Woofun AI reports that Alibaba Cloud has deployed the Zhenwu M890 super node on the Bailian platform to deliver inference services for Qwen3.8. Alibaba Cloud states this represents the first domestic super node capable of executing large models with parameter counts exceeding 20 trillion. Qwen3.8 serves as Alibaba’s latest flagship model, featuring a parameter scale reaching 24 trillion. The Zhenwu M890 is a next-generation training and inference integrated AI chip developed by PingTouGe. The super node instance links 64 Zhenwu M890 chips using the ICN Switch 1.0 interconnection chip, delivering an interconnection speed of 800GB/s and a video memory capacity of 9TB. This configuration enables a single instance to meet the expert parallel processing requirements of large-scale 20 trillion parameter MoE models.