Huawei unveils Atlas 960 SuperPoD as China steps up the AI hardware race with Nvidia

On September 17, Chinese telecoms giant Huawei announced new AI chip technologies and its latest supercomputer‑class cluster, the Atlas 960 SuperPoD, sharpening its challenge to U.S. chip leader Nvidia. The new system succeeds a cluster Huawei launched only months ago and is designed to significantly boost performance for both training large language models and running them in production, a phase known as inference.

U.S. export controls have tightened access to cutting‑edge GPUs for Chinese firms, turning homegrown AI compute into a strategic priority for Beijing and its tech champions. By highlighting a full stack of in‑house chips and servers, Huawei is pitching the Atlas 960 SuperPoD as a “national” AI supercomputer platform for cloud providers, research institutes, and state‑linked projects. Analysts note that Nvidia still holds advantages in power efficiency and software tooling, thanks to its CUDA ecosystem and mature libraries. The real test for Huawei will be whether developers find its hardware and software sufficiently compatible and performant to serve as a viable alternative for training and deploying frontier‑scale models.