Alibaba Cloud Launches Lingjun Zhenwu M890 Super Node Instance, Now Available in Ulanqab

On August 12, Alibaba Cloud announced the official launch of its Lingjun Zhenwu M890 Ultra Node instance, now available in the Ulanqab region of Inner Mongolia. This instance enables enterprise customers to provision 64-GPU high-speed interconnected computing units directly on the cloud—without needing to build their own data centers—and is designed for inference tasks involving MoE (Mixture-of-Experts) large models with parameters reaching the tens of trillions scale. As China’s first ultra node capable of successfully running large models exceeding two trillion parameters, it is already powering services such as KimiK3 and Qwen3.8-Max. The Lingjun Zhenwu M890 Ultra Node supports low-precision FP8/FP4 computations. Leveraging the ICNSwitch 1.0 chip, it scales up interconnect capacity from 16 GPUs to 64 GPUs, achieving an inter-GPU bandwidth of 800 GB/s. In training scenarios such as autonomous driving and embodied intelligence, its training performance is three times higher than that of the previous-generation Zhenwu 810E. Ulanqab hosts one of Alibaba Cloud’s five global hyperscale data centers, with approximately 90% of its power sourced from renewable energy. Built with a fully modular architecture, the facility achieves a 90% modularity rate across power supply, cooling, security, intelligent management, and fire protection systems—significantly shortening deployment timelines. It is compatible with mainstream heterogeneous chips and computing architectures and supports at least three generations of chip iterations. Alibaba Cloud plans to more than double its global production capacity for modular data centers by 2024, with related ultra node instances gradually rolling out to additional regions.