Alibaba Cloud has made its Lingjun Zhenwu M890 supernode instance available in China, with Ulanqab a...
The AMW Read
Alibaba's M890 supernode-as-a-service is a new offering in AI infrastructure, directly targeting MoE inference at scale, building on its existing cloud presence.
Alibaba Cloud has made its Lingjun Zhenwu M890 supernode instance available in China, with Ulanqab as the first deployment region. The offering lets enterprise customers provision 64-card, high-speed-interconnect computing units through the cloud, avoiding the need to build their own data centers. The instance targets inference for mixture-of-experts models with up to 10 trillion parameters, and is already powering Kimi K3 and Qwen3.8-Max services.
This move signals a strategic push by Alibaba Cloud to offer massive-scale, ready-to-use AI compute to enterprises, particularly those needing to run cutting-edge MoE models. By packaging 64 GPUs with high-speed interconnect as a cloud service, Alibaba is commoditizing what would otherwise be a complex and expensive infrastructure build. The inclusion of Kimi K3 and Qwen3.8-Max shows Alibaba is actively courting both its own Qwen models and rival labs, positioning the M890 as a neutral yet powerful inference platform in China's increasingly competitive AI cloud market.
For builders and investors, the M890 offers a low-friction path to deploying trillion-parameter MoE models, reducing the upfront capital and operational overhead of AI inference. This could accelerate enterprise adoption of frontier-scale models in China, while also intensifying competition among domestic cloud providers on price and performance. Watch for whether Alibaba extends this supernode-as-a-service model beyond Ulanqab to other regions, and how it prices against GPU-rental alternatives.
