Sugon Unveils scaleX 10,000-GPU Supercluster, Marking a Breakthrough in China’s AI Computing Power
Want to read in a language you're more familiar with?
China has taken a major step forward in AI infrastructure with the first real-world debut of a domestically built 10,000-GPU supercluster.
On December 18, 2025, at the Photonics Organization 2025 Artificial Intelligence Innovation Conference (HAIC 2025) held in Kunshan, Jiangsu Province, Sugon (Dawning Information Industry) officially unveiled and showcased its scaleX 10,000-GPU supercluster, marking the first time a China-made AI cluster of this scale has been presented as a fully operational system.
Designed for trillion-parameter foundation models, scientific AI, and other highly complex workloads, the scaleX supercluster represents a next-generation computing infrastructure solution. Sugon reported breakthroughs across multiple core technologies, with some capabilities already exceeding the performance targets set by overseas competitors for 2027.
The system’s key strengths include:
- World-first 640-GPU ultra-node per cabinet, with 16 ultra-nodes interconnected to deploy 10,240 AI accelerator cards, delivering over 5 EFLOPS of total computing power and achieving a PUE as low as 1.04;
- A self-developed 400G native RDMA high-speed network, offering 400 Gb/s bandwidth, sub-1 microsecond latency, and 30% lower costs;
- Deep optimization across storage, computing, and data transmission, improving AI accelerator utilization by 55%;
- An integrated digital twin and intelligent scheduling system, delivering 99.99% availability.
The scaleX 10,000-GPU supercluster supports multiple accelerator brands and mainstream computing ecosystems. It has already been adapted to more than 400 leading large models and world models, and can be widely applied to scenarios such as foundation model training, financial risk control, and geological and energy exploration—providing powerful computing support for the continued growth of China’s AI industry.
