China AI data centers: $295B plan, 80% local chips

China's AI data center plan: $295 billion over five years

Bloomberg broke the news, citing people familiar with the matter who asked not to be named. According to them, the plan is still in early discussions, and the figure had not been made public before.

Shares of Chinese data center companies jumped on the report. GDS Holdings rose as much as 12% in US pre-market trading, and Vnet Group gained 17%.

The NDRC plans one network linking China's data centers

China's state planning agency, the National Development and Reform Commission, is one of the bodies drafting the blueprint. The aim is one network of linked computing hubs instead of scattered regional sites that each serve only their own area.

Pooling those regional resources would open high-performance computing to more companies, Charlie Dai, a principal analyst at Forrester Research, told Bloomberg. He expects firms in finance, manufacturing, health care and logistics to get cheaper and more flexible AI capacity.

He also thinks inland provinces could attract more digital investment and talent. Beijing wants the network to carry AI into public services such as health care, transport and city management as well.

China Mobile and China Telecom would run most of the hubs

The two state-owned carriers would run most of the new data centers and keep them linked to one another. Neither replied to Bloomberg's requests for comment.

The telecoms trade publication Capacity sees this as a new role for both carriers. Beyond selling connectivity, they would supply computing power for the whole country.

A 2028 target builds on China's five-year plan through 2030

A national computing network is not a new idea. China's latest five-year plan, running through 2030, already put data infrastructure first. What the new blueprint adds is a deadline: the scattered sites are to form one network by 2028.

The data center push also belongs to the "Six Networks" program Beijing announced earlier in 2026, which covers basic infrastructure from water and electricity to computing. Bloomberg's report says it is unclear how private hubs run by companies such as Alibaba and Tencent would fit into the state network.

An 80% local-chip target would squeeze out Nvidia and AMD

Under the blueprint, at least 80% of the technology in the new data centers, AI chips included, would have to come from Chinese suppliers. Huawei is the only company the reports name, and Nvidia and AMD would be left to compete for no more than the remaining 20%.

The 80% sets a minimum, not a fixed share. Neither report says how the rest would be divided. Capacity notes that the rule pushes China away from Western chips from the buying side, just as US export controls have from the selling side.

Huawei and other homegrown AI chipmakers stand to gain

Huawei is not China's only option. In May 2026, nine kinds of domestic AI chips passed a security review by a Chinese technology-security agency. Their makers include Huawei, Alibaba, Shanghai Biren Technology and Moore Threads.

Clearing that review opens the chips to wider use in sectors with strict security rules. A state network required to buy at least 80% local technology would draw on this group of suppliers.

Nvidia H200 sales to China: approved, not shipped by June

Washington has approved Nvidia's sale of the H200 to Chinese customers, a real easing of controls meant to slow China's AI progress. The H200 is about one generation older than Nvidia's Blackwell chips.

Shipments had not begun by the time of the June report, and Bloomberg read the delay as a sign of Beijing's growing confidence in replacing some computing capacity with Chinese hardware. Even if the H200 ships, the 80% floor would limit foreign chips in the state network to whatever share is left.

Government debt would fund most of China's AI investment

The state, not companies, would put up most of the $295 billion, with bank loans and private capital only filling gaps. That stands out because other public spending in China is being cut as government debt rises.

Bloomberg sees the plan as a sign of how determined Beijing is to keep paying for advanced technology regardless. The order of funding sources below is as Bloomberg's sources described it.

Ultra-long bonds and state funds first, then bank loans

The biggest source would be sovereign debt (bonds issued by China's central government), including ultra-long-term special government bonds that usually run more than 10 years. The second would be state funds that invest in strategic industries.

1. Main source: sovereign debt, including ultra-long-term special government bonds
2. Main source: state funds for investment in strategic industries
3. Supplement: bank loans
4. Supplement: private capital

Neither the Ministry of Finance, which issues the bonds, nor the NDRC answered Bloomberg's faxed requests for comment. The plan's financing details therefore rest on the unnamed sources alone.

Grid spending could bring the total to $735 billion or more

The $295 billion pays for the AI facilities themselves: data centers and faster communications links. China also plans to tie the power grid into the project, and the sources told Bloomberg that including that work could take total projected investment to at least $735 billion.

China vs. US AI investment: five years against one year

US companies such as Meta and Microsoft have budgeted $725 billion for AI in 2026 alone, while China's $295 billion is spread over five years. Set side by side, the two totals look closer than they are, since one covers a single year and the other five.

US firms' 2026 budget is about 12 times China's yearly pace

Neither report gives a year-by-year schedule. Spread evenly, China's plan would come to about $59 billion a year.

That yearly figure is simply $295 billion divided by five, not a number from the reports. Measured that way, the $725 billion US companies plan for 2026 is roughly 12 times China's annual pace.

Private spending is excluded, and China builds for less

Two factors suggest the real gap is smaller than 12 times. The $295 billion counts only the state plan, so spending by private firms such as Alibaba and Tencent is not in it.

Data centers also tend to cost less to build in China than in the US. Labor, parts and construction are cheaper, and local governments offer incentives.

Neither report quantifies either factor, so the 12-times figure compares state money in China with company money in the US without any cost adjustment. The blueprint is also still in early discussions, and the $295 billion could change before anything is announced.