_________

Published on Oct. 07, 2026

China treats artificial intelligence as just another piece of infrastructure. Computing power for AI, in Beijing’s eyes, features alongside the electrical grid and water supply as a basic service that must exist before businesses can use it and grow.

Such a utility requires “patient money” as investment and must be constructed ahead of demand. That is why the People’s Bank of China (PBOC) on September 29, 2026, lowered the rate charged on Pledged Supplementary Lending (PSL), the facility that provides inexpensive funding to China’s state-owned policy banks. The central bank announced the step on its website (in Chinese).

To “give the policy banks a stronger incentive to support the real economy and serve national strategies,” the one-year rate was reduced to 1.5 percent from 1.75 percent, the PBOC wrote.

Building more computing power is explicitly mentioned as an objective. “Construction of the ‘six networks’ is included in the scope of PSL. These are the water network, the new-type power grid, the computing network, the next-generation communication network, urban underground pipeline networks and the logistics network”, the statement says.

This means that data centers and connections between them can now be financed with exactly the same cheap state credit as antennas for mobile phones or power lines.

The central bank’s move was a political decision. One day earlier, the executive meeting of the State Council chaired by Premier Li Qiang stated that “the major projects set out in the 15th Five-Year Plan and in the implementation plan for the ‘six networks’ must start construction as early as possible,” the official news agency Xinhua reported (in Chinese).

“A power grid transmits electricity and a water network transmits water. What the computing network actually dispatches and moves are computing tasks, the data they need and the results,” wrote Xiao Hongwei, director of the policy simulation laboratory at the State Information Center under the National Development and Reform Commission (NDRC), in a commentary for Xinhua (in Chinese).

The term “network” is not randomly chosen. The plan is to wire China’s computing facilities into one system, so that a user anywhere in the country can locate spare capacity and dispatch work to it.

The underlying logic is that the Chinese leadership measures “winning” in AI differently from U.S. President Donald Trump, who believes that building the strongest frontier model might hand America a silver bullet to make it suddenly great again, even with its crumbling bridges and insufficient power supply.

Powerful AI models matter to China too, but so does the number of hospitals, ports and factories using AI. The new technology is seen more like water that can lift all boats, rather than approached with an “oil mindset” where powerful Silicon Valley barons harvest a scarce resource sold to the masses at premium prices.

The task is to “advance the upgrading of traditional industries, the growth of emerging industries and the early planning of future industries, and let AI enable all industries,” President Xi Jinping said at the World Artificial Intelligence Conference in Shanghai in July. Our Deep Dive “AI as Water in China, AI as Oil in the U.S.” already informed readers about that speech.

Measuring success by how many users can use AI cheaply means putting capacity in place before they even turn up.

“First, build moderately ahead of demand, combine construction and use, and plan and build computing infrastructure scientifically,” was the prescription given at the China Development Forum in March by Liu Liehong, head of the National Data Administration (in Chinese). Utilities operate like that. A city has to lay a water main before people can move to a new development area.

Along the way toward their goal, Beijing and local governments alike are coping with temporary overcapacity. Only about 36.8 percent of the AI computing power provided in China in 2025 was used. More than 60 percent of the capacity sat idle, statistics from the China Academy of Information and Communications Technology show (in Chinese).

Chinese state media criticize this overcapacity, but call it a “structural imbalance.” Most data centers that are empty were developed by local governments using their financing vehicles and state-owned platforms “with the mindset of property development,” according to the financial daily Zhengquan Shibao.

In Shanghai and Inner Mongolia, high-end computing cluster operators report no spare capacity. The clusters that China’s large internet companies built for themselves are likewise fully occupied. The idle capacity in some locations “does not come from a surplus in the industry as a whole. It is a structural imbalance,” the newspaper concludes.

Seen from abroad, such overcapacity in China is little understood and judged by aggregate figures only. The market has “settled on two lazy assumptions: China is big, and China is empty,” writes the U.S. research firm SemiAnalysis, which has mapped more than 1,000 data center facilities in the country.

By its count, China’s fleet stands at more than 24 gigawatts, compared with 56 gigawatts expected in the U.S. by the end of 2026.

Demand for AI computing power is growing very fast in China, however. The daily volume of tokens, the units in which the work of AI models is billed, rose in China from 100 billion at the start of 2024 to more than 140 trillion in March of this year, according to Liu Liehong, the head of China's Guojia Shujuju (National Data Administration). That was a a 1,400-fold increase in just over two years.

And while current overcapacity may sound wasteful, the whole approach has advantages. Treating computing capacity as infrastructure allows its planning to be coordinated with the electricity grid. Beijing has decreed that for new computing facilities at the national hubs, at least 80 percent of the power they consume has to come from renewable energy.

A computing campus in Inner Mongolia illustrates this approach in our previous Deep Dive, with 2 gigawatts of planned capacity and an exceptionally high proportion of green electricity supplied directly. Bringing the two networks into alignment nevertheless faces bottlenecks in planning, construction and pricing, as Xiao Hongwei acknowledges.

A lot of money is flowing into this new infrastructure buildout. This year’s investment in all six networks and associated areas, including transport and the “AI Plus” program, is estimated by the NDRC at more than 7 trillion yuan, about US$1 trillion. The computing network alone is to add 4 trillion yuan, roughly US$600 billion of direct investment between 2026 and 2030, NDRC-Chef Zheng Shanjie was quoted by the finanical news agency Cailianshe in July (in Chinese).

Paying the entire bill out of state coffers is not the state’s intention. Enterprise funding accounts for most computing facility construction, according to Jiang Yi.

In his Xinhua commentary, Xiao Hongwei points to the NDRC’s explicit statement that “in the actual construction, market forces will play the decisive role.” Whatever public money also flows, however, is seen in Beijing as a stimulus for the sluggish economy. The central bank’s package that includes AI infrastructure was described by Reuters as “a fresh bid to shore up its slowing economy.”

The new computing network should generate jobs “on the order of a million” while also redirecting computing demand from the east of the country to the resource-rich west, Xiao writes.

Seen from a higher perspective, betting on cheap and ubiquitous AI is part of China’s strategy to foster new growth drivers for its economy while it wrestles with demographic and geopolitical challenges.

If Beijing can really make AI as readily accessible as water and electricity, manufacturers in China stand to benefit a lot. That includes foreign manufacturers operating plants in China.

For everybody else, it will mean a huge new challenge. When competitors in China can get cheap intelligence from a network of data centers just as they take electricity from the grid, they will be able to win on the world market with even lower prices than now.