Huawei details silicon push for agentic AI infrastructure

Huawei opened HUAWEI CONNECT 2026 in Shanghai with a strategy centered on AI infrastructure, with systems and architectures built around SuperPoDs and SuperClusters. David Wang, Deputy Chairman of the Board and Rotating Chairman at Huawei, said foundation model parameters are rapidly approaching 10 trillion and are projected to exceed 100 trillion by 2030, while China’s daily inference token consumption has risen to around 500 trillion.

The company introduced the Atlas 960E SuperPoD, described as the industry’s first NPO-based SuperPoD. A single system can scale up to 4,096 NPUs, deliver 8 EFLOPS of FP8 compute performance and provide up to 1 petabyte of HBM capacity. Huawei said the use of 5,500 Hi-ONE units removes the need for 48,000 800G optical modules, cutting power consumption by over 550 kilowatts and achieving 99.8% system availability.

Huawei also upgraded the TaiShan 950 SuperPoD and launched OceanStor M900 for context memory storage, combining them with UnifiedBus to support agentic SuperClusters that can scale to one million NPUs. The company highlighted its open ecosystem push, including 4.16 million Kunpeng developers, CANN’s community-driven open-source development, Ascend support across over 90 third-party open-source projects, and official PyTorch accelerator backend support.