The ambition of a city's computing power is rarely written in a PPT, but rather on a practical specification. On July 21st, under the spotlight of the 2026 World Artificial Intelligence Conference, Shanghai Yidian Zhi'sai took the initiative to do something that the industry has long called for but no one had taken the lead on: it jointly established the Intelligent Computing System Architecture Alliance with more than 20 upstream and downstream industry chain entities including GPU chip, interconnection, system integration, large model, and software sectors. At the same time, the alliance released its first achievement, "The Super Node System Architecture Specification." The Shanghai Municipal Commission of Economy and Information Technology stood as an advisory unit, while the Yidian-Zhongxing Joint Laboratory took over the daily operations.

image.png

Intelligent computing power has long been the heart of the digital economy, but this heart does not beat smoothly. There is a mismatch between the supply and demand of computing resources, and there are high walls of compatibility between software and hardware. Even the strongest single point cannot build a complete network. As a core enterprise of the Shanghai Intelligent Computing Public Service Platform, Shanghai Yidian has invested a lot of effort in the infrastructure in recent years: several intelligent computing centers have been completed, dozens of domestic AI core components have passed testing and verification, more than a hundred test reports have been accumulated, and the Yidian Intelligent Computing Cloud Platform was independently developed, which has truly made the unified management and intelligent scheduling of heterogeneous computing power work effectively. The mission of the alliance is thus clearly defined: to gather the full-stack ecosystem and build an intelligent computing system architecture, and to practice the responsibility of state-owned enterprises to create an autonomous and controllable infrastructure. It provides systematic and integrated planning and construction solutions for scenarios such as intelligent computing centers, supercomputing clusters, and industry computing bases, rather than another conceptual blueprint.

The value of the alliance lies in its member list. Led by Yidian Zhi'sai, it is backed by 23 manufacturers covering the entire industrial chain: Stepwise Star, Xiyu Technology, Tiangshu Zhixin, Biren Technology, Pingtouge, Hanwuji, Eastern Computing Chip, Muxi Co., Ltd., Suiyuan Technology, Silang Technology, ZTE Communications, HuaXin San, Inspur Group, Suanfeng Information, Sunlight Information, Xizhi Technology, YuXun Network, Qingcheng Jizhi, Qujing Technology, Jiliu Technology, Silicon Flow, Yunmiecunlian, and Yunhezhiwang. From the underlying chips all the way up to the application layer, the purpose of this collaboration chain is clear: to combine heterogeneous computing power and technical resources into one force, break through the computing power barriers, and promote the compatibility and coordination of software and hardware. For alliance members, this membership ticket offers three tangible benefits: participating in the discussion and preparation of industry standards such as the super node construction specifications, using the real machine testing and verification environment of the alliance to efficiently complete compatibility and optimization, shortening the R&D iteration cycle, and leveraging the platform to connect leading enterprises and national-level laboratories to carry out joint research on cutting-edge technologies, co-apply for major scientific research projects and demonstration engineering, and accurately connect the supply and demand of the industrial chain.

What truly transforms the alliance from a loose coalition into a binding constraint is the simultaneously released "Super Node System Architecture Specification" version 1.0. This specification is divided into hardware and software. On the hardware side, it specifies the use of cold plate liquid-cooled rack-mounted cabinets, uniform node dimensions, and blind insertion of electrical, liquid, and network three bus lines for front maintenance. It also includes centralized power supply, liquid cooling, and leak detection isolation. It completes the Scale-up deployment through high-speed interconnection backplanes—effectively setting a common physical language for machines from different manufacturers to be compatible and stable. On the software side, it clearly outlines requirements for operation and maintenance management, cluster integration and topology awareness scheduling, collective communication, storage and data loading, security and trustworthiness, and conformity acceptance, ensuring that upper-level scheduling follows clear guidelines. Shanghai Yidian has clearly stated that it will strictly follow this specification to build its own super node cluster.