Zayo Reserves Corning Fiber Manufacturing Capacity Through 2030

Telecommunications infrastructure provider Zayo has secured manufacturing capacity with Corning to guarantee fiber supply for a massive network expansion running through 2030, a move that underscores growing concerns about supply constraints as AI workloads reshape data center connectivity requirements.

The agreement covers a material portion of the fiber Zayo needs for 15,000 route miles of new and overbuilt infrastructure, representing 20 million fiber miles across 28 routes. The expansion includes 10 entirely new routes and 18 overbuilds of existing paths, with 24 of the 28 routes tied to Zayo's separate agreement with Nvidia.

Why it matters

Fiber availability is emerging as a structural constraint alongside power in AI infrastructure buildout. By reserving manufacturing capacity years in advance, Zayo is protecting construction timelines for projects that take years to plan and execute. The move signals that telecommunications providers see fiber supply as a potential bottleneck as AI drives unprecedented demand for high-capacity, geographically diverse network routes between data centers.

From reactive ordering to capacity planning

Historically, Zayo ordered fiber as individual projects required it. The new approach with Corning represents a fundamental shift to collaborative capacity planning, according to Chief Network Officer Troy Lupe.

"We're reserving capacity with them," Lupe explained in an interview. "We're planning with them so that they're opening the manufacturing lines and buying the supplies they need to deliver all of this fiber."

While Zayo has not yet experienced material fiber shortages, Lupe said the company is seeing concerning signals about future supply pressure. "If you look at the amount of demand coming, the amount of historical demand needed, it takes time to build factories across the stack," he noted.

Corning has positioned itself as a critical supplier for AI network infrastructure, recently signing major optical supply agreements with Meta and Amazon and partnering with Nvidia to expand US optical manufacturing capacity. The company's Optical Communications business grew 36% year-over-year in the first quarter.

AI changes network scale, not fundamentals

AI workloads don't require fundamentally different fiber technology, but they're changing network design in significant ways. Jimmy Yu, vice president at Dell'Oro Group, said the primary difference is scale: "The amount of fiber per conduit will need to be in the hundreds, if not thousands."

AI networks also demand more diverse paths for resiliency, direct routes to minimize latency between data centers, and fewer add/drop points along routes. The traffic pattern itself differs because machines are communicating with other machines, transferring substantially more data than human-to-human or human-to-machine interactions.

Zayo is seeing demand for "triversity" and "quadversity"—three or four physically diverse routes between critical network endpoints—as AI operators seek protection against outages.

Construction timeline and risks

Zayo has delivered four routes this year and expects to complete another four by year-end. Most of the Nvidia-related build will occur over the following several years, with projects currently in various stages from engineering to environmental permitting.

Permitting and construction across multiple jurisdictions pose the biggest remaining execution risks, according to Lupe, even though Zayo has experience building in many of the territories involved.

The optical transport market is already reflecting AI-driven demand. Optical transport equipment revenue grew 15% year-over-year in the second quarter, while data center interconnect revenue from WDM systems and IPoDWDM optics surged 45%, according to Dell'Oro Group.

As AI workloads potentially shift from training toward inference, network requirements could evolve further. Personalized AI-generated content that can't be cached locally could materially increase traffic between users and compute facilities, particularly inside metropolitan areas.

These details were first reported by Data Center Knowledge.