China's First Fully Domestic 100,000-Card AI Supercluster Debuts at HPC Conference
At the 2026 National High Performance Computing Conference in Shenzhen on October 18, China's first fully domestically produced 100,000-card AI supercluster made its public debut. The system, named Sugon 8000 and completed in 2026, is expected to release ten major application achievements during the conference.
The 100,000-card cluster consists of 100,000 AI accelerators interconnected at high speed. Its core breakthrough lies in natively integrating supercomputing and intelligent computing on a unified compute platform, supporting both high-precision scientific computation and large-model training and inference.
According to the conference, the computing power of the system is sufficient to create a digital twin brain in a computer or even simulate the evolution of a universe spanning ten billion light-years. Rather than simply pursuing scale, Sugon 8000 adopts a native supercomputing-intelligent computing fusion route, allowing the same platform to handle both physics-based deduction and data-driven inference.
Supercomputing works by encoding physical equations and simulating future states step by step, while intelligent computing summarizes patterns from empirical data, with large language models as a representative example. Relying only on data correlations, however, does not mean AI truly grasps underlying causality, which can lead to AI hallucinations.
As previously reported, the first fully domestic 100,000-card AI supercluster was officially put into operation in August 2026, marking a new stage of 100,000-card-scale deployment in China's computing infrastructure. The cluster's peak computing power per second is equivalent to 200 years of continuous computation by all humanity, and it currently supports more than 300 computing tasks across 26 fields, including new materials, innovative drugs, and artificial intelligence.
