2,185 EFLOPS, 15.56 million racks, 2,000 EB, over 70 routes...
Individually, each set of figures may seem unremarkable, but when brought together, a clear signal emerges—China's computing power is entering a new phase.
As of June 2026, the number of standard racks in use in China reached 15.56 million, and the scale of intelligent computing power in use reached 2,185 EFLOPS, a year-on-year increase of 177%, with storage capacity exceeding 2,000 EB. Recently, the 2026 China Computing Power Conference was held in Langfang, Hebei. China's computing power is shifting from "scaling up" to "system integration", from "having computing power" to "making computing power highly usable", and from standalone data centers to a truly coordinated and scheduled national computing power network.
From "Building More" to "Computing Better": Computing Power Enters a New Phase of Systematic Development
The rapid development of the computing power industry is inseparable from the forward-looking guidance of top-level design. At the opening ceremony of the conference, government leaders and academician experts, focusing on the construction of the national integrated computing power network, clarified the future direction for the industry from multiple dimensions including industrial direction, technical path, and regional practice, dispelling development uncertainties.
Zhong Zhihong, Chief Engineer of the Ministry of Industry and Information Technology (MIIT), proposed four core guidelines to outline the action framework for the high-quality development of computing power. First, strengthen systematic planning to build a multi-level and hierarchical computing architecture of "hub-region-edge", relying on the China Computing Power Platform to coordinate and monitor resources. Second, strengthen the integration of computing and networking to create high-speed interconnected and ultra-low latency computing power transmission channels, and promote the construction of a computing power interconnection node system. Third, strengthen technological innovation to tackle key challenges in high-end chips and computing-power-electricity coordination, and accelerate the formulation of the computing power standard system. Fourth, strengthen application empowerment to lower the threshold for SMEs to use computing power, transform inclusive computing power dividends into driving forces for the development of the real economy, and support the implementation of the "AI+" initiative.
As a computing power highland around Beijing, Hebei became the focus of the conference with its impressive infrastructure data. According to Zhao Dachun, Vice Governor of Hebei Province, the province's intelligent computing power scale reached 556 EFLOPS (FP16), and the number of racks in use reached 2.5 million standard racks, both ranking first in the country; the comprehensive computing power index has ranked first among the 31 provinces nationwide for two consecutive years. Innovative initiatives such as "computing power vouchers, token vouchers, model vouchers" and the "computing power+" industrial fund have attracted nearly 200 upstream and downstream enterprises to settle in. The computing power utilization rate has exceeded 80%, and 441 vertical large models and 98 intelligent agents have been deployed in 26 industries including steel, chemical, and healthcare, promoting the transformation of computing power resources into industrial value.
Wu Hequan, an academician of the Chinese Academy of Engineering, also "diagnosed" the computing power industry from the technical foundation. He proposed that in the era of intelligent agents, computing power competition is no longer a mere comparison of total computing power, and efficiency first has become the new criterion. The average daily token invocation volume in China has reached 140 trillion. The development of computing power should break away from the misconception that "the more consumed, the better". It is necessary to promote the linkage of computing power clusters in Zhangjiakou, Langfang, and other places, make good use of green power resources, develop a collaborative model of local preliminary computing and remote precision computing, and rely on abundant computing power to spawn new tracks such as AI chips and the intelligent agent economy.
He Baohong, Chief Engineer of the China Academy of Information and Communications Technology (CAICT), also mentioned in his interpretation of the "2026 Comprehensive Computing Power Panoramic Analysis" that domestic computing power presents a pattern of "high-level leadership by a few provinces, central agglomeration by most provinces, and gradient differentiation in different dimensions". Some regions focus on computing power construction while neglecting the supporting storage and transport capacities. In the future, policies should be implemented in categories according to regional endowments, activating existing resources and removing the bottlenecks in transforming resource advantages into service capabilities.
From "Having Computing Power" to "Using Computing Power", the Industrial Chain Redefines the Value of Computing Power
If policies and experts answer the question of "where the computing power industry is going", then operators and upstream and downstream enterprises in the industrial chain answer another question from a practical perspective: how exactly should computing power be built, connected, and used?
As the "main force" in the construction of computing and network integration, operators have formed systematic practical solutions. Liu Guiqing, General Manager of China Telecom, pointed out that China Telecom is comprehensively promoting the cloud reform, digital transformation, and intelligent benefit strategy, and building a "five-in-one" intelligent cloud system. At the level of computing power supply, 16 hundred-MW-level data center parks have been built, and four major domestically produced 10,000-card intelligent computing clusters have been created. The total scale of "self-owned + accessed" intelligent computing exceeds 118.8 EFLOPS. He specifically mentioned the need to deepen the innovation of computing-power-electricity coordination, promote the interconnection of computing power, improve the token production efficiency of domestic computing power, and gather the joint efforts of the industrial chain to promote the construction of the computing power network.
Chen Yangfan, General Manager of China Mobile, introduced that China Mobile is aligning with the national "East Data and West Computing" project deployment, laying out gigawatt-level data parks, and innovating the operating models of computing-power-electricity coordination and computing power enclaves. The total contracted scale of AIDC has exceeded 7GW. At the same time, it continues to optimize the intelligent computing layout of "centralized center, ubiquitous edge, and integrated training and inference", accelerate the construction of inference clusters, and deploy multiple 10,000-card intelligent computing centers and distributed inference networks.
Zhu Hanwu, Deputy General Manager of China Unicom, shared China Unicom's practice in building the foundation of computing and network, implementing the "East Data and West Computing" layout, building 7 hundred-MW-level data center parks, and promoting the large-scale application of liquid cooling technology. In 2026, the proportion of liquid-cooled delivery for newly added cabinets will be about 50%. Over 300 integrated training and inference resource pools have been built to adapt to the diverse needs of the AI era with a green and intensive computing power base.
Meanwhile, enterprises' understanding of the value of computing power is also changing. Zheng Xin, Vice President of YOFC (Yangtze Optical Fibre and Cable), mentioned that YOFC's self-developed HollowBand® hollow-core fiber has optimized attenuation to 0.032dB/km, refreshing the global lowest loss record again. The cumulative delivery exceeds 10,000 fiber-kilometers, and more than 13 commercial and pilot projects have been implemented at home and abroad.
Zhou Chaonan, Chairman of Renge Technology, put forward her observations from the perspective of infrastructure. In 2026, Renge Technology built a single-scale 200-megawatt AIDC intelligent computing center in Langfang, and has currently laid out 9 AIDC intelligent computing parks globally. She believes that the computing power industry is undergoing three major transformations: the demand structure is accelerating the shift from general computing power to intelligent computing power; the value logic is shifting from "scale" to "tokens" as the carrier of industrial value; and the development concept is shifting from simple construction to the coordinated development of computing power and green energy.
From "Computing Power Landing" to "Industrial Take-off", Responding to Value with Practice
To judge exactly what stage an industry has developed to, one cannot only look at planning and investment, but also at whether there are truly tangible, implemented results.
The answers given by the 2026 China Computing Power Conference are quite rich. The conference officially announced that the China Computing Power Platform has achieved the integrated planning and monitoring of national computing power, and the development pattern of "one network, one coordinated strategy, and integration" for China's computing power has basically taken shape.
Currently, the China Computing Power Platform has promoted the docking and networking of platforms in 31 provinces, autonomous regions, and municipalities nationwide. The operational layer of the Computing Power Supermarket has gathered over 10,000 registered enterprise users, more than 200 computing power service providers have settled in, over 2,000 computing power products have been listed, and more than 300 various large models have been accessed. This also means that national computing power resources are moving from "dispersed existence" to "unified perception and coordinated scheduling".
The "2026 Comprehensive Computing Power Panoramic Analysis" was heavily released. The newly upgraded Comprehensive Computing Power Matrix System 5.0 innovatively uses the coordinate matrix atlas to evaluate the computing power development level of 31 provinces, autonomous regions, and municipalities nationwide from four major dimensions: computing power, storage capacity, network transport capacity, and environment. The report directly points out the existing challenges in the industry: obvious gradient differences in regional development, insufficient depth of integration in the computing power industry, and constraints on the implementation of cross-regional scheduling. It is recommended that all localities base themselves on their endowments, coordinate the planning of computing, storage, and transport, improve the efficiency of transforming resources in the central and western regions into services, and promote the upgrading of computing power from a scale advantage to a quality and efficiency advantage.
In addition, the National Computing Power Interconnection Regional (Hebei) Node was officially launched, opening up the key channel for Beijing-Tianjin-Hebei computing power to access the national unified market; the first batch of inclusive computing power empowerment SME development service lists was officially unveiled, targeting the pain points of "expensive and difficult to use computing power" for SMEs to release policy dividends; the "One Network, Three Integrations" computing power financial ecosystem co-construction action was officially launched, with financial capital deeply entering the computing power track to provide financial support for computing power projects and make up for the shortcomings of industrial finance.
Multiple places in Beijing-Tianjin-Hebei jointly launched the "Initiative to Co-build the Beijing-Tianjin-Hebei Space Computing Power Industrial Corridor", exploring a new direction for integrated space-ground computing power. The "China Computing Power · Computing in Langfang" computing power scheduling service platform was officially launched to coordinate and schedule regional dispersed computing power and serve a wide range of market entities. Moving from "single-point breakthrough" to "systematic innovation" is exactly another change happening in China's computing power industry.
Why Can China's Computing Power "Fly"?
China's computing power can truly "fly". What "flies" is the synergy of computing and network integration, breaking down geographical and manufacturer barriers, allowing computing power to be scheduled on demand like water and electricity; what "flies" is the hard power of technological autonomy, tackling key challenges in chips, clusters, and scheduling systems, enabling domestic computing power to move from "usable" to "highly usable"; what "flies" is the transformative power of industrial empowerment, connecting large models, intelligent agents, and thousands of industries, transforming the "0"s and "1"s in data centers into driving forces for the transformation and upgrading of the real economy; what "flies" is inclusive sharing, lowering the threshold for SMEs to use computing power, and allowing computing power dividends to benefit more market entities.
Therefore, what truly makes computing power "fly" is the increasingly close connection among network, technology, industry, application, and ecology. When computing power evolves from a mere infrastructure into a true new quality productive force, the next growth space of China's digital economy is also being "calculated" out bit by bit.
Managing Editor/Layout: Zhu Wenfeng
Proofreader: Mei Yaxin
Supervisor: Liu Qicheng