China's first 100,000-card AI supercluster, “Sugon 8000,” has arrived

On July 10, during the Photosynthesis Organization 2026 Intelligent Computing Application Conference, Sugon announced that China's first fully China-developed 100,000-card AI supercluster —Sugon 8000 (Dengfeng)was officially completed and simultaneously connected to the National Supercomputing Internet. This marks that AI infrastructure construction is beginning to move from the 10,000-card level to 100,000-card-scale deployment.

▍True 100,000-card deployment, a new standard for AI infrastructure

As large models, scientific intelligence, and agent applications accelerate, computing-power infrastructure is evolving from thousand-card and 10,000-card clusters toward still larger scale. Facing high-concurrency, high-throughput, mixed-precision, multi-task composite loads, a 100,000-card AI supercluster is no longer only a numerical concept, but an important entry capability for next-generation AI infrastructure.

Compared with 10,000-card systems, 100,000-card deployment tests more than accelerator-card count and theoretical peak performance. It also tests system architecture, network interconnect, memory-access efficiency, energy-efficiency control, and a strong ecosystem of applications. The official completion of Sugon 8000 (Dengfeng) offers a reference sample for 100,000-card AI infrastructure and shifts evaluation of large computing-power centers from scale stacking and performance contests toward more comprehensive system capability, with greater weight on real application efficiency and industrial economic productivity.

▍Native HPC-AI fusion, full-chain independent R&D

Sugon 8000 (Dengfeng)adopts “HPC-AI fusion” technology path, discarding traditional partitioning to truly achieve all-type computing native integrated fusion, supporting the composite needs of high-precision scientific computing and low-precision intelligent computing, and supporting full precision from FP64 to INT8, covering scientific computing, large-model training, AI inference, industrial simulation, and other research and industrial scenarios.

In system construction, Sugon 8000 (Dengfeng)featuring full-chain, fully in-house AI infrastructure capabilities spanning “chips, computing, storage, networking, cooling, applications, and services.” Among them, Hygon and other China-developed chips provide the underlying support; scaleFabric IB-like native RDMAThe high-speed network enables 100,000-card cluster high-reliability connectivity；ParaStorDistributed storage supports massive data reads and writes in large-model training and scientific computing, and in the 2026 global IO500 list took first place on both the production full-node and 10-node performance rankings; globally leading immersion phase-change liquid cooling can support single-cabinet MW-class high power-density deployment, and improves cluster energy efficiency through China-developed coolants and year-round free cooling. At the same time, relying on an open computing ecosystem and computing-power service capabilities, it can offer various research institutions, industry users, and application developers from underlying computing power through application adaptation with fully supports .

▍Connecting to the national integrated computing network and turning computing power into productivity

Sugon 8000 (Dengfeng)Relying on the National Supercomputing Internet to connect to the national integrated computing network, it will be fully opened to research universities, enterprises, and individual users inclusive, efficient, and convenient computing-power services.

Currently, on the 100,000-card core node, it has already completed more than 300 HPC-AI fusion applicationsoptimizing , covering large models, robotics, automobiles, innovative drugs, new materials, quantum computing, astronomy and meteorology, and other more than 20 fields. Among them, more than 70 applications have achieved 10,000-card-scale expansion, verifying the stability and reliability of the core node under large-scale, high-load research workloads. For key large applications, it has already achieves protein-folding simulation, trillion-atom-scale water molecular dynamicshundred trillion-grid turbulence simulation and other .

As Sugon 8000 (Dengfeng)has officially come online as a fully China-developed 100,000-card The technology, ecosystem, application, and service standards of AI infrastructure have completed closed-loop validation. During the conference, Sugon also, together with Beijing scientific intelligence research Academy reached a strategic partnership and launched a second fully China-developed 100,000-card HPC-AI fusion computing-power systemdevelopment and construction. Facing AI4S, large-model, and other large-scale computing-power demand, 100,000-card full-precision computing centers are expected to move from demonstration projects to scaled replication. In a new stage in which the global AI industry is becoming more “rational and application-oriented,” they are set to become standard next-generation infrastructure supporting comprehensive “AI+” deployment and empowering scientific discovery and the industrial economy.#Sugon