As we enter the digital era, AI (Artificial Intelligence) applications have experienced explosive growth, driving China's daily average Token invocation volume to exceed 140 trillion in March this year, more than 1,000 times that of two years ago. However, various computing power infrastructures were previously mostly built independently by regions, forming numerous "computing power islands" with scattered resources and diverse standards.
Recently, the 2026 China Computing Power Conference opened in Langfang, Hebei Province. At the opening ceremony, the China Computing Power Platform announced the realization of integrated national computing power coordination and monitoring, orderly promoting the connection of platforms from 31 provinces, autonomous regions, and municipalities across the country into the network. A development pattern of "one network, one game of chess, and integration" has basically taken shape. Moving from building computing power well to utilizing it effectively, and achieving the organic integration of resources, networks, applications, and services, has become the industrial focus for the next stage.
Computing Power Network: From Integration to Operation and Then to Revitalization
A computing power network connects data centers, supercomputing centers, and intelligent computing centers scattered across the country with high-speed networks for on-demand usage.
Zhong Zhihong, Chief Engineer of the Ministry of Industry and Information Technology, stated in his speech at the conference: "We will accelerate the construction of a multi-level and hierarchical computing power architecture of 'hub-regional-edge', continuously improve the functions of the China Computing Power Platform, build high-speed interconnected, ultra-low latency, and highly resilient computing power transmission channels, and deploy metropolitan millisecond access on demand."
A landmark event at the 2025 China Computing Power Conference was the China Computing Power Platform completing the access of sub-platforms from 10 provinces, achieving integration among platforms. This year, this number has grown to 31. "The China Computing Power Platform has orderly promoted the connection of platforms from 31 provinces across the country into the network, built a unified, standardized, and collaboratively efficient national computing power situation awareness and monitoring system, and realized the ability to clearly understand the national computing power baseline, perceive the situation, and monitor operations," introduced He Baohong, Chief Engineer of the China Academy of Information and Communications Technology, at the conference.
At the opening ceremony of the conference, the National Computing Power Interconnection Regional (Hebei) Node was officially launched, becoming the first regional node activated in the country. This marks that Hebei has opened up a key channel for accessing the unified national computing power market. These advancements will propel China's computing power coordinated governance into a new stage of "full-domain monitoring, whole-network collaboration, and full-chain scheduling."
At the operational service level, the Computing Power Supermarket has become a key instrument. In May 2026, the "Computing Power Supermarket · SME Zone" of the China Computing Power Platform went online and began operation in the Xiong'an New Area. It implemented diversified payment methods such as billing by "GPU-hour", "core-hour", and Token, and focused on three directions: matching computing power supply and demand, empowering industry scenarios, and implementing policy subsidies, to create an "instant selection and use" computing power service portal for small and medium-sized enterprises. As of now, the operational layer of the Computing Power Supermarket on the China Computing Power Platform has gathered over 10,000 registered enterprise users, hosted more than 200 computing power service providers, listed over 2,000 computing power products, and connected more than 300 various large models. It provides computing power invocation services for the vast number of developers, and the platform has accumulated billions of pieces of computing power monitoring big data.
However, from a national perspective, resource revitalization remains a practical challenge. The "2026 Comprehensive Computing Power Panoramic Analysis" shows that only the intelligent computing centers in 7 provinces nationwide have a deployment rate exceeding 75%. This means that the construction progress of computing power facilities in most regions is faster than the growth of actual business demand, posing a certain risk of resource idleness. Industry insiders estimate that there are over 500 intelligent computing centers in use nationwide, with an actual average utilization rate of only 20% to 30%, while an intelligent computing center usually needs to reach a deployment rate of around 40% to cross the break-even point. In contrast, for leading manufacturers, Alibaba CEO Eddie Wu said at the earnings call that there is hardly a single idle GPU card in Alibaba Cloud servers. However, on a national scale, many intelligent computing centers are still in a state of vacancy or struggling to cover costs.
Notably, the boundaries of the computing power network are extending from the ground to space. During the conference, the Beijing Economic-Technological Development Area, Tianjin Economic-Technological Development Area, Xiong'an New Area, and Langfang City jointly initiated the proposal to "jointly build the Beijing-Tianjin-Hebei Space Computing Power Industrial Corridor." The Beijing E-Tech Development Area is responsible for satellite and rocket manufacturing to enable computing power to go to space; the Tianjin E-Tech Development Area is based on supercomputing to promote the integration of supercomputing and space computing power; the Xiong'an New Area relies on China SatNet to connect satellite internet and the computing power network; and Langfang utilizes its low-latency advantage around Beijing to serve as a connection point for space-ground collaboration.
Token Economy: Banks Start Lending Based on Tokens
In September 2026, the first batch of "Yiqi Token Loan" business was implemented in the Beijing E-Tech Development Area. Six banks provided credit support totaling nearly CNY 2 billion to multiple AI (Artificial Intelligence) industry chain enterprises in the area. The so-called "Token Loan" refers to taking the enterprise's Token production and consumption, as well as the value of computing power service contracts, as the core credit basis, exploring a new path of sci-tech finance that empowers credit approval with computing power data.
Why do banks look at Tokens? This starts with Token becoming the value measurement unit of the AI industry. The National Data Administration has given the official Chinese translation of Token as "Ciyuan". As of March this year, China's daily average Token invocation volume has exceeded 140 trillion, an increase of more than 1,000 times compared to the beginning of 2024.
Calling it "Ciyuan" is not just a translation issue; it means that the computing power industry now has a unified accounting unit. In the past, the units for measuring value in the computing power industry were "GPU-hour" and "core-hour", which are supply-side units. However, a Token is the smallest unit for large models to process information. Every AI dialogue and every content generation involves invoking Tokens. It measures the value flow on the demand side and the delivery side. Li Ji, Chairman of the Information Security Branch of the China Computer Users Association, explained: "Pioneering the cross-chip Token measurement benchmark and switching the pricing unit from 'GPU-hour' to 'effective Token' is to establish unified pricing rules for the computing power network."
However, quantity is not the only standard for measuring Token consumption. "The emergence of agents has driven Token consumption. With technological progress, Token consumption is not about taking pride in large quantities, but in efficiency," said Wu Hequan, Academician of the Chinese Academy of Engineering, at the conference.
The surge in Token consumption is also forcing changes in the supply side of computing power. Zou Xuan, Vice President of Shanghai Iluvatar CoreX, proposed at the main forum of the conference: "In the next three to five years, intelligent computing power will continue to maintain a state of vigorous demand and insufficient supply." The popularization of agents directly drives a surge in Token consumption. How to fill the gap in intelligent computing power? Part of the answer lies in computing power, and part lies in electricity.
Computing-Power and Electricity Coordination: Accelerating from Concept to Implementation
In 2026, "computing-power and electricity coordination" was written into the Government Work Report for the first time and included in the outline of the 15th Five-Year Plan. Four departments jointly issued the "Action Plan on Promoting Two-Way Empowerment of AI and Energy," deploying ten aspects of tasks including computing-power and electricity coordination. Liu Liehong, Director of the National Data Administration, defined it as "a new infrastructure project to promote dynamic matching and optimized allocation of resources," with the goal shifting from "reducing PUE" to "a virtuous cycle of strengthening computing with electricity and promoting electricity with computing."
At the High-Quality Development Forum on Computing-Power and Electricity Coordination of the conference, the China Academy of Information and Communications Technology released three achievements: the first batch of the computing-power and electricity coordination project library included 54 industry practice samples; the Computing-Power and Electricity Coordination Public Service Platform 2.0 provides services such as data monitoring, coordinated scheduling, and industrial docking; and the computing-power and electricity coordination evaluation and certification faces computing power centers, scenario verification, and application platforms, carrying out special certifications such as green power access and energy efficiency optimization. This marks that computing-power and electricity coordination is moving from conceptual planning to substantive implementation.
Yue Hao, Senior Expert of State Grid Jibei Electric Power Company, stated: "The computing power industry has a high proportion of green power, becoming a key supporting force for absorbing green power." Local governments and enterprises are also actively exploring: Ulanqab has built the country's first source-grid-load-storage integration project that realizes direct connection of green power, achieving "determining the source by the load and the source following the load"; China Unicom's Sanjiangyuan Green Power Intelligent Computing Integration Demonstration Park generates 10 million kWh of green power annually, achieving 100% consumption within the park, and is the largest intelligent computing center currently in operation on the Qinghai-Tibet Plateau; State Grid Jibei Electric Power's new energy installed capacity has exceeded 82 million kW, and green power transactions have accumulated over 100 billion kWh during the "14th Five-Year Plan" period.
How to meet the requirements of high-proportion green power usage in computing power centers faces practical difficulties in terms of trading and consumption. Su Jian, Chief Engineer of the Energy Usage Institute of China Electric Power Research Institute, pointed out at the forum that the average construction period of an intelligent computing center from commencement to production is about 18 months, while the construction cycle on the power supply side is generally more than twice that, and the actual construction cycle of supporting power transmission and transformation projects requires at least 2 to 3 years. The mismatch between the construction cycles of computing power and electricity is a practical challenge currently faced by computing-power and electricity coordination.
The core value of computing power lies in its practical application. In the next three to five years, computing power will gradually move from dedicated resources to inclusive implementation, allowing more entities to afford and use it well. Future competition in the computing power industry does not lie in who builds the largest scale, but in who can weave a more connected network, make service pricing more transparent, and integrate the computing power network more deeply with the new power grid and the new generation of communication networks.
Author | Liu Lu
Editor | Wu Lilin
Art Editor | Ma Liya Supervisor | Lian Xiaodong