en.Wedoany.com Reported - On July 18, during the 2026 World Artificial Intelligence Conference (WAIC), the Sugon 8000, China's first domestically produced 100,000-card AI supercluster and a "treasure of the venue" awardee, released its latest operational data: it experienced overwhelming demand and full-capacity operation in its first week online, processing over 150,000 tasks daily, with a peak of more than 500,000 tasks in a single day.
To date, the Sugon 8000 has completed nearly a thousand super-intelligent fusion application adaptations, covering large-scale model training, high-throughput inference, scientific computing, and other multi-ten-thousand-card-scale scenarios. Its rapid saturation upon launch reflects the accelerating release of real-world demand from research institutions, enterprises, and developers for ultra-large-scale, high-quality computing power, and indicates that 100,000-card-level computing power has begun to continuously undertake broader scientific research and industrial tasks.
Full-capacity operation does not imply resource congestion. Leveraging dynamic resource management mechanisms such as intelligent scheduling engines, data affinity algorithms, and multi-fusion scheduling strategies, the Sugon 8000 intelligently distributes and allocates tasks based on task type, data location, and resource status. Facing millions of concurrent user requests, the system maintains smooth operation under high load, achieving "no congestion under high load, no queuing for multiple tasks," and continuously supports major strategic tasks such as the "Four Orientations" and "Four Poles," as well as complex applications like large-scale model training and high-throughput inference.
Stable operation at the 100,000-card scale tests not single performance metrics but the overall system coordination capability. The Sugon 8000 adopts a globally pioneering high-density cabinet structure, with computing density per unit 20 times higher than other super nodes, enabling higher computing deployment within limited space. Its self-developed scaleFabric, a native IB-like RDMA high-speed interconnect technology, achieves stable interconnection at the 100,000-card scale. Meanwhile, the ParaStor distributed storage, ranked first in both IO500 production categories, and globally leading immersion phase-change liquid cooling, along with other key technologies, work in synergy to ensure efficient data flow and long-term stable system operation, supporting sustained effective computing output under high load.
As a core node of the National Supercomputing Internet, the Sugon 8000 is continuously lowering the barrier to high-end computing through initiatives such as the "100,000-Card Co-Creator Incentive Plan," the "Intelligent Agent Ecosystem Co-Creation and Recruitment Plan," and the "Core Ecosystem Partner Recruitment Plan." Enterprises, research teams, and developers can easily access and on-demand utilize high-quality computing resources without building their own ultra-large-scale clusters, transforming national-level high-end computing from major infrastructure into a more accessible and usable public service.
The full-capacity operation in the first week, adaptation of nearly a thousand applications, and stable performance under high load indicate that the evaluation focus of ultra-large-scale AI infrastructure is shifting from system scale to actual service capability. The latest operational data from the Sugon 8000 also provides a new practical sample for observing how a domestically produced 100,000-card AI supercluster continuously serves real-world scientific research and industrial tasks.
