Huawei releases super node products overseas for the first time to break through the bottleneck of computing power with system-level innovation.

On February 28th, local time in Barcelona, Spain, during MWC26 Barcelona, Huawei for the first time showcased overseas its latest super node products created based on the new Inter-Switch Protocol "Networking Quo", including the AI super node Atlas 950 SuperPoD and the industry's first arithmetic super node TaiShan 950 SuperPoD. With the advent of the trillion-parameter era for AI large models, traditional server stacking faces the dilemma of decreasing computing power utilization and frequent training interruptions after cluster scale expansion. Huawei has introduced the Networking Quo Inter-Switch Protocol through the "cluster + super node" system-level architecture innovation, achieving interconnection of up to 8192 cards within the super node, featuring ultra-high bandwidth, ultra-low latency, and unified memory addressing, making the entire cluster efficient in learning and thinking like a single computer in logic. It is reported that products such as Atlas 950 SuperPoD and Atlas 850E can meet diverse scenarios of AI training and inference; at the same time, the new generation of arithmetic servers TaiShan 950 SuperPoD, TaiShan 500, and 200 series provide customers with high, medium, and low-grade computing power options, dedicated to providing solid computing power new options for the world.