Zayo Group and NVIDIA are collaborating on a long-haul network infrastructure program designed to carry traffic between distributed AI compute sites. The partnership, reported by Data Center Frontier, targets the transport layer that connects data centers running large-scale AI training and inference workloads.

The announcement places Zayo, one of the largest independent fiber providers in North America, directly in the supply chain for distributed AI infrastructure. NVIDIA supplies the dominant compute platforms used for AI model training, and workloads of this type increasingly span multiple facilities. That shift creates demand for high-capacity, low-latency links between sites rather than only within them.

The project's framing as a "backbone" indicates point-to-point long-haul routes between data center campuses, as opposed to metro fiber or in-building interconnects. Zayo operates an extensive long-haul fiber network across North America, which gives the partnership an existing physical footprint to build on. NVIDIA has previously positioned networking — through its InfiniBand and Ethernet product lines — as a critical bottleneck for AI clusters whose GPUs sit in different locations.

For data center operators, the arrangement signals that distributed AI training, where a single model is split across multiple sites, is moving from concept to deployment. This architecture requires transport capacity well beyond what conventional internet backbone traffic demands, since synchronized training traffic flows continuously between locations.

The collaboration also illustrates a broader pattern in the sector: compute vendors are partnering directly with fiber operators to guarantee the network capacity their hardware needs. Latency, capacity and route diversity become limiting factors when GPUs in separate facilities must behave as a single cluster.

Zayo has spent recent years expanding and lighting new long-haul routes across the United States, a build-out that aligns with hyperscaler and AI-driven demand for inter-facility bandwidth. Financial terms of the agreement with NVIDIA were not disclosed.

Data Center Frontier reports the project under the headline "Zayo, NVIDIA Build the Long-Haul Backbone for Distributed AI." Further technical specifications — route counts, capacity targets, deployment timelines — had not been published at the time of the report.