At the Ascend AI Developer Summit on May 26, 2025, Huawei unveiled its new Ascend SuperNode architecture, the foundation of its CloudMatrix 384 high-performance AI compute cluster. Comprised of 384 Ascend 910C processors housed across 12 compute cabinets and 4 bus cabinets, the system delivers 300 petaflops of processing power and 48 terabytes of high-bandwidth memory—nearly double the BF16 compute throughput of Nvidia’s GB200 NVL72 system.