Wanka super-cluster system to debut, tackling AI ecosystem challenges

Dec 13, 2025 | AI | Lian Zheng

Wanka super-cluster system to debut, tackling AI ecosystem challengesChina is preparing to unveil its first large-scale domestic super-cluster system at the upcoming Photosynthetic Organization 2025 Artificial Intelligence Innovation Conference (HAIC 2025), scheduled for December 17-19 in Kunshan. The system, known as the Wanka super-cluster, is designed to address pressing ecological challenges in the country's computing power industry.

As generative AI accelerates adoption across industries, large-scale intelligent computing clusters have become essential infrastructure. However, difficulties in integrating diverse computing resources and inefficiencies in cross-vendor equipment collaboration have slowed domestic progress. The Wanka super-cluster aims to provide a technical solution to these issues.

The system is built on Shuguang's scaleX640 super-node technology. This architecture uses a "one-to-two" high-density design and incorporates domestically developed high-speed interconnection networks. By supporting compatibility across multiple domestic accelerator cards through an open framework, the Wanka super-cluster breaks down barriers within the AI ecosystem.

Earlier reports highlight that the scaleX640 is the world's first single-cabinet-level super-node capable of housing 640 cards. It combines ultra-high-density blades, submerged phase-change liquid cooling, and high-voltage direct current power supply. Compared with traditional industry solutions, the system doubles overall computing performance. For mixture-of-experts (MoE) trillion-parameter large models, training and inference efficiency can improve by 30% to 40%.

This debut marks a significant step in China's efforts to build independent, high-performance computing infrastructure. By enhancing efficiency and compatibility, the Wanka super-cluster is expected to play a key role in supporting next-generation AI applications and reducing reliance on foreign technologies.

Latest news