Huawei unveiled the Atlas 960E SuperPoD, an AI computing system for increasingly large AI models, on September 17 at HUAWEI CONNECT 2026 in Shanghai. Huawei says it is the industry’s first SuperPoD based on Near-Packaged Optics (NPO) and supports training and inference for AI models approaching 10 trillion parameters. The system combines UnifiedBus, also known as Lingqu, with Huawei’s Hi-ONE optical system and uses unified memory addressing. A single SuperPoD can support up to 4,096 Ascend AI processors and deliver up to 8 EFLOPS of FP8 performance and 16 EFLOPS of FP4 performance, with a fully liquid-cooled design and an orthogonal architecture. Huawei says the Hi-ONE NPO optical engine offers 7.2 terabits per second of transmission capacity and includes an integrated light source. The Atlas 960E uses about 5,500 Hi-ONE units, replacing roughly 48,000 traditional 800G optical modules. Huawei claims the design can reduce power consumption by more than 550 kilowatts, double fault-free operating time and provide 99.8% system availability. Huawei also plans to connect multiple Atlas 960E SuperPoDs into larger AI clusters using UnifiedBus and RoCE networking, saying its architecture can connect up to 512,000 AI processors, while a multi-rail topology can scale an Ascend SuperCluster to