Woofun AI reports that the epicenter of China's artificial intelligence compute race has shifted unexpectedly from coastal metropolises like Beijing, Shanghai, and Shenzhen to Ulanqab in Inner Mongolia. This region, historically defined by livestock and coal mining, is now rapidly consolidating as a critical node for AI infrastructure, a transformation highlighted by recent coverage in WIRED and financial analysis from Goldman Sachs.

The scale of this infrastructure expansion is unprecedented, with nearly 100 data centers operational or under construction since 2016. Chinese firms have committed to facilities totaling 12.5 GW of capacity, with over 70% of these announcements made in the past year alone. By comparison, OpenAI's Stargate project, backed by a $500 billion investment plan, targets approximately 10 GW upon completion. Consequently, Ulanqab's announced capacity is already a quarter higher than the global benchmark set by OpenAI.

This regional boom is fueled by direct capital allocation from domestic tech leaders. DeepSeek, ByteDance, Alibaba, and REDnote are all actively investing in local data center construction. These entities are moving beyond reliance on third-party cloud services, choosing instead to build proprietary physical infrastructure to secure long-term compute resources for their growing AI workloads.

Structurally, the location offers distinct geographical and climatic advantages. Situated on the Inner Mongolia Plateau, the area benefits from high altitude and cold winters, allowing facilities to utilize natural low temperatures for cooling.

Furthermore, the region's abundant coal resources, combined with rapidly developing wind and solar power, keep electricity costs significantly lower than in other parts of China, directly enhancing the economic viability of energy-intensive hash rate operations.

Woofun AI data shows that network infrastructure upgrades have further eliminated historical barriers to entry. The construction of two dedicated fiber-optic lines in 2017 and 2019 reduced average network latency to below 5 milliseconds. This improvement makes Ulanqab suitable not only for intensive model training but also for AI inference services that require instant data exchange, satisfying the triple requirement of cheap energy, natural cooling, and low latency to major population centers.

The strategic alignment with national policy has accelerated this development. Huawei established its first data center in Ulanqab in 2016, followed by Apple in 2019. In 2021, the area was designated a key node in the "East Data, West Computing" initiative. This policy aims to process data generated in high-cost eastern cities like Beijing and Shanghai using the abundant land, coal, hydro, wind, and solar energy available in the western region.

Andrew Stokols, a professor at Singapore Management University, notes that since the rise of AI in 2022, the industry has recognized that remote data centers are ideal for model training. Unlike search, video, or e-commerce services that demand real-time interaction, training large AI models can take weeks or even months. Therefore, network latency becomes much less critical for these long-duration tasks, validating the use of distant infrastructure for core AI development.

This shift reflects a broader convergence with U.S. industry trends. Previously, Chinese AI companies were more conservative in capital spending, relying heavily on cloud providers. Now, firms like DeepSeek, ByteDance, Alibaba, and REDnote are mirroring American tech giants by becoming infrastructure providers themselves. This transition from model companies to infrastructure owners marks a fundamental change in how AI compute is secured and deployed globally.