Huawei unveils Atlas 960 SuperPod to challenge Nvidia in AI computing

Huawei Technologies on Thursday launched the Atlas 960 SuperPoD computing cluster at its annual Huawei Connect 2026 conference in Shanghai, marking the company's most ambitious move yet to replace Nvidia ↗ 2.78% as the backbone of China's artificial intelligence infrastructure.

A New AI Supernode

Rotating chairman David Wang introduced the Atlas 960 SuperPod as the first in the industry to use near-package optics (NPO), a placement structure that enables higher speeds and lower energy consumption. The system pairs Huawei's UnifiedBus interconnect technology with its Hi-ONE optical engine, allowing a single SuperPoD to connect up to 4,096 processors with round-trip latency near two microseconds. Huawei said the cluster delivers 8 EFLOPS of FP8 compute performance and up to one petabyte of high-bandwidth memory capacity.scmp+2

The Atlas 960, powered by Ascend 960 chips, promises a 2.3-fold improvement in training and a 2.5-fold improvement in inference for 10-trillion-parameter models compared to the predecessor Atlas 950 SuperPoD. Huawei also disclosed that its SuperPoD systems have already shipped more than 1,000 units to over 370 customers. Two next-generation chip variants — the Ascend 960DT and Ascend 960PR — are planned for the first and third quarters of 2027, respectively, ahead of the company's previously announced schedule.pandaily+2

'We Want to Be Nvidia'

The product launch accompanies an increasingly explicit strategic declaration. Guo Ping, chairman of Huawei's supervisory board, recently told new employees that "Huawei's goal in the ICT and computing fields is to become Nvidia," according to remarks posted to the company's internal bulletin board. Like Nvidia, Huawei's advantage would not lie in building large language models but in providing the underlying infrastructure to run them, Guo said.chosun+1

Huawei designs its own chips through subsidiary HiSilicon and places large orders with SMIC Semiconductor Manufacturing International Corporation, China's largest advanced foundry, while optimizing major Chinese AI models such as DeepSeek for its Ascend chip ecosystem.chosun

AI Safety and Supply Constraints

Separately, rotating chairman Eric Xu told reporters at the conference that Chinese AI developers "may need to speed up their pace to the level that they could also feel the risks from AI development," arguing that Chinese model providers may not yet be advanced enough to encounter the frontier safety concerns reported by leading U.S. firms. Xu also acknowledged that Huawei cannot produce enough AI computing equipment to satisfy domestic demand and has no plans for full-scale international expansion.reuters+2

"Since we don't have enough capacity to even satisfy the demand in China, we don't have a plan to expand into the international market in a fully-fledged way," Xu said, according to Reuters.reuters