China Moves AI Compute to Inner Mongolia to Slash Costs and Carbon

China is aggressively relocating its energy-intensive AI computing infrastructure from coastal cities to the interior, centering a new industrial pivot in Inner Mongolia. This strategic migration aims to solve the critical bottleneck of power availability for AI training while leveraging the region's natural advantages.

At the heart of this expansion is Ulanqab, a key hub in the national 'East Data, West Computing' (EDWC) initiative. The region recently saw the launch of the Envision Galaxy Campus, a facility designed to support up to 1 million AI accelerators with a planned total capacity of 2 gigawatts. The campus's main building alone spans 120,000 square meters, marking one of the largest single-site AI computing footprints globally. Further scaling is expected as AI startup DeepSeek plans to establish its own massive data center in Ulanqab with 1 gigawatt of compute capacity.

The Logic of the Interior

The shift is a direct result of the EDWC initiative proposed in early 2022. The strategy seeks to redistribute digital capacity across eight national hubs to alleviate the soaring electricity demands of AI training and inference in densely populated eastern provinces. By moving workloads to the west and north, China can utilize cheaper land and a cooler climate that naturally reduces the energy required for heat dissipation.

Ulanqab specifically employs a 'source-grid-load-storage' model to manage its power needs. This integrated approach allows the region to maintain a high percentage of sustainable energy, with 67% of its data center electricity generated from green sources. This alignment of compute and renewable energy is essential for maintaining the scale required by modern large language models.

The Industrial Impact

This transition represents a fundamental shift from the 'storage era' of data centers—which acted as passive power consumers—to an 'AI-training era' where facilities require custom-designed power grids. By integrating power system engineering, such as dedicated wind farms, directly into the AI infrastructure, Chinese firms are attempting to lower the cost of computing.

In an emerging 'token economy,' where the cost per token determines the competitiveness of AI services, reducing the overhead of electricity and cooling is a primary strategic goal. By treating compute as a resource tied to energy production, China aims to position itself as a global 'token factory,' providing high-volume AI services at a lower cost than competitors tied to expensive urban grids.

Future Outlook

As more firms pivot toward Ulanqab, the focus will shift toward the stability of these massive power loads and the efficiency of data transmission back to the eastern economic centers. While the physical infrastructure is scaling rapidly, the long-term success of the EDWC strategy depends on whether the latency between the interior hubs and the coastal users remains low enough to support real-time AI applications.