China unveils platform for domestic supercomputing chips

China's Computer Network Information Center of the Chinese Academy of Sciences (CAS), the University of Science and Technology of China (USTC), the CAS Institute of Mechanics, and supercomputing company Sugon unveiled the Yisuanfangzhou ("Computing Universe") software platform in Beijing on June 30, 2026, according to China Daily. The platform addresses a long-standing bottleneck: although China has built world-class supercomputers using domestic CPUs and GPUs, NVIDIA's CUDA ecosystem - built over nearly two decades with contributions from millions of developers - has remained the dominant software dependency, making migration to non-CUDA hardware costly and slow. Lead developer Wang Yangang of CAS told China Daily: "That ecosystem has become a deep moat that is very hard to cross. Even if your machine ranks first in the world, you won't be able to unlock its true performance."

Platform architecture Yisuanfangzhou comprises three components, as described to China Daily by the development team:

• •Jiuyanshu is an algorithm library packaging 16 high-performance solvers covering linear algebra, parallel computing, fluid dynamics, biological computing, and deep learning. Key modules deliver over 10x the performance of generic implementations, per the team.
• •BoundX is an AI-driven CUDA code translation engine. Wang told China Daily that code migration previously requiring approximately 10 hours of manual line-by-line engineering can now be processed in roughly 30 minutes, with the converted program submitted directly into a domestic supercomputing environment. The team reports a 71% automated conversion success rate, more than 20 percentage points above other domestic tools.
• •Agent-HiReFlow is a natural-language simulation agent for engineering applications including aerospace fluid dynamics. Professor Cheng Wan of USTC described it as translating user language into physical models and software instructions, then configuring parameters, launching solvers, monitoring faults, and delivering visualization results without any user code. In testing, the system completed a 10-million-cell hypersonic flow simulation in approximately one hour, matching performance previously achievable only on top-end NVIDIA hardware, per the development team.

Nie Hua, chairman of Sugon subsidiary Zhongke Controllable Information Industry, told China Daily that while China has "largely solved the problem of 'having nothing'" in chips, "software-hardware co-design is the next big hurdle." The platform will integrate with the Orient supercomputing system and the National Supercomputing Internet, giving approved researchers access to domestic chip capacity without rebuilding software stacks.

Context This announcement follows China's LineShine system topping the TOP500 ranking in June 2026 with all-domestic chips. Hardware milestones alone do not translate to workload adoption - the BoundX migration data is the practical signal to watch, as real-world conversion rates will determine how quickly Chinese research and enterprise teams can shift CUDA-dependent workloads.