Ulanqab – a city several hundred kilometers northwest of Beijing whose Mongolian name means “red cliff pass” – was known for its windswept pastureland, where cattle grazed and potatoes grew under long hours of sunlight. Its economy hinged on the weather, driven by the cool, strong breezes sweeping through the Yinshan mountain range. Today, giant wind turbines stand next to livestock across those sloping grasslands. As the global artificial intelligence boom fuels a scramble for more computing power and electricity, the same climate that has sustained Ulanqab’s pastoral economy is turning this Inner Mongolian city into one of China’s most coveted industrial assets.
Earlier this month, Chinese renewable energy company Envision Group brought its Galaxy Campus online in Ulanqab. The AI computing facility has, at its heart, a 120,000-square-meter AI data center, roughly the area of 20 standard football pitches. Designed to accommodate up to one million AI accelerators, the campus has a planned power capacity of over two gigawatts, a scale that would make it one of the world’s largest AI computing facilities. For comparison, Colossus 2, the massive U.S.-based AI data center run by Elon Musk’s SpaceX AI, has a projected 1.5-gigawatt IT power capacity, according to a ranking by consultancy Epoch AI. Envision selected the Ulanqab location because it could pair computing infrastructure directly with renewable power, with the Galaxy Campus partly powered by Envision wind farms. Ulanqab accounts for about one-tenth of China’s total wind resources and one-third of the Inner Mongolia autonomous region’s. “Much of that wind once simply blew past, wasted,” said Yang Wenlong, Project Director at Envision Group. “Now it can be turned into electricity for computing, and then into tokens.” About 67% of Ulanqab’s energy was powered by green electricity, including solar power, Yang noted.
Ulanqab’s shift from raising cattle and potato harvesting to the AI frontier has followed a decade-long path. In 2013, Huawei Technologies began working with the Ulanqab municipal government to expand its data center. Five years later, Apple announced plans to host its iCloud service on data centers in the city, the same year Chinese cloud platform UCloud began building its own facility in Ulanqab. Short video firm Kuaishou initiated construction of a data center in 2020, and completed the first phase designed for 300,000 servers several years later. Traditional data center operators VNET, Centrin Data and Chindata have since added green, AI-focused computing projects in Ulanqab, and Hangzhou-based AI model start-up DeepSeek was reportedly set to add one gigawatt of computing capacity in the city earlier this year.
As of June, Ulanqab had signed 89 data center deals, 86 of which were AI computing projects, according to local government figures. The city has a combined 12.5 gigawatts of capacity under development or planned, three times the 4.2 GW projected for Johor, the southern state of Malaysia that has emerged as one of the largest data center sites in the Asia-Pacific region, according to Goldman Sachs. “We are no longer selling computing power; we are selling tokens,” said Liu Huansuo, Deputy Director of Ulanqab’s Big Data Management Bureau. China’s appetite for tokens has exploded in recent years. Daily use rose from about 100 billion at the beginning of 2024 to 100 trillion by the end of 2025, before exceeding 140 trillion in March 2026. This surge in token demand is driving a greater need for the electricity and computing infrastructure required to produce them. For Ulanqab, the wind blowing across the grasslands is no longer merely an energy resource – it helps to power an AI query. “Every one million tokens generated requires two to three kilowatt-hours of electricity,” said Luo Xiyong of data center operator Zhonglian Data. In Ulanqab, data center operators have access to some of China’s cheapest power. Rates vary by supply arrangement, but some operators have secured electricity for about CNY0.35 per kilowatt-hour, less than half the roughly CNY0.70 charged in eastern provinces. Electricity can account for 60% to 70% of a data center’s operating costs, making even modest savings significant.
At full utilization, a one-gigawatt AI computing center might consume as much as 8.76 billion kilowatt-hours of electricity a year. Cutting the electricity cost by just CNY0.1 per kilowatt-hour might reduce a center’s annual power bill by about CNY876 million. For China’s burgeoning AI industry, which is still searching for viable ways to monetize its models, cheap power is no marginal advantage – it is a central part of the business model. But it is only half the equation. The electricity consumed by computing equipment ultimately becomes heat that has to be removed. That is how Ulanqab’s cold, dry climate provides another advantage. With an annual average temperature of about 5 degrees Celsius, the city can use outside air as a natural cooling resource for much of the year, reducing reliance on energy-intensive mechanical systems and leaving more electricity available for computing.
Power usage effectiveness (PUE) measures the ratio between a data center’s total electricity consumption and the electricity delivered to its servers. A PUE of 1.0 means every watt goes directly to computing, while a higher figure indicates more power is consumed by supporting systems such as cooling and power conversion. Leading data-center clusters in the region have a PUE below 1.15, against a national average of close to 1.5, meaning that more of each watt reaches its chips. A difference of only a few cents can translate into hundreds of millions of yuan in savings at gigawatt scale. Those efficiencies matter because the amount of electricity consumed by computing has been rapidly rising, the South China Morning Post reports.
