "source-grid-load-storage", a system that coordinates power generation, grid connection, computing demand, and battery storage into a single dispatch framework.

A 300 MW renewable energy station, 200 MW of wind and 100 MW of solar, sits just about 16 kilometers from the computing base. Power is transmitted through a dedicated 220 kV line to the data center, supported by a 45 MW/180 MWh battery-storage system that smooths fluctuations in wind and solar output.

Sandbox demo of Centrin Data "source-grid-load-storage" integrated system [Photo/Wang Kai]

The company has also deployed what it describes as China's first integrated computing-power and energy-control platform, allowing the data center's computing load to automatically adjust to fluctuations in output from its wind and solar farms.

The results are measurable: 848 million kilowatt-hours of self-generated, self-consumed green electricity per year, a renewable substitution rate of 38.74 percent, and annual reductions of 580,000 tons of carbon emissions.

It is learned a second phase of the project is expected to begin construction by the end of this year, adding 380 MW of new energy capacity.

Where nationwide computing demand lands

In the first half of the year, fixed-asset investment and electricity consumption in Ulanqab's computing-power industry increased by 38.1 percent and 89.52 percent, respectively, from the same period last year.

Once all of the current projects are completed, computing power capacity is expected to reach 1.5 million P.

With project agreements still increasing, the city aims to build a green computing power foundation for nationwide large-model training and inference.

The companies that have established a presence here cover the main drivers of China's AI computing demand: leading enterprises such as Huawei, Alibaba, Apple, ByteDance, Kuaishou, DeepSeek, and Baidu have set up data centers or computing businesses in this rising AI hub.

Their workloads span model training, intelligent driving, digital twins, social-media analysis, image recognition and smart assistants: DeepSeek's large-model training, ByteDance's recommendation and generation business, and XPeng's autonomous driving intelligent computing center all run in Ulanqab. Huawei Cloud has launched its Ascend AI super-node cloud service here, while Alibaba Cloud has deployed Fudan University's Qiewen No. 1 intelligent computing cluster. A film and television computing power industrial park has also been unveiled, with four AI-generated films in preparation.

The rapid expansion is creating a demand for people as well. The companies are working with local academic and research institutions to train professionals for data-centre operations, AI data labelling, model training and related fields. Together, these programmes produce more than 3,000 skilled workers in intelligent computing each year.

By the end of next year, the city aims to achieve 100% green electricity supply for its computing centers, powering an AI industry ecosystem that integrates computing services, token trading, and intelligent applications.

It is learned that enterprises engaged in token generation, measurement, evaluation, and security will also be introduced, alongside the construction of a token trading platform, with the aim of becoming a national hub for high-quality token supply.