Alibaba Cloud launches new AI supernode, first public cloud service

At the 2026 World AI Conference, Alibaba Cloud launched the Lingjun Zhenwu M890 supernode instance, marking its first public cloud AI computing service. It's now available for beta testing in Ulanqab, Inner Mongolia.

The instance supports low-precision FP8/FP4 computing with 800GB/s interconnect speed, capable of running trillion-parameter MoE models on a single node. In training tasks like autonomous driving and embodied AI, it delivers three times the performance of the previous generation.

With HPN 8.0 architecture, a single cluster can handle up to 130,000 heterogeneous GPUs, scalable to one million. The system offers 99.7% average availability and minute-level fault recovery, keeping computing power focused on training.