Distributed AI Training Pushes Datacenter-to-Datacenter Networks Toward Much Higher Bandwidth

Large-scale AI training is increasingly spread across multiple datacenters, with Google, Microsoft, AWS, Meta, and CoreWeave cited as examples.

Because synchronized GPU clusters exchange data in bursts, the links between sites can become a bottleneck. Cisco estimates such inter-site networks may need aggregate bandwidth about 14 times that of a conventional datacenter interconnect (DCI) baseline. The estimate is Cisco's own projection as reported by The Next Platform, not a measured outcome from any named operator.