Nvidia backs an 8,000-mile fibre build for AI
Zayo will construct new long-haul routes with Nvidia as an anchor customer to ease AI interconnect bottlenecks across North America.
Nvidia is backing Zayo’s plan to build more than 8,000 miles of new long-haul fibre across North America, with the chipmaker becoming an anchor customer for the infrastructure project.
The expansion is designed to support the growing demand for high-capacity connections between data centres as AI workloads become larger and more distributed. Zayo will build and operate the network, while making additional capacity available to AI developers, cloud providers, neoclouds and enterprises.
Why AI needs more fibre
Modern AI systems depend on massive computing clusters, but those clusters also need to exchange huge amounts of data. As AI infrastructure spreads across different regions, the networks connecting data centres are becoming just as important as the computing hardware inside them.
Long-haul fibre refers to high-capacity optical networks that connect cities and regions. Zayo plans to build six new routes and expand capacity across 10 high-demand markets. The company expects its AI-focused network in North America to exceed 15,000 route miles following the expansion.
The need is being driven partly by the changing structure of AI infrastructure. Large training clusters may be located in areas with abundant power, while inference, the process of running trained models to generate responses, often needs to happen closer to users. That makes fast connections between data centres increasingly important.
Building new routes will take time
The fibre expansion also highlights a less visible challenge in the AI infrastructure race: physical construction.
New long-haul fibre routes can take more than a year to build and, in some cases, approach two years.
That timeline is comparable to the construction period for some large data centres, meaning network capacity needs to be planned well ahead of demand. Upgrading existing routes is not always a quick solution either.
The industry is facing constraints in optical equipment, including components such as pump lasers and coherent transponders used to transmit data across fibre networks. Extended equipment lead times can make additional physical routes an attractive way to increase capacity.
Nvidia wants AI infrastructure to stay connected
Nvidia's involvement as an anchor customer gives Zayo greater visibility into demand as it develops the new routes. The company will retain ownership and operate the network, while selling unused capacity to other customers.
For Nvidia, the investment reflects how its infrastructure strategy extends beyond GPUs. AI factories require computing, power, cooling, networking and connectivity to operate as a coordinated system.
The wider AI industry could also benefit if the new routes reduce congestion and improve resilience between major data centre markets.
The expansion points to a broader reality in the AI infrastructure race. Building more powerful GPUs is only part of the challenge. Those systems also need reliable, high-capacity networks to communicate with each other.
As AI workloads continue to scale, fibre could become just as critical to the infrastructure buildout as chips, servers and electricity.