Alibaba Cloud announced today that the Lingjun Zhenwu M890 hypernode instance is officially available, with initial sales in the Ulanqab region of Inner Mongolia.
According to the company, enterprise customers can access 64-card high-speed interconnected computing units in the cloud without building their own data centers. The units can support inference for Mixture-of-Experts (MoE) large models with up to 10 trillion parameters. The Lingjun Zhenwu M890 hypernode is also China's first hypernode computing solution to successfully run a large model with more than 2 trillion parameters. Kimi K3 and Qwen3.8-Max are already providing services through this instance.
The Lingjun Zhenwu M890 hypernode instance supports low-precision FP8/FP4 computing. Its ICN Switch 1.0 chip increases the Scale-up interconnection scale from 16 cards to 64 cards, with inter-card interconnect bandwidth reaching 800 GB/s. In training scenarios such as autonomous driving and embodied intelligence, training performance is 3 times higher than that of the previous-generation Zhenwu 810E.
According to the company, Ulanqab, the launch region for this hypernode instance, is one of Alibaba Cloud's five super data centers. Renewable electricity accounts for approximately 90%, providing a lower-carbon operating environment for high-density computing. The data center also features an industry-first fully modular design architecture. The overall modularization rate for power supply, cooling, security, intelligence, and fire protection reaches 90%, which can significantly shorten data center delivery cycles, maximize compatibility with mainstream heterogeneous chips and computing architectures, and support at least three generations of chip iterations.
This year, Alibaba Cloud plans to more than double the global production capacity of its modular data centers. Related hypernode instances will also gradually become available in other regions.
Link:
https://www.aliyun.com/product/lingjun-hypernode