Since its full launch in 2022, China’s “East Data, West Computing” initiative has become the backbone of the country’s computing network. By the end of July 2026, China’s total AI computing capacity had reached 2.45 million PFLOPS on an FP16 basis, equivalent to about 2,450 EFLOPS. For comparison, the Ministry of Industry and Information Technology reported that nationwide AI computing capacity stood at 2,185 EFLOPS (FP16) at the end of June 2026, up 177% year on year. The eight national computing hubs and three compute–power coordination development zones together account for more than 85% of the country’s completed AI computing capacity, with AI computing representing over 80% of their capacity on average.
Note: This article compiles the latest available computing-capacity data for the eight national hubs. Sources include the Ministry of Industry and Information Technology, the National Development and Reform Commission, the National Data Administration, the State Information Center, and authoritative regional media. All figures are drawn from publicly available online information and are provided for reference.
- Beijing–Tianjin–Hebei Hub (Starting Cluster: Zhangjiakou)
The Beijing–Tianjin–Hebei hub is the largest of the eight hubs by computing capacity and one of the fastest-growing. As of June 2026, Hebei Province had 556.1 EFLOPS of AI computing capacity and 2.504 million standard racks in operation, ranking first nationwide on both measures. Its AI computing capacity represented about 25.4% of the national total. In Zhangjiakou, the hub’s starting cluster, 49 data centers were in operation as of May 2026, with more than 810,000 standard racks and total computing capacity of 30,100 P, including 14,690 P of AI computing capacity.
In addition, once Phase I of three major AI computing projects in Chicheng County, Zhangjiakou, is completed, they are expected to provide more than 200,000 P of computing capacity. Leveraging wind and solar resources in Zhangjiakou, Langfang, and other areas, Hebei is emerging as one of China’s largest green-computing supply bases.
- Yangtze River Delta Hub (Starting Cluster: Wuhu)
Wuhu serves as the starting cluster for the Yangtze River Delta hub. As of July 2026, the city had built 224,700 standard racks and reached 64,000 P of AI computing capacity, while Anhui’s provincial computing platform had aggregated more than 71,000 P of AI computing capacity. The Wuhu cluster officially entered service in 2024. Earlier references to “50,000 racks already in operation” reflected an early planning figure; the cluster has since expanded rapidly.
Wuhu has now signed 15 major projects with combined investment exceeding RMB 270 billion. Huawei Cloud’s East China (Wuhu) Data Center is already in service and is planned to support up to 3 million servers. Wuhu has set a 2027 target of 45,000 P of AI computing capacity.
- Guangdong–Hong Kong–Macao Greater Bay Area Hub (Starting Cluster: Shaoguan)
Shaoguan has become one of the fastest-growing hubs and is now the largest computing cluster in South China. As of August 2026, 23 computing projects had signed agreements to locate there, representing total investment of RMB 111.6 billion. More than 1.66 million standard racks are planned, 45,000 P of AI computing capacity has been deployed, six 10,000-accelerator clusters have been completed, and the city has developed capacity to host 170,000 standard racks.
Latency is another key advantage. One-way latency from Shaoguan to major Greater Bay Area cities such as Guangzhou and Shenzhen is only 1.3–1.66 milliseconds, supported by 13 completed 400G all-optical computing-network links. On the demand side, 3.1 GW of computing-capacity orders are expected to generate more than RMB 50 billion in supporting MEP demand and a server-procurement market worth RMB 500 billion.
- Chengdu–Chongqing Hub (Starting Clusters: Tianfu and Chongqing)
The Chengdu–Chongqing hub follows a dual-cluster model. In Sichuan, 139 data centers have been built, providing 48 EFLOPS of available computing capacity, while Jianyang has surpassed 15,000 P of AI computing capacity. In Chongqing, signed cross-regional computing agreements total 19,000 P. The latest official disclosure (Yu Big Data Letter [2025] No. 272) reports 350,000 racks completed or under construction, 3,000 P of general-purpose computing capacity, 3,300 P of AI computing capacity, and 50 P of supercomputing capacity.
- Inner Mongolia Hub (Starting Cluster: Horinger)
Horinger has moved away from counting servers and now reports primarily in terms of computing capacity. As of 2026, Hohhot’s total computing capacity had exceeded 150,000 P, with AI computing accounting for more than 97%. Across Inner Mongolia, total computing capacity reached 345,000 P, including 326,000 P of AI computing capacity.
Green computing is Horinger’s central competitive advantage. Data centers there use green electricity for more than 85% of their power needs, and the cluster’s green-computing index has ranked first nationwide for four consecutive years, making it a benchmark for low-carbon development under “East Data, West Computing.” The 2026 Green Computing Power Conference also saw 13 major projects signed, with combined investment of RMB 136.1 billion.
- Guizhou Hub (Starting Cluster: Gui’an)
Guizhou is one of China’s most concentrated regions for AI computing resources. The latest figures show provincial computing capacity exceeding 176 EFLOPS, with AI computing accounting for more than 98%. Fifty key data centers are under construction or in operation across the province, including 27 large-scale or larger computing facilities. Gui’an New Area, the core hosting zone, had exceeded 113 EFLOPS of computing capacity as of May 2025, with AI computing representing more than 98% and domestically produced equipment accounting for about 90%, making it one of China’s strongest regions for domestically supplied AI computing capacity.
One point requires clarification: earlier claims that Gui’an already had “1 million servers in operation and 200,000 racks” were planning figures. As of August 2023, the actual figures were approximately 645,000 servers in operation and about 85,000 standard racks completed.
- Gansu Hub (Starting Cluster: Qingyang)
Qingyang is one of the eight hubs with the fastest growth, the largest incremental additions, and the highest share of AI computing. As of August 2026, its AI computing capacity had reached 215,000 P, close to 10% of the national total, with AI computing accounting for 99% of local capacity. The figure is expected to exceed 300,000 P by year-end.
Under its “green-power aggregation” model, Qingyang sources more than 80% of its electricity from green power. Plans call for 300,000 additional 2.5 kW standard racks by 2025 and 800,000 by 2030, positioning Qingyang as a representative green-computing hub in western China.
- Ningxia Hub (Starting Cluster: Zhongwei)
Ningxia has built 130,000 standard racks province-wide, with computing capacity of approximately 31,000 PFLOPS on an FP16 basis. Its composite computing-power index has entered China’s top ten for the first time, while the Western Cloud Base in Zhongwei can accommodate more than 650,000 servers. In the first half of 2026, Ningxia signed eight new hyperscale AI computing center projects with combined investment of RMB 188.5 billion.
Ningxia’s target for 2027 is 800,000 racks and 800,000 PFLOPS of computing capacity, with AI computing accounting for at least 80%.
Key Notes on the Data
- Distinguish the reporting basis. “Installation capacity,” “planned scale,” and “actually commissioned/powered-on capacity” are different concepts. For example, Gui’an’s “1 million servers” is a planning figure, while about 645,000 were actually in operation. Likewise, Wuhu’s “50,000 racks” was a planning figure, whereas only 21,000 were actually in operation at the relevant time. Always state the reporting basis when citing these numbers.
- Be careful with unit conversions. 1 EFLOPS = 1,000 PFLOPS. Unless otherwise stated, “P” in this article refers to the FP32 basis. Ningxia’s 31,000 PFLOPS figure is reported on an FP16 basis and should not be compared directly with figures from the other hubs.
- Watch the disclosure date. AI computing capacity expanded rapidly from 2024 to 2026, and figures for individual hubs can change substantially within six months. This article gives the disclosure date for each data point; future citations should use the latest official figures available at the time.
The data in this article are based on public disclosures from the Ministry of Industry and Information Technology, the National Data Administration, provincial and municipal big-data authorities, and official media, and are provided for industry reference only.
🔥 At DITExpo 2026, taking place in Beijing on December 10–11, representatives from the country’s eight national hubs and ten major clusters have been invited to discuss the development and achievements of “East Data, West Computing.” Follow DITExpo for further updates.
