Login to Continue Learning
[Ad_1]
Huawei has announced new iterations of its ‘SuperPod’ AI clusters at the Huawei Connect 2025 event. These include the Atlas 950 and the Atlas 960, with the former featuring the new Ascend AI chips. The firm aims to compete directly with NVIDIA’s Rubin lineup in the rack-scale segment.
Huawei’s Atlas 950 SuperPod Competes With NVIDIA’s Rubin NVL144 & NVL572 Configurations
Chinese AI company Huawei has been at the forefront of competing with NVIDIA in China’s AI market, particularly with rack-scale solutions. The firm recently announced its CloudMatrix 384 AI system, which reportedly surpassed NVIDIA’s Blackwell AI system a few months ago. Now, at Huawei Connect 2025, Huawei has introduced new SuperPod AI clusters: the Atlas 950 and the Atlas 960.
The Atlas 950 SuperPod is claimed to feature 8,192 of the Ascend 950 AI chips. It offers a cumulative performance of eight EFLOPS FP8 and 16 EFLOPS FP16 with an interconnect bandwidth of 16.3 PB/s. Based on these specifications, Huawei expects the SuperPod to be on par with NVIDIA’s NVL144 Vera Rubin AI rack.
Huawei plans to combine multiple Atlas 950 SuperPods into a larger system called the Atlas 950 SuperCluster, which will feature 524,288 of the Ascend 950 AI chips. Huawei claims that this is the world’s largest AI cluster in terms of onboard AI chips.
Huawei also announced the Atlas 950 SuperCluster and Atlas 960 SuperCluster, which can scale up to support between 500,000 and 1 million processors. These are the “largest AI compute clusters” in the world, according to the company.
Huawei has achieved such high performance by focusing on squeezing out immense computing power rather than efficiency or cost. With over 500K AI chips onboard, the power consumption is expected to be extremely high.
[Ad_2]
