Alibaba Cloud has officially launched its Lingjun Zhenwu M890 supernode instance, with initial availability in the Ulanqab region. Enterprise clients can now access a 64-card high-speed computing unit directly from the cloud without needing to build their own data centers, supporting MoE large language model inference with up to ten trillion parameters.

The Lingjun Zhenwu M890 supernode is the first supernode-form computing power in China to successfully run a large model with over two trillion parameters. Both Kimi K3 and Qwen3.8 Max are already providing services through this instance. The supernode supports FP8 and FP4 low-precision computing, and its Scale-up interconnect has been expanded from 16 cards to 64 cards using the ICN Switch 1.0 chip, with inter-card bandwidth reaching 800 GB/s.

In training scenarios such as autonomous driving and embodied intelligence, the M890 delivers a threefold improvement in training performance compared to the previous generation Zhenwu 810E. The supernode cluster is powered by Alibaba Cloud's unified infrastructure, the Lingjun Intelligent Computing Platform. Leveraging the HPN 8.0 training-inference integrated network, a single cluster supports up to 130,000 heterogeneous computing cards and can scale to million-card levels.