At MWC Barcelona 2026, Huawei has debuted its latest SuperPoD product Atlas 950 SuperPoD, TaiShan 950 SuperPoD and a series of computing solutions to the global market.
With AI technologies evolving rapidly and models now using trillions of parameters, agentic AI is beginning to penetrate into core production processes in many industries. This is driving up demand for larger computing scale and lower latency.
However, these massive models are beyond the reach of conventional horizontal scaling; larger clusters often suffer from lower utilization and frequent training interruptions.
Huawei has tackled these challenges with its UnifiedBus interconnect for SuperPoDs. The groundbreaking “cluster + SuperPoD” system architecture is tailormade for growing computing demands and driving AI progress.
At MWC, Huawei debuted its latest SuperPoD offerings on a global arena, including the Atlas 950 SuperPoD and Atlas 850E. Built on UnifiedBus, these products are fit for a diverse range of AI training and inference scenarios.
The Atlas 950 SuperPoD, for instance, connects up to 8,192 NPUs via UnifiedBus, delivering ultra-high bandwidth, ultra-low latency, and unified memory addressing. It operates as a single, logical computer for learning, reasoning, and processing.
Huawei also exhibited TaiShan 950 SuperPoD – the industry’s very first general-purpose computing SuperPoD – alongside next-generation servers like the TaiShan 500 and TaiShan 200. These provide flexible computing options for computing workloads on a scale of high to low intensity.
Huawei has fully open-sourced its CANN heterogeneous compute architecture. Through layered decoupling, all software components – from operator libraries, acceleration libraries, and graph computing to programming languages – are openly available for developers. CANN also supports open source communities and projects typified by Triton, TileLang, PyTorch, vLLM, and verl, which tangibly facilitates developers in terms of accessibility and efficiency.