H3C at Apsara 2026: AI infrastructure race shifts to Token efficiency

H3C unveiled its AI infrastructure lineup at Apsara Conference 2026, including the UniPoD S80000 SuperPod scaling from 32 to 1,024 GPUs and up to 16,384, Scale-Up/Scale-Out/Scale-Across switches, and UniStor X20000 storage. The company argues that adding more GPUs does not yield proportional performance gains, pushing system-level optimization instead.

• UniPoD S80000 supports 32 to 1,024 GPUs, scaling to 16,384
• S9828-128EO 102.4T switch cuts latency by 15% via co-packaged optics
• UniStor X20836 delivers up to 200GB/s and 3 million IOPS per node
• XCache cuts time-to-first-token latency by up to 90%