On July 17, the 2026 World Artificial Intelligence Conference (WAIC) opened in Shanghai under the theme “Intelligent Partners, Creating the Future Together.” During the conference, China's first fully China-developed 100,000-card AI supercluster—Sugon 8000 (Dengfeng)made its first public appearance before a global audience and was selected as a WAIC 2026 “centerpiece exhibit.”

As an important global exchange platform for artificial intelligence, WAIC is both a window on frontier technologies and innovative applications and an important barometer of industrial evolution. At this year's conference, from large models and agents to embodied intelligence, AI is moving deeper into research innovation and industrial application. Meanwhile, as the key foundation supporting AI's continued evolution, computing-power infrastructure has moved from backstage to center stage and become a major focus of the event.

As artificial intelligence accelerating toward the scaled-application stage, The development focus of AI infrastructure is shifting from raising standalone compute performance to strengthening system-level capability. Facing more complex scenarios such as world models, physical AI, multimodal applications, and AI for Science, combining mixed-precision computing, high-density deployment, efficient interconnect, and sustained stable operation has become an important direction for next-generation AI infrastructure.

As the industry's development direction the first to answer, Sugon 8000 follows an HPC-AI fusion technology path, achieving native fusion of scientific computing and intelligent computing. It supports full-precision computing from FP64 to INT8 and covers scientific computing, large-model training and inference, industrial simulation, and more. The system links chips, computing, storage, networking, cooling, applications, and services, and through full-chain coordination meets 100,000-card-scale, multi-type composite computing needs.

achieves 100,000-card scale running efficiently, places greater demands on ultra-large-scale Behind the cluster's systems-engineering innovation capabilities. Sugon 8000 adopts a globally pioneering high-density cabinet structure. Computing power density per compute unit is 20 times that of other supernodes, packing more computing power into limited space. It also adopts in-house scaleFabric IB-like native RDMA high-speed interconnect, delivering ultra-high-speed, stable interconnection at 100,000-card scale and providing network support for ultra-large-scale parallel computing and resource coordination.

Currently, Sugon 8000 has already relied on National Supercomputing Internetconnected to the national integrated computing-power network, completed optimization of more than 300 key applications, covering large models, robotics, innovative drugs, new materials, quantum computing, astronomy and meteorology, and other more than 20 scientific and industrial fields, with more than 70 applications have achieved 10,000-card-scale expansion and support continued operation of scientific intelligence workloads such as protein-folding simulation, trillion-atom-scale water molecular dynamics, and ultra-large-scale turbulence simulation, as well as AI applications including large-model training, scientific agents, and industry agents.

As artificial intelligence continues to move toward scientific intelligence, industry agents, and physical AI and other more complex scenarios, ultra-large computing-power systems are moving from supporting model training toward supporting research innovation and industrial intelligence. As the first fully China-developed 100,000-card AI supercluster to enter operation, Sugon 8000 shows that AI infrastructure is shifting from single-point technology competition to system-capability competition, offering a new practical reference for observing the next generation of global AI infrastructure.#Sugon