Digital News

Advancements in artificial intelligence (AI) have pushed computing power to the forefront as an essential foundational resource. One pressing challenge is how to enable computing resources across different regions to work together efficiently and be accessed on demand.

China’s 15th Five-Year Plan (2026-2030) outlines the need to further advance the East Data, West Computing initiative, build a multi-tier computing infrastructure system, and establish a national integrated computing power network.

Today, the construction of this national integrated computing power network is advancing on multiple fronts. But what exactly is a computing power network? What are its key features, and how should it be built? Our reporter sought answers to these questions.

The computing power network doesn’t move or schedule abstract “computing power”; rather, it handles data and computational tasks.

At the First People’s Hospital of Jingzhou in Hubei Province, radiologist Zhang Liren uploaded computed tomography (CT) images into an AI-assisted diagnostic system. In less than 10 seconds, a preliminary assessment of a patient’s condition was generated.

Such remarkable speed gains rely on robust network support. China Mobile Group Hubei Co., Ltd. (Hubei Mobile) recently unveiled its Lingban Computing Power Network platform, engineered to build a multi-level latency coverage system. It delivers a 1-millisecond latency for intra-city access, 5 milliseconds across the province, 7 milliseconds between Wuhan and key cities along the Yangtze River Economic Belt, and 10 milliseconds linking Wuhan to the nation’s eight national computing hub nodes.

Previously, computing tasks were sometimes processed locally and sometimes dispatched hundreds of kilometers away, resulting in delays that users simply had to endure.

Hubei Mobile has deployed 27 municipal-level data centers and 42 computing nodes, either operational or under construction, across the province. “By integrating municipal edge computing resources and bringing idle social computing capacity under unified management, combined with an intelligent computing power scheduling platform, AI startups can access fast computing services without heavy upfront investment,” said Gong Jian, head of the Changjiang Research Institute under Hubei Mobile.

This local example s a broader challenge in building a national integrated computing power network. China’s computing power resources are unevenly distributed across both geography and time: demand is concentrated in the east, where infrastructure costs are high, while the west has abundant resources but relatively limited demand. Demand for computing power resources also fluctuates significantly across industries and over time. “As a result, computing tasks need to be proactively routed based on urgency, budgets, and required chip types, so they can be executed at the most appropriate locations,” said Zhang Huawei, product manager of China Mobile’s computing power network scheduling platform.

While electricity flows through the power grid and water through the water network, a computing power network works differently. It doesn’t move or schedule abstract “computing power”; rather, it handles data and computational tasks. These data and tasks are transmitted across the network to locations where they can be processed using available computing infrastructure and processing capacity.

Different tasks require different types of computing resources. “To use a highway analogy, ambulances use emergency lanes while delivery trucks stay in regular lanes. In practice, applications such as remote surgery occupy the equivalent of emergency lanes, whereas large-model training and film rendering can wait in line and utilize resources with less stringent latency requirements,” said Zhang Xianghong, head of the data infrastructure standards working group under the National Technical Committee 609 on Data Standardization Administration of China. The national integrated computing power network serves as a digital infrastructure that uses information networking technologies to facilitate the nationwide integrated scheduling and operation of computing resources at high utilization and large scale. “Put simply, it ensures that every computing task finds the most suitable lane.”

Through large-scale deployment and centralized development, it serves as a “reservoir” for the national supply of computing power.

A national integrated computing power network represents not only a technological breakthrough but also a fundamental restructuring of how computing power, data, networks, electricity, and other resources are integrated and allocated.

Inside the monitoring system of a national integrated computing power network monitoring and scheduling test platform at the Pengcheng Laboratory in Shenzhen, Guangdong Province, real-time data on computing power resources nationwide is displayed clearly on screen.

Deng Qing, director of the computing power network ecosystem at Pengcheng Laboratory, said the monitoring system provides a clear picture of intelligent computing capacity and resource distribution across regions. Currently, the system covers the 10 major clusters within the eight national computing hub nodes established under the East Data, West Computing initiative, as well as computing resources from certain non-hub regions. At present, 1.37 million PFLOPS of intelligent computing capacity has been incorporated into the monitoring system, accounting for approximately 72 percent of China’s total intelligent computing capacity.

“The monitoring system serves as the eyes of the computing power network. It is a prerequisite for efficient scheduling, resource optimization, and decision-making. Because computing resources vary significantly across different architectures, visibility is the first requirement. Only when integrated real-time monitoring provides data on resource distribution, workload, and utilization can resources be effectively scheduled and used efficiently,” Deng Qing said.

Inside the server halls of China Mobile’s Qingyang Data Center in Gansu, rows of black server cabinets run at full capacity. This year, 22 new data center buildings are planned in Qingyang, and by the end of the year, the region’s computing capacity is expected to surpass 200,000 PFLOPS. In 2021, Qingyang was approved as one of the eight national computing hub nodes under the East Data, West Computing initiative, even though it had not yet developed any large-scale server halls. By the end of last year, local computing capacity had already exceeded 110,000 PFLOPS.

From a blank slate to a computing hub, Qingyang’s transformation reflects a centralized approach to computing infrastructure development. “Centralized development is a defining feature of China’s national integrated computing power network. It promotes the large-scale deployment of diverse computing power resources, including general-purpose computing, AI computing, and supercomputing, at national computing hub nodes, forming a national ‘reservoir’ of computing power,” said Guo Mingjun, director of the computing economy division at the State Information Center under China’s National Development and Reform Commission.

By the end of March this year, the intelligent computing capacity built within the eight national computing hub nodes of the East Data, West Computing initiative accounted for more than 80 percent of the national total, ing a trend of increasingly concentrated development.

Enable dynamic matching between computing power and electricity, and use green electricity to power green computing.

In Zhongwei, Ningxia, vast photovoltaic arrays stretch across the Gobi Desert. Four dedicated power transmission lines snake along the ridgelines, carrying power from the solar farms to a nearby computing park.

On May 2, China commissioned its first large-scale computing-power and electricity synergy project, the 500 MW photovoltaic power station at the China Datang Zhongwei Cloud Base. The facility is now fully connected to the grid, with an expected annual generation of 970 million kilowatt-hours.

“The power from this solar station is delivered directly to the data center in a point-to-point setup, bypassing the main grid. When the solar panels are offline, wind power supplements the supply, creating a complementary system that ensures round-the-clock coverage,” explained Jin Liang, Senior Engineer for the China Datang Zhongwei New Energy Project. From the perspective of computing applications, AI inference, e-commerce promotions, and other workloads generate highly spiky demand. Meanwhile, western computing hub nodes rely primarily on wind and solar power, whose variable output does not always match demand. “Computing-power and electricity synergy is the key to resolving this mismatch. By dynamically matching computing workloads with power supply and using green electricity to support green computing, we provide essential support for the development of the national integrated computing power network,” Jin Liang said.

Building a national integrated computing power network is also a complex system-level project that involves standards setting, platform construction, and institutional innovation.

Computing resources with different architectures, managed by different institutions, and distributed across different regions can only be identified, managed, and utilized within a unified scheduling framework if common standards are in place. Zhang noted that the National Technical Committee 609 on Data Standardization Administration of China has issued nine technical documents, largely establishing the standard framework for computing power networks. Moving forward, these standards will continue to be refined based on practical experience to further enhance their authority and guiding role.

Unified monitoring and efficient scheduling of computing power resources are key features of the national integrated computing power network. The National Data Administration will accelerate the engineering deployment of a national integrated computing power network monitoring and scheduling platform, enhance computing power access and precise matching capabilities, and better meet the diverse demand for computing power resources from all sectors of society. It will also help reduce costs and entry barriers for small and medium-sized enterprises and promote better coordination of computing power resources across eastern, central, and western regions.

“The goal of building a computing power network is to match supply and demand for computing resources. Issues such as how computing resources should be scheduled, whether a unified scheduling authority is needed, and how benefit-sharing mechanisms should be defined all need to be addressed in practice,” said Zhang Huawei. He added that efforts should be made to explore and establish a sustainable scheduling and operations mechanism. By leveraging mature market-based matching and price discovery mechanisms, the aim is to enable faster matching and more flexible transactions between supply and demand, allowing the national integrated computing power network to operate efficiently.

Rapid growth in AI applications, along with the large-scale deployment of embodied intelligence and other scenarios, is fueling surging demand for computing power. The National Data Administration will focus on the East Data, West Computing Project, the national integrated computing power network monitoring and scheduling system, and computing-power and electricity synergy, while continuing to strengthen the underlying infrastructure to support AI’s rapid growth.

Source: People’s Daily