H3C Shifts AI Infrastructure Focus From More GPUs to Token Efficiency

H3C argued at the 2026 Apsara Conference that AI infrastructure competition is shifting from adding GPUs to maximizing useful Tokens per GPU. The company showcased its UniPoD S80000 SuperPod, supporting 32 to 1,024 GPUs and scaling to 16,384, alongside switches for Scale-Up, Scale-Out, and Scale-Across interconnects.

It also pitched its UniStor X20000 storage, which it says delivers up to 200GB/s bandwidth and cuts GPU waiting time by 30%.