Huawei Rolls Out Faster AI Cluster to Close Gap With Nvidia

Huawei’s new computing cluster and planned chips arrive as the United States maintains export controls on advanced semiconductors, tightening the contest for leadership in artificial intelligence systems. The company unveiled the Atlas 960 SuperPoD at its annual conference in Shanghai on September 17, claiming faster performance in both training large models and running inference tasks than the Atlas 950 model released late last year. It also outlined plans to release the Ascend 970 chip in 2028 and the Ascend 980 chip in 2029.

The timing places the announcement less than two weeks before the scheduled September 24 meeting in Washington between President Donald Trump and President Xi Jinping. U.S. restrictions have blocked Chinese firms from the most advanced foreign chips and manufacturing equipment, prompting Beijing to accelerate domestic alternatives. George Chen, partner and chair of the digital practice at The Asia Group, noted that the rapid upgrade cycle “will only heighten U.S. anxiety about the accelerating AI competition.”

Analysts cited in the reporting observed that many Chinese AI developers still depend on Nvidia hardware for the most demanding training workloads, yet domestic chips from Huawei are gaining adoption. Trump has stated that the United States must retain its lead, declaring that “whoever wins in AI, wins.” Chen added that developments move too quickly for any single country to assume permanent dominance.

The Atlas 960 SuperPoD announcement follows a pattern of successive hardware releases at shorter intervals than earlier generations. No independent benchmark data comparing the new cluster against current Nvidia systems was released at the event.