The Huawei Connect conference featured the Atlas 960E SuperPoD as the centerpiece of Huawei’s new Peerium Computing Architecture, completely rebuilding AI infrastructure for massive scale including 4,096 NPUs that can expand into SuperClusters with over 500,000 NPUs. Ascend chips are advancing rapidly with specialized training and inference variants while Huawei strengthens its lead in interconnect through UnifiedBus and the new Hi-ONE optical engine, all analyzed through a digital AI transformation strategy lens. Memory pooling via OceanStor further expands KV cache capabilities across tiers, according to digital AI strategy consultant Jeffrey Towson.





