Regions · China · updated 31 Oct 2026
China: data centres and AI infrastructure
How much computing China has, who is building it, the policy behind it, and what chips and AI models it runs on. For engineers, operators, buyers, consultants and investors. Every figure is dated and linked to its source.
Who is building it: companies and spending
Confirmed · Filed Alibaba Group
RMB67.7 billion (US$9.98 billion) of capital expenditure in the quarter to 30 June 2026, up 75% year on year
Alibaba says the rise reflects AI infrastructure investment, procurement timing, more CPU capacity for AI agents and higher chip component prices. AI Cloud and Compute Services revenue grew 45%. In February 2025 Alibaba committed RMB380 billion over three years to cloud and AI infrastructure.
Sources: Alibaba Group (HKEX filing): Results announcement for the quarter ended 30 June 2026 (opens in a new tab) Published 20 Aug 2026. Accessed 31 Oct 2026. · Alibaba Group: Alibaba Group announcement on cloud and AI infrastructure investment (opens in a new tab) Published 24 Feb 2025. Accessed 19 Oct 2026.
Confirmed · Filed
Tencent
RMB52.8 billion of capital expenditure in Q2 2026, up 176% year on year
Tencent says large AI-related prepayments for compute supported its Hy model, AI products and external cloud demand. Free cash flow was negative RMB13.8 billion in the quarter.
Source: Tencent: Tencent announces 2026 second quarter results (opens in a new tab) Published 12 Aug 2026. Accessed 31 Oct 2026.
Confirmed · Filed China Mobile
112,700 PFLOPS (FP16) of intelligent computing capacity and RMB61.0 billion of capex in the first half of 2026
The capacity figure includes both self-built and rented computing. China Mobile says utilisation of in-service resources exceeded 90%. China Daily reports its AI data centre contracted scale exceeds 7 GW (reported, not in the filing).
Sources: China Mobile (HKEX filing): Interim Report 2026 (opens in a new tab) Published 27 Aug 2026. Accessed 31 Oct 2026. · China Daily: Nation's smart computing scale surges (opens in a new tab) Published 15 Sept 2026. Accessed 31 Oct 2026.
Confirmed · Filed GDS Holdings
684,977 sqm in service, 79.2% utilised, 170,355 sqm under construction at 30 June 2026
GDS is a carrier-neutral data centre operator. Total area committed and pre-committed rose 18.2% year on year to 784,802 sqm.
Source: GDS Holdings (SEC filing, exhibit 99.1): GDS Holdings Limited reports second quarter 2026 results (opens in a new tab) Published 13 Aug 2026. Accessed 31 Oct 2026.
Quarters and half-years are different periods, so these figures are not added together. Compare with US hyperscaler spending on Investment and Market intelligence.
Policy: where capacity is allowed to go
Live · Official statistic · 18 Feb 2022
In February 2022 China approved eight national computing hubs (Beijing-Tianjin-Hebei, Yangtze River Delta, Guangdong-Hong Kong-Macao Greater Bay Area, Chengdu-Chongqing, Inner Mongolia, Guizhou, Gansu and Ningxia) and ten data centre clusters. The NDRC said eastern regions face land and energy shortages, while western regions are rich in renewable energy.
Source: National Development and Reform Commission: China launches East Data West Computing project (national computing hubs) (opens in a new tab) Published 18 Feb 2022. Accessed 19 Oct 2026.
Projected · Target · 23 Jul 2024
Special action plan for green and low-carbon data centres
Issued by the NDRC, MIIT and two central bureaus. Targets an average data centre PUE below 1.5 by 2025 and a 10% yearly rise in renewable energy use in data centres. By 2030 PUE and energy and carbon efficiency per unit of computing should reach internationally advanced levels. Officials expected data centre power use to rise by about 15% a year.
Source: gov.cn English (Xinhua): China sets green targets for data centers (opens in a new tab) Published 24 Jul 2024. Accessed 31 Oct 2026.
Live · Official statistic · 8 Oct 2026
National computing network: corridors and interconnection nodes
More than 70 computing transmission corridors now link the national hubs and key regions, and 17 regional nodes for national computing interconnection have been approved. A national platform now monitors computing capacity across the country.
Source: Digital China Summit (gov.cn site, from Economic Daily): 全国一体化算力统筹迈出新步伐 (Integrated national computing coordination takes a new step) (opens in a new tab) Published 8 Oct 2026. Accessed 31 Oct 2026.
Dated policy records for every country: Policy tracker.
AI chips and export limits
Confirmed · Company statement Huawei
Ascend roadmap: 950PR (Q1 2026), 950DT (Q4 2026), 960 (Q4 2027), 970 (Q4 2028)
Huawei says it will release a chip a year and double compute each time. The Atlas 950 SuperPoD is planned with up to 8,192 Ascend 950DT chips in 160 cabinets over about 1,000 m², and the Atlas 960 SuperPoD with 15,488 chips. These are company plans, not shipped volumes.
Source: Huawei: Groundbreaking SuperPoD Interconnect: Leading a New Paradigm for AI Infrastructure (opens in a new tab) Published 18 Sept 2025. Accessed 31 Oct 2026.
Confirmed · Company statement Alibaba (T-Head)
Zhenwu M890 AI processor used by more than 650 external customers through Alibaba Cloud
Alibaba's figure, covering more than 20 industries. Alibaba does not publish shipment volumes.
Source: Alibaba Group: Alibaba's Full-Stack AI Accelerates Monetization with 22-Quarter-High Cloud Growth (opens in a new tab) Published 20 Aug 2026. Accessed 31 Oct 2026.
Confirmed · Filed
NVIDIA
US licences allow small H200 shipments to approved Chinese customers, but Chinese restrictions limit sales
NVIDIA says H200 shipments under the US licensing programme were less than 1% of Data Center revenue in the quarter to 26 July 2026, and it took a US$0.4 billion H200 charge in the first half of fiscal 2027. Its outlook for the next quarter assumes no Data Center compute revenue from China.
Source: NVIDIA (SEC Form 10-Q): Quarterly report for the period ended 26 July 2026 (opens in a new tab) Quarter ended 26 July 2026. Accessed 31 Oct 2026.
Chip power and rack density context: Chip power.
Power and cooling
The IEA expects Chinese data centre electricity use to grow by about 175 TWh between 2024 and 2030 (forecast). National policy asks for an average PUE below 1.5 by 2025 and pushes new large sites towards western hubs with more renewable power (see Policy).
Confirmed · Company statement
Offshore wind-powered underwater data centre in full commercial operation off Shanghai (May 2026)
HiCloud's 24 MW facility in the Lingang Special Area seals servers in modules about 35 m below the surface and uses seawater for cooling. Chinese media claim a PUE below 1.15. It is a single specialist project, not a mainstream design.
Source: DatacenterDynamics: HiCloud's offshore wind-powered underwater data center up and running off coast of Shanghai, China (opens in a new tab) Published 20 May 2026. Accessed 31 Oct 2026.
Cooling and power suppliers, vendor-neutral: Suppliers.
Companies are named for information only. CoolPower Index is vendor-neutral and this is not investment advice. Compare countries on Compare markets. Other regions: Regions and Asia Pacific. Industrial robots and humanoids in China and APAC: Robotics.
