The Fiber Race Behind America’s AI Buildout

For years, fiber-optic networks were treated as the quiet plumbing of the internet. That is changing. As artificial-intelligence companies build larger and more geographically distributed computing systems, access to high-capacity fiber is becoming a strategic infrastructure decision.

The shift is visible in the market for dark fiber, the unused strands in optical networks that companies can lease or operate for their own traffic. Zayo, a major U.S. network operator, reported that long-haul dark-fiber demand doubled between 2024 and 2025, while new fiber sales tripled. Hyperscalers and carriers accounted for about 95% of its long-haul purchases.

The underlying issue is not simply that companies need faster internet. AI workloads can require enormous volumes of data to move between computing facilities, and interruptions or congestion can make expensive computing capacity less productive. For companies investing billions of dollars in data centers, the network connecting those facilities increasingly matters almost as much as the facilities themselves.

Why AI changes the fiber equation

Traditional cloud computing already created substantial demand for fiber. But AI introduces a different traffic pattern.

Training advanced models can involve large computing clusters operating simultaneously. Data must move between processors, storage systems and facilities with very high bandwidth and low latency. As companies distribute computing across multiple sites, those sites need reliable connections to function as part of a broader system.

That makes network capacity a potential bottleneck.

A data-center developer can build additional computing capacity relatively quickly compared with the time required to secure rights-of-way, construct long-distance fiber routes or obtain suitable connections. Existing routes therefore become strategically valuable, particularly where new infrastructure is difficult to build.

Research published by the Fiber Broadband Association in 2025 estimated that the U.S. could require 2.3 times more fiber to support the performance, scalability and security requirements associated with AI and data-center growth. The organization also projected at least a tripling of hyperscale data-center capacity by 2029.

The exact scale of future demand remains uncertain, but the direction is clear: computing growth is increasing the value of the physical networks underneath it.

Companies are buying certainty

One reason companies are moving early is that fiber is not easily added at the last minute.

A business can lease capacity on an existing network, purchase dedicated strands or participate in new network construction. Dark fiber is particularly attractive to large customers because it can provide greater control over capacity, routing and network configuration.

The economic calculation is similar to securing electricity or land before constructing a data center. Paying for capacity in advance can appear expensive, but the cost of being unable to connect a newly built facility may be greater.

Recent transactions illustrate the strategy.

In January, Meta announced a multiyear agreement worth up to $6 billion with Corning for optical fiber, cable and connectivity products used in its U.S. data-center infrastructure. Corning said the agreement would support expanded manufacturing capacity in North Carolina.

Verizon has also positioned fiber as an AI infrastructure business. In July, the company disclosed a deal worth more than $1 billion to connect Google data centers using Verizon's dark-fiber routes. Verizon's expanded fiber footprint, following its acquisition of Frontier Communications, gives it a larger network from which to pursue similar opportunities.

Uniti Wholesale, meanwhile, announced a 1,100-mile expansion of its dark-fiber network across the south-central U.S., backed by a 20-year customer agreement valued at more than $500 million. The routes are intended to connect major markets with emerging AI and data-center locations.

These deals show that the market is evolving from simply selling telecommunications capacity toward long-term infrastructure partnerships.

The value of location and redundancy

Fiber demand is also being shaped by geography.

AI developers are increasingly looking beyond established technology centers because electricity, land and permitting can be easier to obtain elsewhere. That creates a connectivity challenge: new data-center clusters must be connected to established networks and, in many cases, to one another.

The shortest route is not necessarily the best route. Companies may want multiple paths so that a single fiber cut, construction accident or network failure does not isolate a facility.

That makes route diversity economically valuable.

A network with several independent paths can cost more to build, but the additional expense may be justified when computing infrastructure represents billions of dollars of capital. In effect, fiber becomes a form of operational insurance.

This is particularly important as AI infrastructure becomes more distributed. The economics of a large computing campus increasingly depend on the reliability of the ecosystem surrounding it, including power, cooling, land and communications.

Telecom companies see a new opportunity

For telecommunications operators, AI offers a potential growth market at a time when traditional connectivity businesses face mature consumer markets and heavy capital requirements.

PwC reported that U.S. telecom deal activity accelerated in early 2026, with fiber consolidation and AI-related demand among the forces shaping the sector. Regional operators and network owners have become attractive acquisition targets because their existing routes can be difficult and expensive to replicate.

That could encourage further consolidation. It could also make previously underused fiber assets more valuable as AI customers seek additional capacity.

But the opportunity has limits. Fiber is still capital-intensive infrastructure. Demand projections depend on the pace of AI adoption, data-center construction and the evolution of networking technology. More efficient optical equipment can increase the amount of information carried over existing fiber, potentially reducing the need for entirely new routes.

There is also competition between different infrastructure priorities. Fiber operators must balance hyperscaler demand with broadband expansion, enterprise connectivity and other customers.

What comes next

The next stage of the fiber market will depend on how quickly AI infrastructure spreads geographically and how much traffic these systems actually generate.

If AI workloads continue moving toward larger and more distributed computing clusters, dedicated fiber and network redundancy are likely to become increasingly important. If computing becomes substantially more efficient or workloads become more localized, some expected capacity requirements could prove excessive.

The geography of the U.S. power system will also matter. Data centers are increasingly being built where electricity can be secured, and those locations are not always where existing high-capacity networks are strongest. Fiber companies therefore have an incentive to follow the data-center map rather than simply expand where traditional telecom demand is greatest.

The broader lesson is that AI is increasing the economic value of infrastructure that was largely invisible to end users. Chips determine how much computing can be performed; data centers determine where it happens; fiber determines how effectively those systems can communicate.

As companies compete to secure computing capacity, they are discovering that control over the connections between machines can be nearly as important as control over the machines themselves.