Our Privacy Statement & Cookie Policy

By continuing to browse our site you agree to our use of cookies, revised Privacy Policy and Terms of Use. You can change your cookie settings through your browser.

I agree

Clusters and fibers: China accelerates AI build-out

CGTN

A liquid-cooled server is on display at the World AI Conference, Shanghai, China, July 20, 2026. /VCG
A liquid-cooled server is on display at the World AI Conference, Shanghai, China, July 20, 2026. /VCG

A liquid-cooled server is on display at the World AI Conference, Shanghai, China, July 20, 2026. /VCG

China is racing to build the computing muscle behind its AI industry, and the pieces are starting to connect. New huge clusters are coming online in central and southern China, a hair-thin fiber is set to quadruple data capacity between key hubs, and more than 60% of the country's AI compute is now visible to a national scheduling platform that entered trial operation recently.

China has switched on its first AI super cluster built from 100,000 domestically produced chips, the National Development and Reform Commission (NDRC) said Sunday, marking a new phase in the country's push to scale up computing power.

The cluster went live at the Zhengzhou node of China's national supercomputing network, in the central province of Henan, where engineers have combined traditional scientific computing with AI-focused hardware to prepare for emerging workloads.

The system can hit peak speeds equivalent to all of humanity calculating continuously for 200 years, said Bu Jingde, chief engineer at the National High-Performance Computer Engineering Technology Research Center. It is already running more than 300 computing tasks across 26 fields, including new materials, drug discovery and AI research.

Cluster dedicated for AI models

A second major build-out is underway in the Greater Bay Area, a southern tech hub and one of the national nodes in China's "East Data, West Computing" infrastructure plan. China Telecom has brought a new 10,000-card cluster online aimed at training the next generation of large AI models.

Liu Zhijun, who leads planning for the carrier's Greater Bay Area data center, said Chinese models are moving from hundreds of billions to trillions of parameters, and that the new system can cut training time for a trillion-parameter model from roughly a year to about six months.

China's overall AI computing capacity grew 177% year on year by the end of June, reaching 2,185 exaflops, according to NDRC figures.

The 'four-in-one' fiber

To make better use of capacity scattered across regions, researchers at Pengcheng Laboratory in south China's Shenzhen are developing a new type of optical fiber that packs four separate channels into a single hair-thin strand. "It is designed to break through the capacity ceiling of regular fiber," said Li Chao, an associate researcher at the lab.

The team plans to deploy the new cable along a 1,000-kilometer route linking Shenzhen to southwest China's Guizhou Province, creating a national corridor for moving large amounts of data with minimal delay.

Nationwide coordination network

More than 60% of the country's computing power is now visible to a new national monitoring and scheduling platform in trial operation, which routes jobs to the most suitable hardware across regions.

China expects the build-out to draw roughly 4 trillion yuan (about $560 billion) in new direct investment during the 15th Five-Year Plan period (2026-2030), mostly from private capital, according to Guo Mingjun, who heads the computing economy division at the NDRC's State Information Center. Funding will flow toward areas such as coordinating data centers with the power grid, and the next generation of large-scale AI clusters, he said.

Search Trends