China Deploys First Domestic 100,000-Card AI Supercluster as National Grid Expands

China has put into operation its first fully domestic 100,000-card AI supercluster at the National Supercomputing Internet's Zhengzhou node, signaling a shift toward ultra-large-scale compute infrastructure built on domestic hardware. Supported by a national intelligent computing capacity that surged 177% year-on-year to 2,185 EFLOPS by the end of June, the initiative is paired with a trial monitoring and scheduling platform that already tracks over 60% of the nation's compute resources.

The deployment demonstrates China's attempt to build a resilient, domestic compute architecture capable of supporting trillion-parameter frontier AI models despite external hardware restrictions. By integrating scientific and intelligent workloads into a national framework, Chinese infrastructure planners aim to mitigate compute shortages and resource imbalances across regions.

China has officially activated its first fully domestic 100,000-card artificial intelligence supercluster, according to state media reports citing the National Development and Reform Commission. Located at the Zhengzhou core node of the National Supercomputing Internet, the facility marks China's entry into a new tier of ultra-large compute infrastructure deployment. The Zhengzhou cluster integrates scientific supercomputing with intelligent AI workloads, delivering peak calculation capabilities currently supporting over 300 tasks across 26 sectors, including new materials, innovative pharmaceuticals, and general AI research. In parallel, a new domestic 10,000-card hyper-node cluster was brought online in the Guangdong-Hong Kong-Macao Greater Bay Area hub node. Officials from China Telecom noted that the hub aims to halve the training time required for trillion-parameter large language models from one year to approximately six months. Data reported by state media shows that China's overall intelligent computing capacity grew 177% year-on-year by the end of June, reaching 2,185 EFLOPS. To address geographical disparities between computing supply in the west and demand in the east, Beijing has accelerated the roll-out of a national integrated computing network. Central to this networking push is a newly launched trial monitoring and scheduling platform that already encompasses over 60% of the country's total computing capacity. The platform matches compute supply with demand across hundreds of applications in manufacturing, scientific research, and artificial intelligence, supported by millisecond-level computing service circles connecting major eastern and western hubs. To overcome physical bandwidth constraints across long distances, researchers at institutions such as the Peng Cheng Laboratory are developing multi-core fiber technologies. Recent breakthroughs include four-channel optical fibers—roughly the thickness of a human hair—capable of quadrupling transmission capacity to create low-latency, direct optical transit corridors for nationwide computing scheduling.