Alibaba Cloud said on July 23 that its Zhenwu M890 supernode supports Qwen3.8 and is live on its Bailian platform for model inference services.

Alibaba said this makes it the first supernode in China to run a large model with more than 2 trillion parameters.

Alibaba previously said its M890-based Panjiu AL128 supernode server packs 128 accelerators into one rack with petabyte-per-second internal bandwidth.

The M890 chip also has 144 GB of on-chip memory and 800 GB per second of inter-chip bandwidth.

Alibaba has not disclosed Qwen3.8’s architecture or parameter breakdown.

Public Qwen3 releases include mixture-of-experts models with 235 billion total parameters and 22 billion activated parameters, so total parameter counts do not necessarily reflect runtime compute.

Other Chinese companies have also discussed supernode products.

Sugon, a China-based high-performance computing company, said its scaleX platform is a 10,000-card AI supercluster for trillion-parameter workloads.

🔗 Source: Security Times

Recent Alibaba developments