CITIC Securities: The full chain of 100 billion projects has been implemented, and investment in computing power continues to be booming

The Zhitong Finance App learned that CITIC Securities released a research report saying that the 100 billion computing power hub project in Inner Mongolia was implemented centrally, compounded by the long-term plan of 2 million P computing power in 2030 and the autonomous and controllable trend of the computing power industry, and the domestic computing power industry chain ushered in clear growth opportunities. The implementation of the 10,000 card level intelligent computing cluster at the hub accelerated, superimposing supply chain security and autonomous controllable requirements. Demand for domestic AI chips, general computing power chips, and complete servers continued to be released, driving demand for the entire upstream chip, board card, and complete machine chain. Considering that the current core pain point of domestic computing power is on the supply side, more attention is currently being paid to upstream bottlenecks.

CITIC Securities's main views are as follows:

Computing power infrastructure: The scale ranks first in the country, and the green computing power advantage continues to lead

According to the 2026 China Green Computing Power Conference, up to now, the total computing power of the Inner Mongolia Autonomous Region has reached 345,000 P, of which the intelligent computing scale is 326,000 P, and the comprehensive utilization rate of green power in data centers throughout the region has exceeded 85%. The total computing power and intelligent computing capacity are among the highest in the country; Hohhot, as the core carrying area of the computing power hub in Inner Mongolia, currently has exceeded 150,000 P, accounting for 97% of intelligent computing power. By the end of 2026, the computing power scale will exceed 250,000 P, breaking 2 million P of computing power by 2030. The computing power structure continues to be upgraded to high-density intelligent computing; among them, the benchmark project China Telecom Inner Mongolia Information Park has built 24 buildings, with 21,000 racks put into operation, and the total IT power in the park has reached 2-3 GW, which is the leading domestic scale and leading green power ratio. Count the park.

At the network base level, the conference revealed that the hub has built a 400G cross-hub backbone optical cable network, forming a low-latency computing power service circle of “2 ms in the call package, 5 ms in Beijing-Tianjin-Hebei, and 20 ms in the Yangtze River Delta”, which can fully meet the latency and bandwidth requirements of various scenarios such as distributed training of large models, cross-domain computing power scheduling, and inference business sinking. At the same time, the hub continues to promote the construction of regional nodes for national computing power interconnection, deepens computing network integration capabilities, and supports unified scheduling of computing power resources across architectures, subjects, and regions.

In terms of long-term planning, the hub anchors two core positions: one is to build the country's largest green computing power base in 2030 to achieve a quantum leap in computing power; the second is to build a global token supply base, relying on the advantages of low-cost green power and large-scale intelligent computing to undertake large computing power requirements such as AI model training, offline rendering, and scientific supercomputing, and promote the transformation of computing power advantages into industrial value.

Computing and electricity collaboration: low electricity price+emphasis two-wheel drive, supporting a high proportion of green electricity consumption

In terms of cost advantages, according to this conference, relying on the rich landscape resources and mature market-based electricity trading mechanism in the Mengxi region, the electricity price for hub computing power companies is about 0.33-0.36 yuan/degree, which is significantly lower than the electricity cost of computing power hubs in eastern China, providing outstanding cost advantages for energy-intensive businesses such as large-scale model training and word element production. It is equipped with an integrated wind and solar storage power supply project and a millisecond intelligent scheduling system to achieve dynamic matching of computing power load and new energy output, and continuously increase the green power consumption ratio while ensuring 7×24 hour power supply stability. At the policy level, local authorities support special policies such as direct supply of green electricity and nearby consumption of new energy sources to provide institutional guarantees for the supply of a high proportion of green electricity computing power.

National pattern: The scale of intelligent computing power is growing rapidly, and the construction of the computing power network scheduling layer is speeding up

At the national computing power supply level, according to this conference, as of June 2026, the total scale of domestic intelligent computing power reached 2185 EFLOPS (FP16), the computing power listing rate reached 71.4%, the total storage capacity reached 2.53 ZB, the computing power infrastructure supply capacity continued to increase, and the structure was skewed towards intelligent computing. At the level of computing power network construction, the current national computing power network construction focuses on the development of the scheduling layer. The core goal is to achieve a “national computing power game”, promote standardization and commercialization of computing power resources, so that computing power resources can complete the whole process of listing, trading, and scheduling like e-commerce products, further improve the efficiency of the utilization of computing power resources, and optimize the matching between national computing power supply and demand. We believe that the improvement of the unified national scheduling system will further highlight the cost advantages of green computing power hubs in the west and accelerate the shift of large computing power and word element production businesses to western hubs such as Inner Mongolia.

Risk factors:

Demand for AI computing power fell short of expectations, resulting in lower computing power listing rates and return on investment than expected; power support facilities and power grid expansion and implementation progressed slower than expected, limiting the release of computing power production capacity; increased land resource constraints, affecting the pace of implementation of new projects; intensified competition in regional computing power hubs, triggering price wars and resource diversion; adverse changes in green power transactions and electricity price policies; progress in the construction of the national computing power network scheduling system fell short of expectations; the commercialization process and implementation of contracted projects fell short of expectations.