China Activates First Homegrown 100K-Card Supercomputer

China has officially commissioned the Sugon 8000 (codenamed “Dengfeng”), its first fully domestically-produced 100,000-card artificial intelligence supercomputing cluster, in Zhengzhou, Henan Province. The system, built by Dawning Information Industry (Sugon), achieved full load operation within its first week and represents a landmark milestone in the country’s push for technological self-sufficiency amid escalating US-China technology competition.

According to China.com.cn, the cluster has been synchronously connected to the National Supercomputing Internet’s Zhengzhou core node, integrating it into China’s national integrated computing network. The system has already adapted over 300 applications across more than 20 industries, with more than 70 applications achieving stable operation at a 10,000-card parallel scale.

A Full-Stack Domestic Achievement

What distinguishes the Sugon 8000 from previous Chinese supercomputing systems is its full-stack domestic supply chain. Every layer — from chips and computing to storage, networking, cooling, applications, and services — relies on domestically developed technology. The system uses Haiguang (海光) processors as its computational foundation, paired with a proprietary scaleFabric native RDMA high-speed network for 100,000-card interconnect, and ParaStor distributed storage that achieved first place in the IO500 production all-node and 10-node categories in June 2026, as detailed by PConline.

Sugon officials have described scaling from 10,000 to 100,000 cards as a challenge far beyond simply multiplying capacity, noting that system architecture, network interconnection, memory access efficiency, and energy control all become exponentially more complex at this scale.

Supercomputing-AI Fusion

The system adopts what Sugon calls a “supercomputing-AI fusion” technical route, unifying high-precision scientific computing and lower-precision intelligent computing capabilities within a single architecture. As China Daily reported, this allows the cluster to handle workloads ranging from scientific simulations and industrial modeling to large language model training and AI inference — all on the same infrastructure.

The system also employs an immersion phase-change liquid cooling technology developed domestically. All computing equipment is submerged in fluorinated liquid; when operating temperatures rise, the liquid boils and undergoes phase change, transferring heat through a closed-loop recycling system. This supports single-cabinet megawatt-class high-density deployment.

National Supercomputing Internet

The Sugon 8000’s integration with the National Supercomputing Internet is a critical component of China’s broader computing strategy, according to China Daily. The network, officially launched in April 2024, now aggregates over 3.5 million CPU cores and 250,000 GPU cards, serving more than 1.4 million registered users with over 7,300 applications and 1,500 adapted large language models. The platform records over 11.3 million visits per month.

Chinese officials have emphasized that the country’s computing industry must both develop world-class systems with independent technology and build a connected ecosystem through the National Supercomputing Internet, pushing computing power to a broader user base that now ranges from middle school students to professional researchers.

Strategic Implications

The deployment of the Sugon 8000 carries significant weight in the context of US export controls on advanced semiconductor technology and AI chips. As reported by ECNS, every key component of the system relies on domestically developed technology, suggesting China has made considerable progress in reducing its dependence on foreign chip suppliers.

The system’s launch also aligns with China’s 14th Five-Year Plan, which emphasizes breakthroughs in high-end chips, operating systems, and advanced computing. Sugon has announced a strategic partnership with the Beijing Institute of Scientific Intelligence to develop a second domestically produced 100,000-card system, signaling that this is not an isolated achievement but the beginning of a broader deployment phase.

What to Watch

While the Sugon 8000 represents a notable engineering achievement, several questions remain unanswered. International benchmark scores beyond the IO500 storage ranking have not been disclosed, making direct comparison with leading systems such as NVIDIA DGX SuperPOD clusters difficult. The performance characteristics of the Haiguang chips — including manufacturing process, FLOPS, and memory bandwidth — have also not been publicly detailed.

Additionally, whether the system can achieve commercial viability at scale or will primarily serve research and national security applications remains to be seen. The Guanghe Organization, an industrial alliance of which Sugon is a leading member, has aggregated over 6,000 industry partners to address ecosystem fragmentation in China’s domestic computing space — a challenge that will be critical to the long-term success of the country’s computing ambitions.

The Sugon 8000 places China among a very small group of nations — alongside the United States and Japan — capable of deploying computing clusters at the 100,000-accelerator scale. As AI workloads continue to grow in complexity and scale, this capability may prove strategically decisive in the global technology competition.