At the opening of the 2026 World Intelligent Industry Expo in Tianjin on May 28th, numerous innovative AI technologies and products were showcased. Dawning Information Industry Co.,Ltd. presented a massive 1,700 square meter immersive exhibition, focusing on its new-generation computing infrastructure system for the AI era. The highlight was the physical debut of the scaleX ten thousand GPU super cluster, which became a focal point of the event. Concurrently, Li Jun, Chairman of Dawning, delivered a keynote speech at the opening ceremony. He stated that AI for Science (AI4S) is reshaping how humanity understands the world, with foundational scientific breakthrough capabilities becoming a new decisive factor in global technological competition. He emphasized that humanity is accelerating into the "Scientific Intelligence Era," where hyper-intelligent, full-precision computing is beginning to redefine AI forms. In the future, the tight coupling and systematic collaboration between high-precision scientific computing driven by mathematical models and low-precision intelligent computing driven by data models will be a crucial direction for the evolution of next-generation computing and AI infrastructure.
Systematic Victory: Defining a New Benchmark for AI Computing Power As large model parameters grow exponentially and AI4S develops rapidly, the demand for high computing power with full-precision coverage has shifted from an "option" to a "necessity." The scaleX ten thousand GPU super cluster developed by Dawning is precisely a new-generation, ultra-large-scale AI computing infrastructure designed to meet this challenge. It integrates core technologies such as super nodes, high-speed interconnect networks, and phase-change immersion liquid cooling, aiming to overcome industry-wide challenges like insufficient supply of high-end computing power, rising pressure from cooling and energy consumption, and data transmission bottlenecks. To truly unleash the potential of ultra-large-scale AI clusters, their efficiency and stability depend not on any single hardware component but on the system-level collaboration of full-stack capabilities encompassing computing, storage, networking, cooling, and scheduling. At the expo, Dawning systematically showcased its full spectrum of self-developed technologies and products covering "computing, storage, networking, cooling, management, and software" for the first time, providing a model path for the development of AI infrastructure in this new demand phase. Super Node to Super Cluster Spectrum: In addition to the scaleX ten thousand GPU super cluster, praised on-site as a "national strategic asset," Dawning also displayed physical products including the world's first single-rack 640-GPU super node scaleX640, the world's first cable-free box-type super node scaleX40, and various customized multi-specification super nodes. This matrix of realized products is driving the evolution of AI computing power towards full precision, high density, and low energy consumption. Storage + Networking Foundation: To address the AI era's data demands for "large scale, high throughput, and strong collaboration," Dawning provides globally leading FlashNexus high-end all-flash storage, ParaStor distributed storage technologies and solutions, and a full-stack self-developed scaleFabric high-speed networking product matrix. These effectively break through storage I/O bottlenecks and meet the extreme bandwidth requirements for concurrent read/write operations across ten thousand GPUs. Platform + Scheduling Software: Through platforms like the OneScience scientific large model one-stop development platform and the Gridview cluster operation and maintenance system, Dawning enables full-process management from the physical cluster's digital twin to intelligent scheduling, achieving system availability as high as 99.99% and significantly lowering the barrier for AI application and scientific research innovation.
Ecosystem-Driven: Seizing the New Era of Scientific Intelligence Li Jun stated that in the Scientific Intelligence Era, the full-precision, high-computing-power system, represented by new types of ultra-large-scale AI infrastructure, will become an important "admission ticket" to the new era. "Computing power is the 'electricity' of the AI4S era. Without a powerful computing infrastructure, all scientific intelligence innovation is impossible." In this regard, Dawning has taken the lead in industrial deployment. In February of this year, three sets of the scaleX ten thousand GPU super cluster systems were simultaneously put into trial operation at the core nodes of the National Supercomputing Internet, fully covering large-scale AI computing scenarios such as trillion-parameter model training, high-throughput inference, and AI4S. This also marks the country's first open-architecture ten thousand GPU cluster to achieve actual deployment. Currently, Dawning's AI4S computing cluster has been upgraded to a scale of 60,000 GPUs, having completed over 70 tests at the ten thousand GPU scale. Among these, 30,000 GPU computing power has been applied to protein folding simulations, 45,000 GPU computing power has achieved DFT precision simulation with 41.47 billion atoms, and the efficiency of billion-level grid simulations has been improved from the weekly level to the hourly level. Li Jun also pointed out that the challenge of the Scientific Intelligence Era is not just about "having computing power," but more about how to establish a collaborative system of "data, computing, models, and applications" to truly interconnect data, computing power, models, and applications. Simultaneously, it involves driving continuous breakthroughs in underlying technological capabilities such as advanced packaging, silicon photonics interconnects, advanced cooling, and "computing-power coupling." Therefore, ecosystem development is crucial. Dawning has long been promoting the construction of an open ecosystem for domestic software and models. It has now deeply adapted over 400 globally mainstream large models and achieved localization adaptation and migration optimization of professional software in fields such as fluid dynamics, biology, quantum mechanics, and industrial simulation, further advancing AI infrastructure from being "buildable" to being "usable effectively." From super nodes to super clusters, from core components to system collaboration, and from open architecture to an open ecosystem, Dawning is propelling China's AI industry towards larger scale, higher energy efficiency, and more open collaboration, aiming to seize the strategic high ground in the new global era of Scientific Intelligence.
Comments