- The company said its Zhenwu M890 AI supernode is China’s first to support inference for a model exceeding 2 trillion parameters
- The milestone highlights China’s push to compete in AI infrastructure through system-level integration rather than chips alone
Alibaba Cloud said on July 23 that its Zhenwu (真武) M890 AI supernode has been successfully adapted to run inference for its flagship Qwen3.8 foundation model, which has 2.4 trillion parameters, via the company’s Bailian AI platform.
The company said the deployment marks the first time a domestically developed AI supernode has successfully run inference for a model with more than 2 trillion parameters.
Models at this scale require parameters to be distributed across dozens or even hundreds of high-speed interconnected AI chips, making network bandwidth and memory architecture as critical as raw computing power.
The Zhenwu M890 is Alibaba’s latest in-house AI chip for both training and inference, supporting precisions from FP32 to FP4.
Its supernode architecture connects 64 chips through the proprietary ICN Switch 1.0 interconnect, delivering 800 GB/s bandwidth and 9 TB of aggregated memory—enough to support expert-parallel inference for large mixture-of-experts (MoE) models.
Alibaba said end-to-end optimization across chips, networking and cloud software improved performance by as much as 1.5 times in agentic AI inference workloads.
Earlier this month, the Zhenwu M890 chip and Panjiu (磐久) AI supernode server were named among the flagship exhibits at the 2026 World Artificial Intelligence Conference (WAIC).
From chip performance to system architecture
AI infrastructure competition is increasingly shifting from standalone accelerators to tightly integrated computing systems.
Supernodes—clusters that combine chips, interconnects, memory and software into a unified architecture—have become a strategic focus for China’s AI computing industry.
Industry forecasts estimate China’s domestic supernode market could reach 341.4 billion yuan ($47.6 billion) by 2028, with Huawei, Tencent, Sugon and Lenovo among companies investing in the segment.T
Why it matters for global readers
The achievement illustrates China’s broader effort to compete in AI infrastructure through vertically integrated systems rather than individual chips alone.
By combining proprietary AI chips, foundation models and cloud services into a single stack, Alibaba is pursuing an alternative to Nvidia’s ecosystem.
While software maturity and developer adoption remain long-term challenges, the ability to run trillion-parameter models on a homegrown computing platform suggests China’s AI infrastructure is becoming a more credible alternative for large-scale AI deployment.
