Huawei Launches AI Supercomputing Systems to Rival Nvidia
Huawei has announced a major expansion of its AI computing lineup, unveiling new systems powered by its in-house Ascend chips. At its annual Huawei Connect event in Shanghai, the company introduced the Atlas 950 SuperCluster, set to launch next year, which will use more than 500,000 chips to deliver massive computing power despite U.S. sanctions limiting access to advanced semiconductors.
The design builds on Huawei’s strategy of linking thousands of less powerful chips into supernodes, superpods, and ultimately superclusters. The upcoming Atlas 950 supernode will support 8,192 Ascend chips, while a future Atlas 960 model, expected in 2027, could scale up to 15,488 chips per node, pushing full cluster capacity beyond one million chips. Huawei claims this system will outperform Nvidia’s upcoming NVL144 and even Elon Musk’s xAI Colossus supercomputer.
Vice chairman Eric Xu argued the Atlas 950 will deliver 6.7 times the performance of Nvidia’s system slated for next year, and “stay ahead on all fronts” through 2027. Research firm SemiAnalysis has previously noted that Huawei’s CloudMatrix already rivaled Nvidia by multiplying chip count, even though each Ascend chip has only about one-third the power of an Nvidia GPU.
The company also confirmed plans for three new Ascend chip generations through 2028, promising to double computing power with each release. Huawei has already deployed more than 300 Atlas 900 A3 supernodes to 20+ customers in telecom, manufacturing, and other sectors.
Huawei’s announcement comes amid growing China–U.S. chip tensions. This week, China extended an antitrust probe into Nvidia and reportedly ordered tech firms to halt purchases of its RTX Pro 6000D, sending Nvidia’s stock down more than 2%. Nvidia CEO Jensen Huang called the reports “disappointing” while acknowledging Huawei as a “formidable” rival.
