Huawei Technologies unveiled its next-generation Ascend 960 SuperPoD artificial intelligence computing cluster during the Huawei Connect conference in Shanghai.
The newly announced supercomputing node integrates proprietary near-packaged optics and unified fabric technology to connect up to 4,096 processor cards with minimal data latency. Company leadership also announced an expedited hardware schedule, moving the commercial debut of the Ascend 960 accelerator series forward to the first quarter of 2027. The architecture is designed to train massive foundation models exceeding ten trillion parameters by clustering multiple nodes into mega-systems spanning hundreds of thousands of processors.
By emphasizing systems-level clustering and high-speed optical interconnects, the Chinese telecommunications giant aims to circumvent strict Western trade restrictions on advanced semiconductor fabrication. The release highlights China's concerted push to establish autonomous artificial intelligence hardware infrastructure capable of competing directly with industry leader Nvidia.