CITIC Securities has released a research report indicating that the concentrated deployment of billion-yuan projects at the Inner Mongolia computing hub, combined with the long-term plan for 200万P of computing power by 2030 and the trend toward self-reliance in the computing industry, presents clear growth opportunities for the domestic computing supply chain. The rapid rollout of 10,000-card AI computing clusters at the hub, coupled with supply chain security and self-sufficiency requirements, is driving sustained demand for domestic AI chips, general-purpose computing chips, and server systems, thereby boosting demand across the entire upstream chain of chips, boards, and complete machines. Given that the core bottleneck in domestic computing lies in the supply side, the report suggests focusing on upstream chokepoints at this stage.
The key viewpoints of CITIC Securities are as follows:
Computing Infrastructure: The region maintains a leading position nationally in scale, with its green computing advantage continuing to lead the pack. According to the 2026 China Green Computing Conference, the total computing power of the Inner Mongolia Autonomous Region has reached 345,000P, of which intelligent computing accounts for 326,000P, with a comprehensive green electricity utilization rate of over 85% for data centers across the region. Both total computing power and intelligent computing capacity rank among the top nationwide. As the core carrying area for the Inner Mongolia computing hub, Hohhot has seen its total computing power surpass 150,000P, with intelligent computing making up 97% of this. By the end of 2026, computing capacity is expected to exceed 250,000P, and surpass 2 million P by 2030, with the computing structure continuously upgrading toward high-density intelligent computing. A benchmark project, the China Telecom Inner Mongolia Information Park, has completed 24 buildings with 21,000 operational racks and a total IT power of 2-3GW, making it a leading high-standard intelligent computing park in China in terms of scale and green electricity ratio.
On the network foundation front, the conference revealed that the hub has established a 400G cross-hub backbone optical cable network, creating a low-latency computing service circle featuring "2 milliseconds in Hohhot-Baotou-Ordos-Ulanqab, 5 milliseconds in Beijing-Tianjin-Hebei, and 20 milliseconds in the Yangtze River Delta." This fully meets the latency and bandwidth requirements of various scenarios, including large model distributed training, cross-domain computing scheduling, and inference business offloading. Simultaneously, the hub is advancing the construction of national computing interconnection regional nodes, deepening the integration of computing and network capabilities to support unified scheduling of computing resources across architectures, entities, and regions.
In terms of long-term planning, the hub is anchored to two core objectives: first, to build the nation's largest green computing base by 2030, achieving a qualitative leap in computing scale; and second, leveraging its low-cost green electricity and large-scale intelligent computing advantages, to establish a global token supply base. This aims to accommodate high-computing-demand applications such as AI large model training, token production, offline rendering, and scientific supercomputing, transforming computing advantages into industrial value.
Computing-Electricity Synergy: Low electricity prices combined with robust dispatch mechanisms create a dual engine, supporting high-proportion green electricity consumption. On the cost advantage front, according to the conference, leveraging the abundant wind and solar resources in the western Inner Mongolia region and a mature electricity market trading mechanism, the on-grid electricity price for computing enterprises at the hub is approximately 0.33-0.36 yuan/kWh. This is significantly lower than the electricity costs for computing hubs in eastern China, providing a distinct cost advantage for high-energy-consumption businesses such as large model training and token production. Supporting integrated wind-solar-storage power supply projects and a millisecond-level intelligent dispatch system enable dynamic matching between computing loads and new energy output, ensuring 7×24-hour power supply stability while continuously improving the green electricity consumption ratio. On the policy front, local supporting policies for direct green power supply and nearby consumption of new energy provide institutional guarantees for a high proportion of green-powered computing supply.
National Landscape: Intelligent computing power is growing at a rapid pace, and the development of the computing network dispatch layer is accelerating. In terms of national computing supply, according to the conference, as of June 2026, China's total intelligent computing capacity reached 2185 EFLOPS (FP16), with a computing rack rate of 71.4% and total storage capacity of 2.53 ZB. The supply capacity of computing infrastructure continues to improve, with the structure increasingly tilted toward intelligent computing. Regarding computing network construction, the current national focus is on developing the dispatch layer, with the core goal of achieving "a unified national computing network." This involves promoting the standardization and commercialization of computing resources, enabling them to be listed, traded, and dispatched like e-commerce products, thereby further enhancing resource utilization efficiency and optimizing the national matching of supply and demand. The report suggests that the improvement of a unified national dispatch system will further highlight the cost advantages of western green computing hubs, accelerating the transfer of high-computing-demand tasks like large-scale computing and token production to hubs such as Inner Mongolia.
Risk Factors: Demand growth for AI computing power may fall short of expectations, leading to lower-than-expected rack rates and investment returns; the implementation of power supporting facilities and grid expansion may lag behind schedule, limiting the release of computing capacity; intensifying land resource constraints could affect the pace of new project implementation; intensified competition among regional computing hubs could trigger price wars and resource diversion; adverse changes in green electricity trading and electricity pricing policies; the construction progress of the national computing network dispatch system may fall short of expectations; and the commercialization process of the token industry and the implementation progress of signed projects may not meet expectations.
Comments