Huawei Launches 11 AI Chips And Atlas 960 SuperPoD In Direct Challenge To Nvidia

Huawei Launches 11 AI Chips And Atlas 960 SuperPoD

 

OpenLife News reports that Chinese tech giant Huawei Technologies has unveiled 11 new AI-related chipsets in a bold move to challenge the global dominance of Nvidia, Intel and AMD in artificial intelligence computing.

The announcement was made on Thursday at the Huawei Connect 2026 conference in Shanghai by the company’s Rotating Chairman, David Wang.

According to Wang, the 11 semiconductors are built on Huawei’s UnifiedBus 2.0 technology (Lingqu in Chinese) — a high-speed interconnect technology designed to allow thousands of processors, memory modules and computing resources to work together as a single computer.

The chips cover computing, storage, management and interconnect, forming the backbone for Huawei’s large-scale AI infrastructure — its SuperPods and SuperClusters.Ascend 960 Coming Q1 2027Wang announced that Huawei’s next-generation AI accelerator, the Ascend 960DT designed for AI model training, will launch in Q1 2027, three quarters ahead of schedule. It will double the performance of the previous generation.

A second chip, the Ascend 960PR for inference, will arrive in Q3 2027.

He added that Huawei will maintain a one-generation-per-year pace, with Ascend 970 expected in 2028 and Ascend 980 in 2029.“Our main goal is to help industries build vital AI infrastructure and offer an alternative to China and the world. Among these chips, Ascend is the most critical one in the entire system,” Wang said.

Huawei also unveiled its latest Atlas 960 SuperPoD cluster, which promises 2.3x improvement in training and 2.5x improvement in inference for 10-trillion-parameter models compared to the Atlas 950 SuperPoD launched last year.

The company says its new UnifiedBus architecture combined with near-packaged optics and Hi-ONE optical systems will allow a SuperPoD to connect up to 4,000 processors, while its largest systems, SuperClusters, can support up to 1 million AI processors. cio.economictimes.

Wang revealed Huawei has already shipped over 1,000 Atlas SuperPoD supernodes to more than 370 customers, with the Atlas 950 now in large-scale commercial.

The launch is seen as a key part of China’s push for self-sufficiency in AI chips as US export controls continue to restrict Beijing’s access to advanced chips from Nvidia and others.

Huawei’s Chairman of the Supervisory Board, Guo Ping, previously told staff: “Huawei’s goal is to become Nvidia,” aiming to make large language models run on Huawei’s CANN platform just like they do on Nvidia’s CUDA.