Rent a supernode by the hour - Alibaba brings frontier-scale AI compute to the public cloud, but only if you live in this remote Chinese province

Alibaba Cloud has begun offering its Lingjun Zhenwu M890 supernode instance for commercial use, allowing customers to rent a 64-card, high-speed interconnected computing unit. This marks the first time Alibaba has directly sold supernode computing power through their public cloud platform, with availability currently limited to Inner Mongolia Autonomous Region in North China.

The supernode is a set of accelerators tightly wired together, allowing software to treat them as a single, very large chip rather than a network of separate ones. This approach is often necessary for managing models larger than what a single card can accommodate. While not faster than existing alternatives like Nvidia's GB200 NVL72 or AMD's 72-GPU Helios configuration, Alibaba's offering provides advantages to domestic Chinese consumers, including the use of Chinese silicon and lower latency.

The underlying Zhenwu M890 processor, designed by T-Head, features 144GB of HBM memory with inter-chip bandwidth up to 800 GB/s and an ICNSwitch 1.0 chip that enables interconnect bandwidth up to 1,200 GB/s. Alibaba claims the configuration can handle AI models with up to 10 trillion parameters, making it capable of powering commercial services for Kimi K3 and Qwen3.8-Max.

The company has not disclosed pricing or power consumption information, and benchmarks comparing it to mainstream options are limited. Despite these constraints, the move highlights China's increasing self-reliance in the AI industry and its ambitions for further expansion in the near future.