China's AI Gambit: Huawei Says Its SuperPod Crushes Nvidia's Power by 6.7x

Huawei has laid out its most ambitious AI roadmap yet, unveiling a new generation of Ascend-powered SuperPods that could rewire the AI chip landscape in China. Speaking Thursday, rotating chairman Eric Xu detailed a SuperPod design capable of linking up to 15,488 neural processors into a single system. While individual chips may still trail Nvidia's NVDA best-in-class silicon, Xu emphasized that cluster-based computingbolstered by Huawei-designed high-bandwidth memory and next-gen Ascend 950 chips coming next yearcan deliver results at scale. By 2028, Huawei expects to roll out its 970 series, continuing a march toward hardware self-sufficiency. Xu claimed the new Atlas 950 SuperPod could deliver up to 6.7x the compute power of Nvidia's NVL144, leveraging density over node advantage.

The timing isn't accidental. With Washington still restricting Nvidia's high-end chip exports and Beijing now advising local firms to steer clear of Nvidia's RTX Pro 6000D and H20 models, Huawei is stepping into a vacuum. Despite having licenses, Nvidia has yet to ship H20s to China, citing the impact of government guidance. That leaves local chipmakers like Huawei and Cambricon with a rare window of opportunity to fill. Huawei's founder Ren Zhengfei has already acknowledged the company's fabrication gap, but said performance can be recaptured through compensating with cluster-based computing. The pivot seems strategic: if the fabs aren't available, scale up the system instead.

Investors are responding. Cambricon's stock has surged, and Chinese tech namesAlibaba BABA, Baidu BIDU, DeepSeekare gaining traction as AI momentum builds. Huawei, meanwhile, is planning a super cluster that could scale to one million GPUs, using its SuperPod architecture as the backbone. Analysts like Forrester's Charlie Dai frame this as a structural milestone: a systems-level breakthrough, not a chip-level one. While it won't erase the performance delta with US peers overnight, it signals that China's AI hardware ambitions are very much aliveand increasingly domesticated.