Huawei unveils Ascend AI chips and Atlas clusters to rival Nvidia

Huawei on Thursday unveiled a suite of AI chips and cluster technologies in Shanghai that the company said could deliver world-class computing power without relying on Nvidia hardware, a move Beijing has encouraged as tensions over chip access escalate. At the companys annual Connect conference, Huawei executives described new Ascend AI processors, large-scale Atlas cluster systems and a SuperPoD interconnect designed to link thousands of accelerators into single compute fabrics, and set timelines for chip rollouts beginning in 2026.

Eric Xu, Huawei's deputy chairman and rotating chairman, said the company has developed high-bandwidth memory and a domestic chip roadmap intended to reduce dependence on foreign suppliers. Xu outlined plans for new Ascend generations, including an Ascend 950PR the company said is due in the first quarter of 2026, and described two new logical machines, the Atlas 950 SuperPoD and Atlas 960 SuperPoD. Huawei said the SuperPoDs and related Atlas superclusters will house thousands to millions of Ascend neural processing units, and claimed performance advantages over specific Nvidia configurations.

Huawei also introduced SuperPoD Interconnect technology...