China unveils first major projects powered by homegrown 100,000-card AI supercluster
China has unveiled the top 10 application achievements built on its first domestically developed AI supercluster equipped with 100,000 computing accelerators. The results were presented Sunday in Shenzhen at the 2026 CCF National Annual Conference on High Performance Computing.
The supercluster, known as Dawning 8000 or Dengfeng, was completed and put into operation in July this year.
Its significance lies not only in scale. The system combines traditional supercomputing and AI computing on a single infrastructure, allowing it to handle both high-precision scientific simulations and large-scale AI model training and inference.
Among the top 10 applications are the world's first digital twin of the human brain at the scale of 100 billion neurons, the HyperMillennium project simulating the evolution of a universe spanning tens of billions of light-years, and a next-generation weather model capable of producing global forecasts at one-kilometer resolution within one hour.
The supercluster is also being used to study the Yellow River. A team from Tsinghua University's Institute of River and Ecology is collecting suspended sediment data at different depths to build a more accurate digital model of the river.
Previously, such models were constrained by limited computing power. According to institute director Sun Jian, earlier systems relied on grid sizes of tens of meters, while the current target is to reduce that scale to about one meter.
Modeling the interaction between water and sediment in the Yellow River is particularly complex, and researchers hope Dawning 8000 will significantly improve the precision of these simulations.
The computing center is located in Zhengzhou. Just three years ago, the area was undeveloped land; today it has become China's largest AI computing pool.
The system deploys 100,000 domestically produced intelligent computing chips and is currently operating close to full capacity.
Its infrastructure includes 13,440 computing nodes, all of which can be monitored through a unified control system.
More than 6,000 tasks are running on the supercluster in real time. In some cases, a single project can make use of nearly the entire 100,000-card system.
Developers emphasize that Dawning 8000 is designed not only for AI but also for scientific and engineering computing, making it a versatile platform for next-generation workloads.
According to CCTV+, the first major projects running on Dawning 8000 show China moving from building domestic computing infrastructure toward large-scale practical use in science, weather forecasting, natural-system modeling and artificial intelligence.
