China Launches Computing Power Sales Network Amid Data Center Overhaul
China's Push to Build a Nationwide Computing Power Network
China is taking significant steps to establish a network for selling computing power and manage the rapid expansion of data centers. This initiative comes in response to the surge in local government-backed facilities that have led to an oversupply of capacity, threatening the sustainability of these operations.
A nationwide assessment of the sector is currently underway following a three-year construction boom. According to sources familiar with the matter and a document reviewed by Zona Kreasi, the state planner is closely examining the situation. Additionally, Beijing is exploring the creation of a national, state-run cloud service aimed at utilizing surplus computing power.
The Ministry of Industry and Information Technology (MIIT) is working with China's three major state telecom companies to develop a platform that connects data centers and facilitates the sale of computing power. This move is seen as critical in the broader context of the technological rivalry between China and the United States.
Unused computing power and financially unstable data centers could hinder China’s ambitions in artificial intelligence development. "Everything will be handed over to our cloud to perform unified organization, orchestration, and scheduling capabilities," said Chen Yili, deputy chief engineer at the China Academy of Information and Communications Technology, during a recent industry conference.
Chen did not provide specific details about the cloud service proposal, but his presentation materials indicated that China aims to standardize the interconnection of public computing power nationwide by 2028. However, some analysts remain skeptical due to the technological challenges involved.
The Data Center Boom and Its Consequences
China's data center construction boom began in 2022 after the launch of the "Eastern Data, Western Computing" project. This initiative aimed to coordinate data center construction by concentrating facilities in western regions, where energy costs are lower, to meet demand from eastern economic hubs.
At a June event, Chen mentioned that the industry ministry has licensed at least 7,000 computing centers. A review of government procurement documents showed a significant increase in state investment in data centers last year, totaling 24.7 billion yuan ($3.4 billion), compared to over 2.4 billion yuan in 2023. This year alone, 12.4 billion yuan has been invested, primarily in Xinjiang.
Despite this growth, concerns about returns on investment have led to an increase in project cancellations. While only 11 projects were canceled in 2023, over 100 were canceled in the past 18 months. Utilization rates are estimated to be low, with sources indicating they range between 20% and 30%.
Investors and local governments often build data centers without considering real market needs, expecting government and state-owned firms to act as buyers. A project manager from a server company noted that building data centers in remote western provinces lacks economic justification. Charlie Chai, an analyst with 86Research, added that while lower operating costs are a factor, performance and accessibility issues must also be considered.
Regulatory Measures and Challenges
To regulate the sector's growth, China's National Development and Reform Commission (NDRC) initiated a nationwide assessment earlier this year. This has resulted in stricter scrutiny of new data center projects planned after March 20 and a ban on local governments from participating in small-sized computing infrastructure projects.
The NDRC aims to prevent resource wastage by setting specific thresholds, such as requiring a computing power purchase agreement and a minimum utilization ratio. These measures are intended to filter out unqualified projects. However, the NDRC did not respond to requests for comment.
Industry sources and Chinese policy advisers note that forming a computing power network will be challenging. The technology for efficient real-time transfer of computing power remains underdeveloped. The Eastern Data, Western Computing project aimed for a maximum latency of 20 milliseconds by 2025, necessary for real-time applications like high-frequency trading.
Many facilities, especially those in remote western regions, have not met this standard. Additionally, many centers use different chips from Nvidia and local alternatives like Huawei's Ascend chips, complicating integration into a unified cloud service.
Chen remains optimistic, envisioning a cloud that bridges differences in underlying computing power and physical infrastructure. "Users do not need to worry about what chips are at the bottom layer; they just need to specify their requirements, such as the amount of computing power and network capacity needed," he said.