Zayo NVIDIA fiber expansion adds 8,000 miles for distributed AI workloads

This partnership signals a shift in infrastructure strategy where networking is no longer an afterthought to power and land acquisition. By integrating NVIDIA Spectrum-XGS scale-across technology with physical fiber, providers can treat distant GPU clusters as a single unified computing environment. For the streaming and CDN ecosystem, this distributed architecture facilitates moving heavy inference workloads closer to the edge without sacrificing the processing power of centralized training hubs. As AI factories move into secondary markets to find electricity, the availability of high-capacity dark fiber becomes the primary bottleneck for scaling real-time video applications. Watch for whether neocloud providers prioritize these new Zayo corridors to accelerate their GPU capacity deployments.

Zayo's 8,000-route-mile fiber expansion is part of a broader industry push to connect distributed AI data centers with high-capacity optical links. In July 2025, Zayo announced a strategic partnership with NVIDIA to deploy Spectrum-XGS Ethernet networking across its long-haul routes, enabling geographically separated GPU clusters to operate as unified compute environments. The project targets corridors where power availability is attracting new data center construction, particularly in secondary markets outside traditional Northern Virginia and Dallas hubs. Zayo's existing network already spans more than 17 million fiber miles across North America and Europe, giving it one of the largest dark-fiber footprints among independent infrastructure providers.

The competitive landscape for AI-focused fiber infrastructure is intensifying. In early 2025, Lumen Technologies secured a multi-billion-dollar contract with Microsoft to build dedicated fiber routes connecting AI data centers, signaling that hyperscalers are willing to pay premium rates for guaranteed low-latency connectivity between GPU clusters. Crown Castle, which operates roughly 80,000 route miles of fiber in the U.S., has similarly positioned its small-cell and fiber assets as critical for edge AI inference workloads. These moves reflect a market where fiber availability, not just power, is becoming the binding constraint on AI infrastructure scaling.

On the technology side, NVIDIA's Spectrum-XGS platform is designed to extend its Spectrum-X Ethernet fabric across distances that previously required proprietary interconnects. The platform targets scale-across architectures where training and inference jobs span multiple sites separated by hundreds or thousands of kilometers. Corning and Coherent, both named as optical component suppliers in the Zayo buildout, have reported surging demand for high-density fiber optic cables rated for 800G and 1.6T per-wavelength transmission. Corning reported in its Q2 2025 earnings call that enterprise and data center optical revenue grew more than 30 percent year over year, driven by AI-related fiber deployments. Lumentum has similarly flagged rising orders for coherent optical transceivers used in long-haul AI interconnects, underscoring that the component supply chain is scaling alongside Zayo's physical build.