China Unveils First Fully Domestic 100,000-Card AI Supercluster at Intelligent Computing Conference in Zhengzhou

According to Securities Daily, the 2026 Intelligent Computing Application Conference hosted by the Photosynthetic Alliance ran from 9 to 11 July in Zhengzhou, Henan Province. On the exhibition floor stands Sugon 8000 (Dengfeng), the country’s first fully domestic supercluster equipped with 100,000 AI accelerator cards. The installation marks a landmark shift in China’s AI infrastructure rollout, moving development past the 10,000-card tier to the next generation of large-scale compute deployment.

Senior secretariat staff from a special committee under the China Computer Federation shared commentary with Securities Daily on the hardware milestone. Ongoing engineering deployments mean 100,000-card scale AI infrastructure will transition from landmark demonstration projects to standard industrial deployment over the coming period. These unified compute resources will underpin the national integrated computing network, deliver breakthroughs across frontier scientific research disciplines, and accelerate digital transformation for industrial sectors nationwide.

Iterative Upgrades Reshape Benchmarks for Large-Scale Compute Infrastructure

Discussions centred on the 100,000-card architecture dominated panel sessions throughout the conference. Industry frameworks once treated a 10,000-card cluster as the baseline hardware threshold for training trillion-parameter large language models. The explosive growth of ultra-large AI models means systems built around 100,000 accelerators will become a mandatory hardware baseline for next-generation intelligent computing facilities.

A senior vice-president at Sugon Information Industry Co., Ltd told Securities Daily that a 100,000-card cluster demands far more than sheer volume of accelerator hardware. Performance hinges on balanced mastery of system architecture, inter-node networking, memory access throughput, energy efficiency management, and cross-industry application ecosystem integration.

Field assessments of Sugon 8000 (Dengfeng) confirm the platform abandons conventional partitioned cluster layouts to adopt a native hyper-converged supercomputing and AI architecture. The unified framework delivers seamless integration between high-precision scientific computing and low-precision intelligent workloads, supporting deployment across scientific simulation, large model training, AI inference and industrial digital twin workflows.

A senior engineering lead at the High Performance Computing Centre of Nanjing University explained to Securities Daily that linear speed-up scaling represents the core technical hurdle for ultra-large compute deployments. Simple hardware aggregation generates steep efficiency losses driven by network latency and excessive power draw. Sugon 8000 (Dengfeng) validates the capacity of fully domestic hardware stacks to sustain high concurrency and data throughput even at the 100,000-card scale. The breakthrough signals Chinese AI infrastructure development has moved beyond the narrow pursuit of raw theoretical peak performance, shifting focus to practical operational efficiency and holistic system capability. The unified compute architecture eliminates data transmission bottlenecks while cutting overall energy consumption, enabling faster, more stable simulation and modelling for autonomous vehicle development and medical pharmaceutical research pipelines.

Cross-Chain Industrial Collaboration Delivers Full Domestic Tech Stack

A prominent display panel beside the Sugon 8000 (Dengfeng) hardware outlines complete end-to-end independent development spanning core chips, compute nodes, distributed storage, high-speed networking, liquid cooling systems, application software and after-sales technical services, demonstrating tangible progress in coordinated domestic supply chain advancement.

A comprehensive indigenous computing industrial ecosystem has formed around the 100,000-card supercluster. The hardware stack relies on domestically designed Hygon processors at its base layer, paired with integrated system assembly and upper-layer model development tools. Exhibition zones operated by the Photosynthetic Alliance drew consistent foot traffic throughout the event. Sugon Information Industry Co., Ltd showcased joint industry solutions alongside Guizhou Xiangming Technology Co., Ltd, Guizhou University, Guizhou Tuji Technology Co., Ltd and Beijing Suyan Intelligence Technology Co., Ltd, all ecosystem participants building operational tools upon fully domestic compute foundations.

A technical deputy director at Guizhou Xiangming Technology Co., Ltd spoke to Securities Daily about Qianshu, a new trusted intelligent computing base co-developed by Photosynthetic Alliance members. The platform resolves four core operational barriers facing enterprise large model and autonomous agent rollout: restricted secure data utilisation, fragmented compute resource scheduling, slow AI service commercialisation and opaque workflow audit trails. Live deployments are already running across AI pathology operating systems, digital research institute platforms and automated office collaboration suites.

This coordinated cross-industry progress dismantles siloed hardware and software development that previously limited domestic compute rollout. Sugon’s fully self-developed scaleFabric high-speed interconnect network mitigates critical communication bottlenecks within 100,000-card clusters, while driving parallel technical upgrades for domestic switching chips and optical transceiver modules across upstream and downstream supply chains.

A vice-president at the Beijing subsidiary of Sugon Information Industry Co., Ltd stated that Sugon 8000 (Dengfeng) verifies technical viability for fully indigenous development covering every layer from core silicon chips to immersion phase-change liquid cooling hardware. The liquid cooling architecture resolves persistent energy overhead challenges, placing domestic intelligent computing hubs among global front-runners for green low-carbon operational metrics. The platform also drives deep integration and capability expansion across China’s complete domestic computing industrial chain.

Shifting Industry Focus from Infrastructure Build-Out to Operational Utilisation

Hardware deployment constitutes only the initial phase of development. The greater technical challenge lies in fully saturating 100,000-card clusters with complex real-world industrial and research workloads, a priority highlighted across application launch sessions at the conference.

Demonstration showcases moved beyond static server hardware displays to feature live end-to-end industrial modelling pipelines. Sugon 8000 (Dengfeng) connects to the national supercomputing internet and forms part of the national integrated computing network. Engineers have completed targeted optimisation for over 300 hybrid supercomputing and AI workloads spanning large language models, robotics, quantum simulation and advanced new material research. More than 70 distinct applications support stable expansion to 10,000-card operating scales, verifying consistent reliability under sustained high-load research tasks.

A lead researcher at the Intelligent Watershed Laboratory within the Yellow River Institute of Hydraulic Research shared practical case studies with Securities Daily on site. Hydraulic sediment simulation workflows previously operated solely via supercomputing numerical modelling. Twenty years of iterative upgrades moved processing from CPU parallelisation to GPU acceleration, compressing calculation runtimes from hours to minutes, yet only delivering faster computation speeds.

Combined hyper-converged compute resources delivered by Sugon 8000 (Dengfeng), paired with physical artificial intelligence and scientific large model frameworks, enable simulation workflows to deliver precise modelling, early risk forecasting and automated operational optimisation simultaneously. The research institute is now developing high-fidelity hydraulic digital models and dedicated knowledge large language models for water resource management supported by the unified compute cluster. The new architecture replaces sequential workflows of observation, calculation and risk assessment with integrated simulation prediction, intelligent optimisation and safety monitoring. The technology framework delivers transformative upgrades for Yellow River regulation and digital transformation across the wider water conservancy sector.

During the conference, Sugon Information Industry Co., Ltd finalised a strategic cooperation framework with the Beijing Institute of Scientific Intelligence, launching research and development work for a second fully domestic 100,000-card hybrid supercomputing and AI cluster.

A council member at the Capital Enterprise Reform and Development Research Association outlined industrial takeaways from the platform launch. Deployment of Sugon 8000 (Dengfeng) fills a domestic gap in fully independent 100,000-card scale intelligent computing infrastructure. The open collaborative compute ecosystem bridges operational gaps between underlying hardware layers and top-tier industry applications. The launch opens a new development chapter for Chinese intelligent computing, shifting industry priorities from constructing compute infrastructure to maximising practical operational value, and creates robust foundational support for nationwide rollout of the AI Plus industrial initiative.

Engineering teams will continue advancing iterative optimisation for the two deployed 100,000-card superclusters, expanding the portfolio of optimised industrial and research applications running on fully domestic hardware stacks. Supply chain participants will scale production for core indigenous chips, networking hardware and liquid cooling components to match growing demand for large-scale intelligent computing facilities across provincial integrated compute hubs. Joint technical working groups formed by Photosynthetic Alliance members will release standardised deployment toolkits to lower technical barriers for enterprises and research institutions adopting domestic ultra-large compute infrastructure.