Cover | Illustration by Pencil News

China's Most Sizzling AI Construction Hub Takes Shape in Ulanqab, Inner Mongolia.

This August, the city of Ulanqab, boasting a permanent population of just over a million, clinched 12 intelligent computing projects in rapid succession. Based on incomplete statistics, confirmed planned investments have surpassed 130 billion yuan. Five disclosed intelligent computing centers boast a combined planned power capacity of 6.3 gigawatts.

But the action doesn't stop there. GDS Holdings is poised to invest over 30 billion yuan in the region over the next five years. DeepSeek, too, has plans to construct a large-scale data center in Inner Mongolia, deploying no fewer than 160,000 Huawei AI chips.

This northern town, once synonymous with potatoes, has abruptly found itself thrust into the epicenter of China's AI race.

Why Ulanqab? More crucially, with investments running into the hundreds of billions, who stands to gain?

- 01 - Arrival of Huawei, Alibaba, DeepSeek

Ulanqab has been a hive of activity lately.

According to an October 12 report by Xinhua Daily Telegraph, 22 computing power projects are currently under construction in the local Chahar High-Tech Zone, employing nearly 30,000 workers. By midday, boxed meal vendors and fruit stalls spring up around the construction sites, and the afternoon rush hour can even bring traffic to a standstill in this small town.

The most striking development is the expanding roster of participating companies.

In late August, 21Vianet, Chindata Group, ChinaUnicom Cloud Data, GLP, and Huayi Cloud inked agreements in Jining District, Ulanqab, with plans to establish intelligent computing centers boasting capacities of 2 GW, 1 GW, 1.5 GW, 1.3 GW, and 0.5 GW, respectively.

That same month, energy company Envision's Xinghe Base commenced operations, with a planned capacity of 2 GW for the entire park. Envision, initially recognized for its wind turbines, energy storage, and energy systems, has now ventured into constructing its own "token factories." Established IDC companies like GDS Holdings and 21Vianet continue to ramp up their investments.

More intriguingly, customers leasing computing power are now stepping directly into the fray.

DeepSeek has also set its sights on Ulanqab. According to informed sources, the AI unicorn intends to construct a GW-scale data center there, deploying at least 160,000 Huawei Ascend 950DT chips. Based on the latest disclosed prices, the nominal procurement cost for these chips alone exceeds 40 billion yuan, excluding supporting infrastructure such as data center space and power.

As early as April, DeepSeek posted job listings for senior data center operation and maintenance engineers in Ulanqab, offering salaries up to 30,000 yuan per month. A large model team recruiting not in Beijing or Hangzhou, but on the vast Inner Mongolian grasslands.

In fact, Ulanqab's ties with internet giants predate the current AI surge.

In 2013, Huawei made the initial move. By 2016, Phase 1 of Huawei's cloud data center became operational, initially housing just over 1,500 server racks to support government clouds and data storage. Subsequently, Alibaba, Kuaishou, UCloud, and others joined the fray.

In 2022, XPeng Motors and Alibaba Cloud established their autonomous driving intelligent computing center, "Fuyao," in the city. XPeng revealed that this 600 PFLOPS center slashed core model training time from seven days to under one hour.

First came data storage and cloud computing companies; then autonomous driving firms; now large model companies. Each technological wave has ushered in new customers for the same data centers.

Yang Zhi from Zhiyun Data informed Pencil News that computing power sellers can be categorized into two groups: those with "hard assets" (GPUs and data centers) and those with "skills" (customer understanding, scheduling capabilities, and utilization optimization). "Those with hard assets earn relatively stable, long-term profits, while those with skills profit from pricing power. These two categories are increasingly overlapping. In the long run, only those with both hard assets and skills can command premiums and sustainable profits."

The current investment surge in Ulanqab underscores companies' efforts to secure their "hard assets."

- 02 - Why This Small Town?

Viewing Ulanqab as a colossal computing power factory, its greatest asset lies not in chips but in electricity.

This city has long depended on electricity for its economic sustenance. By 2025, its ferroalloy industry is projected to exceed 100 billion yuan in output value, with traditional submerged arc furnaces being major electricity consumers. Now, more power is being directed towards servers. While the industrial equipment consuming electricity has evolved, so have the customers Ulanqab vies for.

AI servers must operate round-the-clock. Beyond costly chips, the most consistent expense is electricity—both for server operation and data center cooling. The larger the scale, the more pivotal electricity pricing becomes.

According to data provided by the Inner Mongolia Industry and Information Technology Department to Xinhua Daily Telegraph, the average delivered price of green electricity for local data centers is approximately 0.35-0.38 yuan per kilowatt-hour.

A straightforward calculation: if a park averages 1 million kW in power load and operates continuously throughout the year, annual consumption reaches about 8.76 billion kilowatt-hours. Saving 5 cents per kilowatt-hour would reduce annual costs by about 438 million yuan. While actual project loads vary, it's evident why major customers are concerned about even minor per-unit cost differences.

Ulanqab boasts another natural advantage: cold weather. For about 10 months of the year, outdoor low-temperature-assisted cooling is feasible, reducing air conditioning expenses. Abundant wind and solar resources also facilitate direct green electricity supply projects.

However, China boasts other regions with cheap electricity and cold weather. Ulanqab's additional edge is its proximity to Beijing.

A mere two-hour high-speed rail ride from Beijing, the city enables large model companies—which require frequent server debugging and cluster management—to dispatch engineers for same-day trips. For training and storage tasks not necessitating local data presence, dedicated lines can transmit computing needs to grassland data centers.

These combined advantages have enabled Ulanqab to attract major customers earlier than many western computing power cities.

More crucially, over a decade of accumulation has erected barriers to entry: Huawei's early presence brought data centers, fiber optics, and operations staff; Alibaba, Kuaishou, and XPeng's involvement drove upgrades in power supply, cooling, and networking; more projects mean swifter subsequent construction and delivery.

A company constructing a data center can commence from scratch with land acquisition, power connection, and fiber laying, but cannot replicate a mature industrial cluster.

This is why competition now extends beyond just electricity prices. 21Vianet's Ulanqab project commenced in March 2024 and achieved its first delivery in October the same year; Alibaba Cloud's new modular data center there boasted a delivery cycle of about 100 days, according to National Business Daily.

With AI companies eager to launch products, earlier data center delivery can translate into months of additional computing power sales.

The small town once competing on land and electricity prices now competes on speed and customer resources.

- 03 - After 130 Billion, Who Reaps the Profits?

Behind the investment boom lie two vastly different sets of calculations.

The first is project-based. A Goldman Sachs report in August this year revealed that as of end-June, Ulanqab had committed data center capacities of about 12.5 GW across operating, under-construction, and planned facilities—ten times the actual operating capacity of 1.2 GW expected by 2025.

The second is operational.

Ulanqab reports: In 2025, the city's computing power industry generated 6.835 billion yuan in revenue and 368 million yuan in tax revenue. In the first half of 2026, revenue reached 3.66 billion yuan while data center electricity consumption hit 3.267 billion kilowatt-hours, up 89.52% year-on-year.

Hundreds of billions in project signings correspond to tens of billions in annual industrial revenue.

However, for a city, more data centers don't necessarily equate to more profits retained locally.

Third-party IDC companies typically construct data centers, install power and cooling systems, then lease server racks to cloud providers or internet giants. Some servers are purchased by customers themselves, with models, chips, and sold tokens belonging to those customers. Local revenues often stem from rack rentals, electricity fees, operations income, and some taxes—but not necessarily the most lucrative segments of the AI supply chain.

Liu Huansuo, deputy director of the Ulanqab Big Data Administration, admitted to Xinhua Daily Telegraph that relying solely on low electricity and land prices risks reducing the city to mere "data center landlords" earning meager profits.

Another challenge is whether the constructed computing power will actually find buyers.

One thousand GPUs scattered across different data centers hold different value from one thousand GPUs interconnected at high speed and operating continuously for months. Chip models, networking, software systems, and scheduling capabilities determine whether they can genuinely support large model training and inference.

Currently, all players are addressing their shortcomings. IDC companies are expanding into green power and computing power scheduling; energy firms like Envision are constructing their own data centers; Alibaba is extending its reach from chips and cloud platforms to data center construction. All aim for the same objective: converting each kilowatt-hour of electricity into more, cost-effective computing services that customers will actually purchase.

Ulanqab itself aspires to transcend mere rent collection. The city now aims to become a "token capital," transitioning from rack leasing to computing power scheduling, token trading, and AI applications.

Success hinges on real orders, chip supplies, and data center utilization rates. Constructing power stations doesn't guarantee timely server delivery; filling servers doesn't guarantee long-term customer payments.

Over a decade ago, Ulanqab retained Huawei's servers with cheap electricity and proximity to Beijing. Today, with DeepSeek and other large model companies arriving on the grasslands, the city has secured its ticket to the next round of AI infrastructure competition.

But the 130 billion yuan in signings is merely the starting point. The true test will arrive in a few years: whether servers here operate at full capacity and how much profit from token sales remains locally.

This article does not constitute investment advice.