- China’s AI computing scale: As of end-June 2025, China operates 10.85 million standard racks, with intelligent computing power reaching 788 EFLOPS (exa–floating point operations per second) and storage exceeding 1,680 exabytes, according to the Ministry of Industry and Information Technology (MIIT). People’s Daily Online
- Explosive growth: Intelligent computing power jumped from 90 EFLOPS in 2024 to 788 EFLOPS by mid‑2025, and is forecast to grow over 40% in 2025 alone, driven by AI demand. english.henan.gov.cn
- Model ecosystem: China has released 1,509 large AI models, roughly 40% of the world’s ~3,755 models, giving it the largest national model count globally. Reuters
- Industrial integration: Intelligent computing is now used in finance, healthcare, energy, manufacturing, transport, agriculture and smart cities, with over 23,000 innovative computing projects piloted across sectors. english.henan.gov.cn
- Recent policy shock: On 5 November 2025, Beijing ordered state‑funded data centres to use only domestic AI chips, forcing many projects to remove or cancel foreign chips such as Nvidia’s, in one of China’s toughest “de‑Americanization” moves so far. Reuters
- Domestic hardware push: China Unicom is building a $390 million data centre in Qinghai powered by AI chips from Alibaba and other Chinese vendors, targeting 20,000 PFLOPS of capacity when complete. Reuters
- Open‑source momentum: Chinese startup Zhipu recently released its open‑source GLM‑4.5 model; by July 2025, China’s 1,509 LLMs led all countries, according to Xinhua via Reuters. Reuters
- Energy and infrastructure challenge: Computing infrastructure in China is projected to consume around 360 billion kWh by 2025, with 5G base stations adding another 140 billion kWh, raising pressure for greener, more efficient data centres. chinaobservers
- Market angle (approximate latest prices, USD): Baidu around $116.6 (BIDU), Alibaba $159.8 (BABA), and Nvidia $187.7 (NVDA) in U.S. trading on 17 November 2025 – all deeply exposed, in different ways, to the AI compute race.
BEIJING, Nov. 17, 2025 — An AI “power grid” behind the scenes
China’s AI boom isn’t just about flashy chatbots or humanoid robots; it’s being driven by a vast, largely invisible infrastructure of intelligent computing centres, dense racks of servers and accelerators that together form something like an AI “power grid.”
A new report from People’s Daily highlights that by the end of June 2025, China’s data centres housed 10.85 million standard racks, with 788 EFLOPS of intelligent computing power and 1,680+ exabytes of storage, putting the country among the global leaders in compute capacity. People’s Daily Online
At the same time, official data and international reporting indicate that China has released 1,509 large AI models, more than any other single country and over 40% of all publicly known models worldwide. Reuters
The combination of massive compute and dense model ecosystems is now central to Beijing’s push to turn AI into a “foundational engine” for economic upgrading, scientific discovery and national competitiveness.
What is “intelligent computing” – and why does EFLOPS matter?
“Intelligent computing” (智算) in Chinese policy documents means compute infrastructure tuned specifically for AI workloads:
- GPU and AI accelerator clusters
- High‑bandwidth networking
- High‑speed storage
- Optimized software stacks for large‑scale training and inference
Unlike traditional “general computing,” which might focus on standard CPU servers, intelligent computing is measured using AI‑friendly metrics such as EFLOPS at FP16 or lower precision, which better reflect performance for deep learning. english.henan.gov.cn
MIIT data show that China’s intelligent compute capacity jumped from 90 EFLOPS in 2024 to 788 EFLOPS by mid‑2025, an almost order‑of‑magnitude leap as AI models ballooned in size and complexity. english.henan.gov.cn
Zhang Xiaorong, director of the Beijing‑based Cutting‑Edge Technology Research Institute, describes computing power as a core battlefield in the global tech race and argues that the surge of capacity means AI is no longer confined to research labs but spreading through society. english.henan.gov.cn
1,509 large models: From leaderboard bragging rights to real‑world deployment
At the World Artificial Intelligence Conference (WAIC) 2025 in Shanghai, officials revealed that China accounts for 1,509 of the world’s roughly 3,755 large AI models, according to Xinhua – the highest model count of any country. Reuters
Reuters has corroborated this figure, noting that Zhipu’s new GLM‑4.5 agent‑oriented model is just one entry in a rapidly growing roster of Chinese LLMs. Reuters
Beyond raw numbers, these models are increasingly open‑sourced. Analyses of Chinese open‑source LLMs show that companies like DeepSeek, Baichuan, Moonshot, Alibaba (Qwen), and Baidu have released a wave of models under Apache/MIT‑style licences, supporting tens of thousands of derivative projects. IntuitionLabs
This model explosion is tightly bound to China’s intelligent computing build‑out:
- Pengcheng Laboratory links computing nodes in over 20 cities via the China Computing Net, with its “Cloud Brain II” platform serving tens of thousands of researchers. People’s Daily Online
- The Shanghai AI Lab, together with local universities and research labs, recently used intelligent computing platforms to identify and validate two new cancer targets within just two months, demonstrating how big compute shortens the loop from hypothesis to discovery. People’s Daily Online
In practice, intelligent computing is allowing Chinese researchers to push into frontier areas like multi‑modal models, embodied AI (robots and drones), and advanced scientific simulations.
Policy pivot: Chip bans, domestic hardware and quantum wildcards
One of the biggest recent shifts came on 5 November 2025, when Reuters reported that China had ordered new, state‑funded data centres to use only domestically made AI chips. Projects under 30% completion must rip out foreign chips or abandon plans to buy them; more advanced projects will be reviewed case by case. Reuters
The move is explicitly aimed at:
- Reducing dependence on U.S. chipmakers like Nvidia, AMD and Intel
- Accelerating the scale‑up of domestic suppliers such as Huawei, Cambricon, MetaX and Moore Threads
According to that report, the guidance could represent one of Beijing’s most aggressive steps yet to purge foreign technology from critical compute infrastructure. Reuters
In parallel, China is spotlighting large‑scale facilities powered by local chips. In September, state media highlighted a $390 million China Unicom data centre in Xining, Qinghai, where about 72% of nearly 23,000 chips are from Alibaba’s T‑Head unit, alongside MetaX, Biren and others. When finished, it aims for 20,000 PFLOPS of compute. Reuters
Commercial players are also stepping up:
- Baidu has unveiled two next‑generation AI chips, M100 and M300, via its Kunlunxin unit, plus new “Tianchi” supernode systems, positioning itself as both AI platform and chip supplier after U.S. export controls squeezed Nvidia’s Chinese business. The Wall Street Journal
- A Chinese firm called CHIPX claims to have built an optical quantum chip allegedly up to 1,000× faster than Nvidia GPUs on some AI workloads, though production is limited to about 12,000 wafers per year and yields remain low – making this more a long‑term wild card than an immediate alternative. Tom’s Hardware
Expert view
Zhang Xiaorong argues that chips, computing power and AI form a “reinforcing loop”: better chips allow more powerful AI, which drives demand for more compute, which in turn incentivizes further chip advances. english.henan.gov.cn
In the near term, however, independent analysts caution that China still lags the United States in access to cutting‑edge AI hardware. A RAND study estimated that as of 2024, the U.S. controlled about 75% of global AI compute, versus roughly 15% for China, although China’s share is rising quickly as it upgrades older, less‑restricted manufacturing nodes and builds enormous clusters of “good‑enough” accelerators. RAND Corporation
“East Data, West Computing”: A national AI infrastructure
China’s AI strategy isn’t just about single data centres; it’s about linking them into a national compute fabric.
Under the “Eastern Data, Western Computing” (东数西算) initiative, China is building eight national computing hubs and dozens of smaller clusters, channelling data‑hungry applications from coastal megacities to energy‑rich western regions like Guizhou and Qinghai. RAND Corporation
Key pieces of this puzzle:
- Guizhou has become one of the strongest “home‑grown” intelligent computing hubs, with around 85 EFLOPS of capacity, 98% of it in AI‑oriented compute, and roughly 90% of hardware domestically produced. bjreview.com
- A DCPulse analysis notes that Guizhou’s data centres reached about 92.6 EFLOPS by September 2025, with roughly 97% dedicated to intelligent computing, positioning the province as a flagship AI compute base. dcpulse.com
- China Mobile reports its own intelligent computing centres already provide 11 EFLOPS across key national hubs, supported by an AI‑centric optical backbone network with millisecond‑level latency. huawei+1
At the 2025 China Computing Power Conference, officials said that the national computing platform has already integrated 10 provincial platforms, with operators’ compute investments expected to grow 20%+ annually, and leading internet firms planning over 500 billion yuan (≈$70 billion) in AI investment over three years. english.henan.gov.cn
At the same time, Beijing is dealing with overcapacity and under‑utilization. After a three‑year, government‑backed building boom, thousands of data centres operate at 20–30% utilisation. Reuters reports that policymakers now plan a state‑run national cloud network to sell surplus computing power and rationalise the sector, with MIIT coordinating China’s three big telecom operators to interconnect facilities nationwide by around 2028. Reuters
From cows to cancer: How intelligent computing is hitting the real economy
- Healthcare & life sciences
- At Nankai University, a high‑efficiency open‑source AI framework on intelligent computing infrastructure boosted the speed of retinal blood‑vessel segmentation by 2.4×, bringing AI‑powered fundus imaging closer to routine clinical use. People’s Daily Online
- The Shanghai AI Lab, LinGang Lab and leading universities in Shanghai used large‑scale compute to identify and validate two new cancer drug targets in only two months, a task that traditionally can take much longer. People’s Daily Online
- Smart manufacturing & transport
- CRRC, China’s top high‑speed train maker, trained an intelligent aerodynamic simulation model using Baidu’s PaddlePaddle and scientific computing suite. The model reportedly cut simulation time from several days on a supercomputer to about 10 seconds on a single GPU, improving efficiency more than 30× while keeping accuracy within 5%. People’s Daily Online
- Agriculture & food
- At Yili Group’s Modern Intelligent Health Valley dairy complex, each cow has a digital health record. Using intelligent computing from Alibaba Cloud, AI models analyze video of animals’ eyes and behaviour to support real‑time health monitoring and precision feeding, which company representatives say has improved milk quality and cut waste. People’s Daily Online
- Supply chains & enterprise AI
- Yili and Alibaba Cloud report building over 800 intelligent agents, with large models now covering about 70% of supply‑chain scenarios, from inventory turnover to logistics optimization. People’s Daily Online
And beyond that flagship article:
- Autonomous driving firms like Haomo.AI and XPeng jointly operate intelligent computing centres with over hundreds of PFlops of power to train self‑driving algorithms, often hosted on Alibaba Cloud or other hyperscalers. People’s Daily Online
- China Southern Power Grid built a “MegaWatt” sector‑specific model on Huawei’s Ascend platform to detect faults in power lines, combining computer vision with natural language processing to triage issues faster and more accurately. Huawei Enterprise
Hong Yuan, chief product officer at intelligent‑computing vendor Tecorigin, summarizes the industrial strategy this way: innovation in intelligent computing depends on close collaboration between enterprises, universities and research institutes, backed by strong talent pipelines and customized hardware‑software stacks. People’s Daily Online
Universities, labs and the “full‑stack” AI industrial policy
China’s intelligent computing build‑out sits on top of a broader AI industrial policy stack:
- A National Integrated Computing Network to link cloud resources nationwide
- Dozens of state‑backed AI labs, such as Zhejiang Lab, Pengcheng Lab, the Beijing Academy of Artificial Intelligence and the Beijing Institute for General Artificial Intelligence, which mix basic research, talent training and standards work RAND Corporation
- Large AI‑focused investment funds, including a National AI Industry Investment Fund (~$8.2 billion) and multi‑billion‑dollar loan programmes from major state banks for AI‑related projects RAND Corporation
A RAND analysis notes that these labs and funding channels blur the line between public and private sectors, with state labs incubating talent and techniques that later flow into commercial champions like Baidu, Huawei, Alibaba, DeepSeek and Zhipu. RAND Corporation
Meanwhile, China’s Politburo has repeatedly framed AI as a “strategic technology”, and think‑tank estimates suggest the country has already reached around 70% of its 2030 AI industry target by mid‑2025, underscoring how central intelligent computing has become to national planning. apcoworldwide.com
The energy and sustainability squeeze
All this compute comes with a cost: electricity.
- China’s computing infrastructure will consume around 360 billion kWh, and
- 5G networks will add roughly 140 billion kWh on top.
That’s pushing policymakers to:
- Move data centres to regions with abundant hydropower, like Guizhou and Qinghai
- Invest in liquid‑cooled racks and high‑efficiency power systems (Tecorigin, for example, touts high‑density liquid‑cooled clusters serving more than 200 universities and enterprises) People’s Daily Online
- Tighten approvals for new data centres, favouring those with green energy and high utilization
The government’s review of a data‑centre glut and plan to sell surplus compute through a state‑run network is partly about economics, but also about avoiding massive stranded, energy‑hungry assets. Reuters
Market angle: What investors are pricing in (and what they aren’t)
While China’s domestic AI infrastructure is largely financed through state‑owned operators and private Chinese capital, global investors are watching through a handful of key stocks and themes.
Prices below are approximate snapshots from U.S. markets on 17 November 2025 and will change over time.
- Baidu (BIDU, NASDAQ) – about $116.6 per share
- Acts as a proxy for Chinese AI platforms and domestic AI chips via its Kunlunxin unit and ERNIE model family.
- The launch of M‑series chips and Tianchi supernode systems positions Baidu as both heavy compute user and provider. The Wall Street Journal
- Alibaba Group (BABA, NYSE) – around $159.8 per share
- Through Alibaba Cloud and T‑Head, Alibaba supplies both cloud‑based intelligent computing and AI accelerators, including those used in China Unicom’s Xining data centre and in industrial deployments such as Yili’s dairy operations. Reuters
- Nvidia (NVDA, NASDAQ) – roughly $187.7 per share
- Once dominant in China’s AI chip market, Nvidia now faces severe restrictions. Reuters reports that its share of the Chinese AI chip market has effectively fallen to zero in key regulated segments, especially after the new domestic‑only chip guidance for state‑funded centres. Reuters
Other likely beneficiaries—Huawei, Cambricon, MetaX, Biren, Moore Threads—are often either unlisted, thinly traded, or available only on mainland exchanges, making them less accessible to global investors but central to China’s domestic compute ecosystem. Reuters
Important: None of this is investment advice. It’s simply a snapshot of how public markets intersect with China’s AI infrastructure push.
Outlook: How far can China’s intelligent computing surge go?
Analysts and officials see 2025 as a turning point:
- Growth trajectory: MIIT‑linked forecasts suggest intelligent computing power will grow more than 40% year‑on‑year in 2025, after already leaping nearly 8‑fold from late 2024 to mid‑2025. english.henan.gov.cn
- Industrial spread: Use‑cases are moving from pilots to wide deployment in finance, healthcare, power grids, logistics, education and agriculture, with tens of thousands of projects trialed nationwide. english.henan.gov.cn
- Geopolitical control: The domestic‑chip mandate for state‑funded data centres suggests Beijing is willing to accept short‑term hardware performance gaps in exchange for long‑term supply security and technological sovereignty. Reuters
If double‑digit annual growth in intelligent computing continues, China’s AI compute capacity could multiply several times again by the end of the decade. That said, a few constraints loom large:
- Hardware limits and export controls
- Even with domestic chips, China still lacks access to the most advanced lithography tools and cutting‑edge GPU designs, making it harder to match U.S. and Taiwanese performance on a per‑chip basis. RAND Corporation
- Energy and cooling
- Power consumption and heat density will force ever more aggressive liquid‑cooling, waste‑heat recovery and green‑energy strategies, especially as compute densifies in western provinces. chinaobservers
- Utilization and ROI
- The very need to build a network to sell surplus compute shows the risk of over‑building. The challenge now is not just to build racks, but to keep them busy with economically valuable workloads. Reuters
- Global fragmentation
- With both China and the U.S. racing to build largely separate AI infrastructure stacks, the world may see two partially incompatible AI ecosystems, complicating cross‑border collaboration, standards, and governance.
Still, the direction of travel is clear. As Wang Peng of the Beijing Academy of Social Sciences notes, intelligent computing is becoming a core driver of high‑quality development across sectors from healthcare to agriculture and education. english.henan.gov.cn
For the global AI landscape, China’s 788‑EFLOP intelligent computing “super‑brain” is no longer just a domestic story—it’s a central factor in how quickly AI advances, how it is governed, and how the balance of technological power evolves over the next decade.
