At the 2026 National High Performance Computing Conference in Shenzhen on Oct 18, China's first fully domestic 100,000-card AI supercluster, the Sugon 8000, was unveiled. It integrates supercomputing and intelligent computing on one platform, supporting both scientific computing and large-model training and inference, and ten major application achievements are to be released.
Sugon released its upgraded AI inference-native storage FN Neo, which uses a pooled shared architecture for cluster-level KV Cache scheduling. It delivers 160GB/s bandwidth per array with sub-millisecond latency, cuts first-token response time by 73%, boosts token throughput by 150%, offers 99.999% reliability, and supports frameworks like vLLM, LMCache, and Mooncake.
The China Meteorological Administration's Earth System Numerical Prediction Center, together with Sugon, achieved a breakthrough in global kilometer-scale weather forecasting by using the domestically built 100,000-GPU AI supercomputing cluster Sugon 8000 and the self-developed MCV model, completing 5-kilometer-resolution, 10-day global forecasts within one hour.
At the 2026 China Computing Power Conference, Sugon Vice President Cao Zhennan disclosed that the National Supercomputing Internet has surpassed 1.7 million registered users, completed over 270 million jobs, and achieved a domestic computing power share exceeding 95%. The platform has aggregated over 3.5 million CPU cores and 250,000 GPU cards, with daily jobs exceeding 300,000 and peaking above 1 million; its app marketplace has over 880 merchants and 240,000 orders.