Baidu Intelligent Cloud Releases Tianchi Supernode Architecture Specification to Standardize AI Compute Infrastructure

On September 21, 2026, Baidu Intelligent Cloud released the Baidu Tianchi Supernode System Architecture Design Specification, making public the architecture design details and engineering trade-offs behind its Tianchi supernode. The document covers six areas — cabinet, power, cooling, node, interconnect, and management — with design requirements, technical specifications, dimensions, and interface standards intended to support the wider adoption of high-density AI compute infrastructure.

According to Baidu Intelligent Cloud, a supernode is more than packing more AI chips into a rack; it integrates computing, power, cooling, interconnect, and management into a high-density, highly reliable system. The released specification spans from key components to full racks. It defines a standard-width 46U rack solution, a complete power delivery path from facility power to node power, a hybrid air/liquid cooling architecture, a 1U one-node pluggable node design, and a one-hop interconnect latency as low as 1.5 microseconds for 32/64-card scale-up configurations. The management architecture uses rack-level and node-level BMCs to enable second-level metric collection and minute-level fault recovery.

Baidu Intelligent Cloud said it hopes the validated architectures, specifications, and interface requirements can be standardized and shared, giving data centers a more reliable basis for planning, deployment, and maintenance. Previously, Shen Dou, Baidu EVP and president of Baidu Intelligent Cloud, said the Tianchi supernode has achieved a fully domestic supply chain, from core components to liquid cooling and from rack architecture to high-speed interconnect. Baidu added that standardized prefabricated modules have cut AIDC construction cycles by more than 30%.