Huawei Cloud will begin commercial operations of its Ascend 950 series AI cluster cloud service in China on September 30 and expand to global markets on November 30, Huawei Cloud CEO Zhou Yuefeng said at Huawei Connect 20261,2. The dates convert the Ascend 950 series hardware into a bookable managed cloud offering on a published calendar, rather than another architecture brief.
The Ascend 950 series system uses Huawei Technologies' Unified Bus interconnect architecture, marketed under the Lingqu name, to assemble a 1,024-accelerator pod. Huawei Technologies said the cluster can deliver up to 1 EFLOPS of FP8 computing power and 2 EFLOPS of FP4 computing power, with 256 TB of globally addressable unified memory. The design is intended to support end-to-end training of large-scale models without additional cluster-level partitioning or adaptation, keeping a thousand-card job coherent under one memory address space.
Rotating Chairman Wang Tao said more than 1,000 Ascend supernodes are already deployed and that Ascend 950 series supernodes have entered scaled commercial use.
Huawei Technologies said its Ascend 960 roadmap is ahead of schedule. The Ascend 960DT is planned for Q1 2027, and the 960PR is planned for Q3 2027. The company indicated a cadence of roughly one major Ascend generation per year thereafter, with longer-range 970 and 980 milestones on the horizon.
ANALYSIS The China-then-global rollout calendar and the published cluster specifications mark a shift from hardware announcements to a concrete cloud-service SKU that prospective customers can plan procurement around. The 1,024-card, EFLOPS-class envelope positions the Ascend 950 series as Huawei Cloud's answer to large-model training demand, backed by the company's broader networking and interconnect investments noted in prior coverage[2].
The November 30 global availability date will be the first test of international demand for the Ascend 950 series cloud service outside China.