Huawei has unveiled new supernode computing cluster systems at its Connect 2025 conference in Shanghai, marking a significant step in its efforts to enhance AI computing power and compete with industry leaders Nvidia and xAI.
The company has introduced the Atlas 950 and Atlas 960 systems, which comprise both superpods and superclusters, designed to bolster its AI capabilities. The Atlas 950 superpod is set to feature 8,192 Ascend neural processing unit (NPU) cards and is slated for launch in the fourth quarter of 2026. Furthermore, by the end of 2027, Huawei plans to release the Atlas 960 superpod, which will boast up to 15,488 Ascend 960 NPU cards, significantly expanding its AI computing capacity.
The supercluster systems, including the Atlas 950 and Atlas 960, are also part of Huawei’s ambitious plans. The Atlas 950 supercluster is expected to be available by the end of 2026, incorporating over 500,000 Ascend NPUs. The Atlas 960 supercluster will take this a step further by containing more than 1 million Ascend NPUs, underscoring Huawei’s commitment to advancing its AI infrastructure.
According to Huawei, the development of these systems is aimed at addressing the limitations associated with chip manufacturing and fostering domestic AI development. By introducing these cutting-edge computing cluster systems, Huawei is positioning itself to play a more significant role in the evolving AI landscape.
