Summary: This Hewlett Packard Enterprise glossary article defines AI data center networking as the fabric that supports rigorous performance, scalability and low-latency needs of training and inference workloads. It contrasts traditional InfiniBand with open Ethernet approaches, and describes architectural elements such as non-blocking fabric topologies, congestion management, and automation strategies that maximize GPU utilization, improve job execution times, and reduce network-induced bottlenecks in large AI clusters. 
Source: https://www.hpe.com/uk/en/what-is/ai-data-center-networking.html