Ulanqab Becomes a China AI Compute Hub

Goldman Sachs has identified Ulanqab as one of the Asia-Pacific region’s largest and fastest-growing AI computing clusters, and the numbers underline why. VNET said its Ulanqab projects cover about 1,100 Chinese mu, or 73.3 hectares, and have roughly 1.3 gigawatts of IT power capacity, enough to support clusters of up to 1 million processors. Across Inner Mongolia, 20 of China’s 30 leading private compute companies were present by June 2026, with investment topping 200 billion yuan, or about $30 billion. Regional computing capacity reached 363,000 petaflops in August, nearly quadruple the level less than two years earlier.

For investors, the story is bigger than a local development. It points to a widening second-order trade in AI: not just chips and models, but land, electricity, cooling, fiber and data-center operators. That benefits infrastructure providers and power-linked developers, while reinforcing demand for Nvidia’s accelerators, which still underpin large-scale model training, and for Chinese AI leaders such as ByteDance, DeepSeek and Baidu, which are using Ulanqab’s compute. It also suggests the investment cycle in China’s AI stack is still early, even as the country faces external constraints on advanced semiconductors and ongoing pressure to localize critical hardware.

Ulanqab’s own economy is already feeling the effect. In the first half of 2026, fixed-asset investment in its compute industry rose 39% from a year earlier, electricity consumption jumped 90% and revenue increased 12%. The city has also launched a three-year plan to build a “token city,” aiming to create a full value chain from wind and solar generation to green compute and token generation. That framing matters: China is treating tokens, the unit AI models consume and produce, as a new industrial metric much as oil is measured by the barrel.

The labor impact is equally significant. The compute buildout is pulling young workers back to a region long associated with outmigration. More than 20 compute firms arrived in Ulanqab in 2026, and local industry-education partnerships have trained more than 2,000 AI specialists. For a city once defined by potato farming, AI has become a source of higher-value jobs in operations, maintenance and systems management.

The bull case is that Ulanqab becomes a template for China’s distributed AI infrastructure, lowering training costs and easing bottlenecks in Beijing. The bear case is that the buildout outpaces monetization, leaving exposed capacity, high electricity loads and dependence on a policy-led investment cycle. But for now, the narrative is clear: China’s AI expansion is pushing well beyond the big coastal tech hubs and into inland regions where cheap power, cooler weather and policy support can turn farmland economies into compute economies.