Why Inner Mongolia Has Become China’s AI Data Center Capital

Inner Mongolia has become one of the most important buildout zones in China’s artificial intelligence race, with the city of Ulanqab emerging as a major center for data centers, new compute projects, and AI infrastructure. The surge matters because it shows Chinese companies are no longer just renting cloud capacity — they are now investing heavily in the physical backbone needed to train and run AI systems.

What was once a sparsely populated landscape better known for sheep, coal, and wide-open grassland is now drawing some of China’s biggest tech names. Nearly 100 data centers have been opened or started in Ulanqab since 2016, and companies have announced plans for projects with a combined estimated capacity of 12.5 gigawatts. That scale places the city among the fastest-growing compute clusters in Asia and highlights how aggressively China is trying to close the gap with the United States.

How did Ulanqab become a core AI infrastructure hub?

Ulanqab became a data center magnet because it combines climate, geography, electricity prices, and improving connectivity in one unusually attractive package. The city sits on the Inner Mongolian Plateau, where cold winters reduce cooling needs, and it is close enough to Beijing to keep latency low for many uses.

Cheap power is another major draw. Electricity in Inner Mongolia is among the least expensive in China, helped by strong wind and solar growth as well as substantial coal reserves. For operators running energy-intensive servers every hour of the day, those lower rates can make the difference between a viable business and an expensive one.

Goldman Sachs recently described Ulanqab as one of Asia’s fastest-expanding compute clusters after tracking the scale of announced projects and their rapid recent growth. More than 70 percent of the commitments counted in the research note were made in the past year, indicating that the acceleration is recent and still building.

Why Chinese AI companies are building their own data centers

Chinese AI firms are making a strategic shift by investing in their own infrastructure instead of relying so heavily on cloud providers. That marks an important change in how the country’s AI sector is developing, especially compared with the United States, where large-scale compute spending has long been concentrated among a few dominant cloud and platform companies.

DeepSeek is reportedly building a major AI data center in Ulanqab, and so are ByteDance, Alibaba, and Xiaohongshu. Their presence suggests that the city is not just a generic hosting location but a place where major Chinese technology firms see a long-term advantage in owning compute capacity directly.

For years, Chinese AI companies were seen as more efficient in software and model development than in physical infrastructure spending. They built competitive models but often lagged in the expensive backing required to train them at scale. The surge in Ulanqab suggests that dynamic is changing.

Analysts and researchers say the expansion reflects a broader realization in China’s AI industry: remote data centers can be highly effective for model training because those workloads do not depend on split-second response times the way consumer apps do.

Why latency matters less for training than for live services

Latency is still important, but it is not equally important for every AI task. Training a model can take months, and the process does not involve constant back-and-forth with end users. That makes distant facilities more practical than they would be for consumer-facing tools requiring immediate response.

By contrast, inference — the process of serving a model to users in real time — is more sensitive to delay. That is why improved connectivity has become so valuable in Ulanqab. Two dedicated fiber-optic cables built in 2017 and 2019 reduced average latency to under five milliseconds, making the city much more usable for modern AI workloads.

What changed after 2022?

The AI boom of 2022 changed how remote infrastructure was viewed across China. Facilities that had once been seen mainly as storage or backup locations began to look far more useful for model training and AI serving.

Andrew Stokols, a professor at Singapore Management University who studies China’s compute infrastructure, has argued that the rise of AI made it clear that these distant data center sites could support serious training workloads. In practical terms, that meant old assumptions about the limits of inland infrastructure no longer held.

Ulanqab also benefited from geography in a relative sense. While it is far from China’s eastern coastal cities, it is still much closer to Beijing and other population centers than the western regions that had traditionally been favored for data center development under national planning. That makes the city a more compelling option for AI than much deeper inland locations.

How does government policy fit into the boom?

Government policy has played a significant role in shaping where compute is built, but the recent surge in Ulanqab appears to be driven increasingly by market demand. China’s “Eastern Data, Western Compute” initiative, launched to shift data infrastructure into western regions, helped establish the broader framework for the boom.

Inner Mongolia was named one of the main hubs under that program in 2021. The idea was to use lower-cost western land and energy resources to support the country’s digital economy while relieving pressure on the eastern seaboard. Ulanqab had already attracted large projects before the AI era, including Huawei’s first local data center in 2016 and Apple’s entry in 2019.

But experts say the latest wave looks different from earlier, more centrally directed phases. Instead of being built primarily for storage and government-aligned purposes, many of the new projects are being driven by companies that need compute for commercial AI products and model serving.

Who is paying for the latest wave?

Chinese AI startups and major internet platforms are increasingly financing the new infrastructure themselves. That trend matters because it suggests the domestic AI market is maturing enough to justify direct capital investment in compute rather than depending mainly on rented cloud capacity.

As consumer demand for AI tools grows, companies such as DeepSeek, Moonshot AI, and Zhiput AI have found a business case for closer-inference infrastructure. Their customers are paying enough for services that the cost of owning or co-locating data centers can make economic sense.

Why is electricity such a big part of the story?

Electricity is central because AI data centers consume huge amounts of power, and the cheapest electricity often determines where they are built. In Inner Mongolia, abundance of land is not the problem; abundant and affordable power is the main advantage.

The region’s energy mix is changing, but not uniformly. Wind and solar are expanding quickly, and policymakers are eager to use data center demand to absorb surplus renewable generation that might otherwise go unused. That creates a powerful policy argument for building compute in places like Ulanqab.

Envision, one of China’s biggest wind turbine manufacturers, announced this month that it will build a 2 gigawatt AI data center in Ulanqab tied directly to its own clean power supply. The project is a notable example of how energy producers and AI operators are beginning to overlap.

Research on the region suggests that places with the most unused renewable power in China often line up with places that have attracted the most data center construction.

But is the power really clean?

Not entirely. Despite the growth of renewable generation, coal still plays a major role in Inner Mongolia’s electricity system. One academic estimate cited in research on Ulanqab put coal’s share of local electricity at around 37 percent.

That reality complicates the clean-energy narrative. Data centers operate continuously, and operators often value the reliability of fossil fuel-based supply even when they are also adding renewables to the mix. For now, the region’s AI buildout appears to depend on a combination of coal, wind, and solar rather than a fully renewable grid.

What makes water the biggest local risk?

Water may become the most immediate constraint on Ulanqab’s AI ambitions. The city is extremely dry, with annual rainfall of roughly 14 inches, a level comparable to Denver’s climate. That makes water management a sensitive issue even before additional industrial demand is fully online.

The local supply system is already under pressure from residents’ needs. Last month, the city’s water company reportedly shut off several waterworks for seven hours each night to manage peak demand, underscoring how tight the margin already is.

Data centers in the region will not need the same volume of cooling water year-round. According to local weather data, extra water demand for cooling is mostly concentrated in just two months of the year. Even so, the scale of the expansion could still create a serious environmental challenge as more facilities are completed.

Why cooling still matters in a cold climate

Cold winters help reduce air-conditioning loads, but they do not eliminate water use. Servers, power systems, and associated infrastructure still require cooling management, and some forms of liquid cooling can be water intensive.

That is one reason Ulanqab’s future will depend not only on cheap power and low latency but also on whether local authorities can balance industrial growth against basic civic supply. The challenge is especially acute in a dry region where competing demands can quickly become visible.

A timeline of Ulanqab’s rise

The city’s transformation did not happen overnight. It unfolded in stages, beginning with generic data infrastructure and gradually evolving into a strategic AI hub.

How does Ulanqab compare with American AI infrastructure?

Ulanqab is still smaller in global strategic influence than the largest U.S. AI infrastructure projects, but its announced scale is striking. The combined 12.5-gigawatt capacity pledged for the city exceeds the 10-gigawatt target associated with OpenAI’s Stargate Project, at least on paper.

That comparison should be treated carefully. Announced capacity is not the same as completed capacity, and not every project reaches full buildout on schedule. Still, the numbers show that China’s domestic AI industry is thinking in very large increments.

For years, the U.S. had the clearest lead in large-scale compute investment, particularly through cloud giants and frontier model developers. The Ulanqab boom suggests China is now scaling its own infrastructure ambitions more aggressively, especially as domestic AI products start attracting paying users.

What does this mean for China’s AI race?

The main implication is that Chinese companies are beginning to close one of the biggest structural gaps in the AI industry: access to compute. Models alone do not define AI leadership; the power, land, network capacity, and cooling systems behind them matter just as much.

By building in Ulanqab, Chinese firms are gaining not only cheaper capacity but also more control over the resources needed to train and serve models. That can improve resilience, lower costs, and reduce dependence on outside cloud platforms.

It also shows that China’s AI competition is becoming more geographically distributed. Instead of being concentrated in coastal megacities, some of the country’s most important new digital assets are rising on remote grasslands once associated with livestock and coal. The shift is a reminder that AI infrastructure follows the economics of power, water, land, and latency — not just software talent.

What happens next?

The next phase will depend on whether the local ecosystem can solve the resource constraints that come with scale. Power is affordable today, but grid composition may evolve. Connectivity is already strong enough for many uses, but demand may keep rising. Water remains the clearest bottleneck.

If the city can manage those pressures, Ulanqab may become a template for how China industrializes AI infrastructure away from the coast. If it cannot, the boom could face delays, curbs, or more complicated environmental trade-offs.

For now, though, the city stands out as one of the clearest examples of how the AI era is remaking the map of Chinese industry. What was once a peripheral inland region is now central to one of the country’s most important technological priorities.