Artificial intelligence is driving extraordinary demand for compute, but moving data among AI factories, clouds, enterprise locations, and edge environments is becoming equally important. Zayo’s 2026 Bandwidth Report, based on purchasing patterns across nearly 6,000 customers, shows this transition is already underway. Demand is rising not only for 400-gigabit wavelengths, but also for long-haul and metro dark fiber, highlighting that AI is beginning to reshape network architectures and capacity plans.
In a recent CUBE interview, Bill Long, chief product and strategy officer at Zayo, identified three forces arriving simultaneously: U.S. reindustrialization, continued economic digitization and the rapid buildout of AI infrastructure. What does this mean for enterprise leaders? Primarily, it indicates that network planning must go beyond incremental upgrades to encompass scale, security, resiliency, and ecosystem connectivity. Watch the full interview below
Three Waves of Demand Converge
Zayo’s report indicates that bandwidth growth is broad-based. Manufacturing, defense and the public sector are expanding as workloads and supply chains move closer to home. Meanwhile, warehouses, factories, healthcare facilities and retail operations increasingly rely on robotics, imaging, analytics and cloud applications.
Long illustrated the magnitude of that shift with the example of a major retailer’s warehouse. A facility that once operated with a 10-megabit connection now requires dual 100-gigabit links to support approximately $100 million in robotics. This is not simply more employee traffic; it represents a transition toward machine-generated, operationally critical data.
AI adds a third wave and will likely accelerate the other two. As Long explained, “It’s really the convergence of those three trends all coming together at the same time.” AI applications will deepen the digital requirements of virtually every industry.
AI Is Changing the Scale of Wide-Area Networking
The report’s capacity findings are striking. For the second consecutive year, 400-gigabit services represented roughly half of all wavelength capacity purchased. Long-haul dark fiber demand doubled, new fiber sales tripled, and metro dark fiber demand was significantly higher in leading AI markets.
Perhaps more revealing is the change in order size. Long said a large long-haul order once consisted of 8, 10, or 12 fibers; Zayo is now seeing orders for as many as 432. The company had expected some fiber investments to provide inventory for a decade, only to see that capacity absorbed within 12 to 18 months.
Distributed AI training is one reason. Power constraints can prevent organizations from housing all required computing infrastructure on one campus, forcing them to connect multiple large sites with enormous bandwidth and predictable performance. Zayo also reported turning up 20 terabits of capacity for one AI lab in 30 days using multiple 400-gigabit connections.
Most enterprises will not need hundreds of fibers or terabits of capacity. However, demand from hyperscalers, neoclouds and AI labs can tighten supply, extend delivery timelines and influence where enterprises can obtain high-capacity services.
Connecting the AI Ecosystem
Zayo’s collaboration with NVIDIA highlights the importance of planning network infrastructure as an ecosystem. The companies examined where AI data centers and available power were likely to be located, then identified network routes requiring additional capacity. Zayo is building or upgrading more than 8,000 route miles across 24 routes, with enough capacity intended to serve NVIDIA and the broader market.
Long summarized the risk plainly: “You’re spending tens of billions of dollars to put data centers everywhere. They’re just going to be really expensive refrigerators if we don’t figure out how to get them connected.”
Zayo subsequently secured fiber supply through Corning and coordinated with optical equipment providers. AI infrastructure requires synchronized investment across chips, power, data centers, fiber, optics and interconnection; a bottleneck in any layer can constrain the entire system.
Metro Networks Take on a New Role
AI is also changing where traffic originates and how it flows. Historically, metro networks were designed primarily around concentrations of people and office buildings. Increasingly, bandwidth demand comes from machines in factories, fulfillment centers, imaging facilities, and edge locations.
These applications can produce more balanced upstream and downstream traffic. Robots continuously exchange information, while personalized AI content may need to reach an inference platform rather than a nearby cache. This places new pressure on metro connectivity and helps explain Zayo’s acquisition of Crown Castle Fiber assets in several key markets.
For enterprises, the metro network is becoming an interconnected extension of the data center. Low latency, route diversity, and access to clouds, colocation facilities, and AI service providers will matter as much as raw capacity.
What Enterprise Leaders Should Do Now
Long compares the architectural shift created by AI to the early evolution of cloud, but at much greater scale. Enterprise data, AI models, and users or agents may all reside in different environments. The resulting infrastructure is distributed, hybrid, and dependent on multiple providers.
Enterprises should begin by mapping AI applications to four requirements: capacity, performance, security, and connectivity. Leaders need to understand where relevant data resides, where training or inference occurs, who or what consumes the output, and which dependencies could disrupt that path. Private connectivity may also become increasingly important as AI moves from experimentation into production and begins using more sensitive corporate data.
Long suggested enterprises consider a three- to four-year planning horizon while prioritizing agility. He recommended evaluating applications individually and asking whether the network can support their future scale, security, and performance requirements.
Why It Matters
The central takeaway from Zayo’s report is that AI readiness requires more than GPUs and data-center capacity. Enterprises also need the reach, resiliency, security, and operational flexibility to move rapidly growing volumes of data among increasingly distributed locations.
The bandwidth transition will not affect every organization at the same pace. However, the purchasing data indicates that the market has already begun to move. Enterprise leaders should assess network readiness before capacity constraints, construction timelines, or architectural complexity become barriers to AI adoption. The question is no longer whether AI will increase network demand, but whether enterprise infrastructure plans are prepared for what comes next.
For more information on Zayo’s 2026 Bandwidth Report, please visit the Zayo website.
