Huawei presented new artificial intelligence computing systems at its Connect 2025 conference that target direct competition with NVIDIA hardware. The Chinese technology company revealed Atlas 950 SuperPoD clusters equipped with 8,192 Ascend 950 processors. These systems deliver eight exaFLOPS FP8 performance and 16 exaFLOPS FP16 computational capacity alongside 16.3 petabytes per second interconnect speeds. The company positions these specifications against NVIDIA's Rubin NVL144 rack configurations.
Huawei expanded its announcement to describe Atlas 950 SuperClusters containing 524,288 Ascend processors. The manufacturer claims these represent the world's largest dedicated AI computing arrays by chip count. Both Atlas 950 and Atlas 960 SuperCluster variants support between 500,000 and one million processors. The company prioritizes raw computational output over power efficiency considerations. These developments reflect Huawei's strategy to minimize Chinese dependence on Western semiconductor technologies while meeting domestic artificial intelligence processing demands.
Huawei expanded its announcement to describe Atlas 950 SuperClusters containing 524,288 Ascend processors. The manufacturer claims these represent the world's largest dedicated AI computing arrays by chip count. Both Atlas 950 and Atlas 960 SuperCluster variants support between 500,000 and one million processors. The company prioritizes raw computational output over power efficiency considerations. These developments reflect Huawei's strategy to minimize Chinese dependence on Western semiconductor technologies while meeting domestic artificial intelligence processing demands.
