On September 17, the 11th Huawei Connect 2026 opened in Shanghai. Huawei deputy chairman and rotating chairman Wang Tao announced that the Ascend supernode has been widely adopted, with over 1,000 sets deployed for more than 370 customers—entering real commercial deployment after engineering validation across chips, servers, interconnect and software ecosystem.
Huawei also revealed progress on the next-generation Ascend 960 chip and supernode. Wang said the Ascend 960 doubles performance versus its predecessor and is ready three quarters ahead of schedule, to be ready in Q1 2027; the Atlas 960 will be the industry's first supernode using NPO (near-package optics), with a liquid-cooled version in Q3 2027 and an air-cooled version in Q2 2027.
Wang outlined Huawei's AI strategy: compute is at the core, with "supernode + cluster" building China's solid compute base; an open-source ecosystem supporting mainstream large models; the Pangu model focused on Huawei product intelligence; and flexible offline-plus-online compute schemes bringing AI to devices and vehicles.
Analysts noted the early disclosure reflects Huawei accelerating the iteration of next-generation AI compute products amid intensifying global AI compute competition, a key node for China's domestic AI compute industry following last year's three-year Ascend roadmap.
Source: National Business Daily / Huawei
