Connectivity - The Networks Carrying Intelligence

Power determines whether AI can run. Connectivity determines how far and how fast it can travel.

Power gets an AI campus online. Connectivity connects it to the economy.

Every time someone opens an AI application, uploads a document or enters a prompt, information begins moving across a physical network. It travels from a device to a carrier, through fiber routes and network exchanges, into a data center and across the high-speed systems connecting processors inside it.

The response then makes the journey back.

All of this may happen in seconds, creating the illusion that AI simply lives in “the cloud.” But the cloud has cables, buildings, routes, access points and geography.

AI doesn’t float above the physical world. It travels through it.

That makes connectivity more than another component of AI infrastructure. It is the distribution system of the Intelligence Economy the infrastructure that allows intelligence to move between processors, facilities, companies, communities and people.

The Signal

AI is creating a networking challenge alongside the computing challenge.

Training a large model requires enormous volumes of information to move between processors. Operating that model requires data to travel between storage systems, cloud platforms, data centers and users. As AI systems become larger and more widely used, the amount of information moving across these networks increases with them.

A global survey of more than 1,300 data-center decision-makers found that organizations expect their data-center interconnection bandwidth requirements to grow by at least six times over five years. Fifty-three percent said AI workloads would be the largest driver of demand over the next two to three years, while 87% believed they would need connections supporting 800 gigabits per second or more per wavelength. Ciena

The message is straightforward:

AI is not only a compute problem. It is a data-movement problem.

The fastest processors in the world are less useful when the networks connecting them cannot keep up.

What Happens After You Press Enter?

Consider what happens when someone sends a prompt to an AI application.

The request first travels across a local or wireless connection. From there, it enters a regional carrier network and moves toward the cloud platform or data center hosting the application.

Inside that facility, the request may need to reach storage systems, databases and groups of processors working together. Those processors exchange information, generate a response and send the result back across the network.

One prompt can involve several different networks, multiple physical locations and thousands of individual computing components.

For simple consumer applications, a slight delay may be inconvenient. For autonomous systems, industrial operations, financial platforms, healthcare applications or critical infrastructure, delays and network interruptions can have much larger consequences.

That is why connectivity readiness cannot be measured by asking whether fiber is nearby.

The more important question is whether the infrastructure can move intelligence at the required speed, scale, security and reliability.

The Three Networks Behind AI

The connectivity supporting AI can be understood as three interconnected systems.

1. The Compute Network

The first network exists inside the AI data center.

It connects processors, servers and storage systems so they can operate as one coordinated computing environment. During model training, processors constantly exchange information. If those connections are slow or congested, expensive computing equipment can spend valuable time waiting for data.

This is often called east-west traffic: information moving between systems within a computing environment rather than directly toward an outside user.

The compute network is what turns thousands of individual processors into an AI factory.

2. The Data-Center Network

The second network connects one facility to another.

AI workloads are rarely contained inside a single building. Data may be stored in one location, processed in another and delivered through several cloud regions. Companies may also distribute workloads across multiple facilities for additional capacity, security and resilience.

These connections rely on long-haul fiber, carrier networks, cloud on-ramps, internet exchanges and dedicated data-center interconnection systems.

If the compute network connects the machines, the data-center network connects the factories.

3. The Delivery Network

The third network carries intelligence to the people and organizations using it.

This includes metro fiber, broadband networks, wireless towers, subsea cables and edge infrastructure located closer to end users. Its performance determines how reliably an AI service can reach businesses, homes, factories, hospitals and public institutions.

The delivery network is where AI infrastructure becomes an economic service.

Without it, intelligence may exist—but it cannot reach the market.

The Fiber Illusion

A line on a fiber map can create a false sense of readiness.

A route may pass near a development site without having available capacity. The fiber could belong to a carrier that does not serve the project. It might be dark, privately controlled or technically difficult to access. Two apparent connections may even share the same trench, creating a single point of failure disguised as redundancy.

“Fiber nearby” does not necessarily mean “connectivity available.”

A serious connectivity assessment must examine:

• Whether the fiber is lit or dark

• How much capacity is available

• Which carriers control or operate the route

• Whether additional strands can be leased

• How quickly capacity can be expanded

• Where the nearest interconnection points are located

• Whether backup routes are physically diverse

• What latency the network can provide to major markets

The most valuable route is not always the closest one. It is the route that offers capacity, access, resilience and a credible upgrade path.

Latency Changes the Map

Distance still matters in the digital economy.

Data may travel extremely quickly through fiber, but every additional mile, connection and processing point adds time. That delay is known as latency.

For email or document storage, a small delay might go unnoticed. For AI inference, robotics, industrial automation and real-time decision systems, milliseconds can influence performance.

This gives certain locations a structural advantage.

Regions near dense fiber routes, cloud on-ramps, carrier hotels and internet exchange points can connect to more networks with fewer intermediate steps. Edge data centers can place computing capacity closer to users, reducing the distance information must travel.

The cloud may appear locationless from a screen, but its performance is shaped by physical geography.

Redundancy Is Part of Capacity

Network capacity receives most of the attention. Network resilience deserves just as much.

An AI facility connected through one high-capacity route remains exposed to construction damage, equipment failure, severe weather or a regional outage. A second connection improves resilience only if it follows a genuinely different physical path.

Two cables inside the same trench are not two independent routes.

Strong network design may require different carriers, separate rights-of-way, multiple interconnection points and alternative paths into and out of a facility. For the largest AI campuses, connectivity planning increasingly resembles utility planning: the question is not merely how much service is available, but how the system performs when something fails.

Reliability is not separate from connectivity readiness.

It is part of it.

The Market Is Responding

The scale of recent agreements shows how quickly fiber is becoming part of the AI supply chain.

Microsoft selected Lumen to provide dedicated fiber, new network routes and expanded capacity supporting growing AI data-center demand. Lumen later said AI-related demand had contributed to $5 billion in new business, much of it involving custom networks and dark fiber. Microsoft and Lumen Lumen

Meta is approaching the problem on a global scale. Its Project Waterworth subsea cable is planned to stretch more than 50,000 kilometers across five continents, creating new routes for digital services and future AI workloads. Meta

Fiber manufacturing is expanding as well. In 2026, Meta announced a multiyear agreement with Corning valued at up to $6 billion for optical fiber, cable and connectivity products supporting its US data-center expansion. Amazon followed with another multiyear, multibillion-dollar agreement intended to expand domestic fiber-optic manufacturing capacity. Corning and Meta Amazon and Corning

These are not ordinary technology purchases.

They are long-term commitments to the physical pathways through which AI will operate.

Following Fiber

Data-center announcements usually receive the headlines. Fiber activity can provide an earlier signal.

New long-haul routes, dark-fiber leases, interconnection expansions, cable orders and manufacturing investments can reveal where companies expect future traffic to grow. The activity may begin before the public sees a finished campus—or even before a major project is formally announced.

This is the idea behind Following Fiber™.

Power tells us where AI can operate. Fiber can help tell us where it is preparing to scale.

Regions where power infrastructure and connectivity routes converge may become natural locations for AI campuses, cloud regions and industrial technology clusters. As more infrastructure gathers around those locations, they can develop into Intelligence Corridors™ connecting multiple communities and economic centers.

Connectivity creates its own form of economic gravity.

Once a region becomes easier to reach, it becomes easier to invest in.

Company Watch: Corning

Everyone is watching silicon. I’m watching glass.

Corning is not usually described as an AI company, yet its optical fiber and connectivity products sit inside the infrastructure supporting AI data centers and the networks connecting them.

Its recent agreements with Meta and Amazon are noteworthy because they extend beyond short-term product orders. They involve multiyear manufacturing capacity, new facilities, workforce development and domestic supply-chain expansion.

This is what the AI buildout looks like below the software layer.

The companies positioned to benefit may include not only chip designers and cloud platforms, but fiber manufacturers, network-equipment providers, carriers, interconnection operators and data-center companies.

That puts companies such as Ciena, Lumen, Zayo, Cisco, Nokia, Equinix, Digital Realty and CoreSite on the connectivity watchlist—not because every company will capture the same opportunity, but because each controls part of the system through which intelligence moves.

The Twin Infrastructure Principle

Power and connectivity are often evaluated separately.

AI projects experience them together.

A location with abundant power but limited fiber may struggle to move data, connect facilities or serve distant users. A location with major fiber routes but inadequate electricity may be unable to support the computing capacity those networks were built to reach.

Neither resource creates AI readiness on its own.

Power + Connectivity = AI Readiness

When these systems align, land becomes more useful, development timelines become more credible and investment becomes easier to justify. When they do not, even an otherwise attractive site can remain stranded.

This is the Twin Infrastructure Principle: the Intelligence Economy grows where energy and information can move together.

Connectivity Across the Eight Pillars

Connectivity does not operate in isolation.

It influences the value of land because sites near scalable, diverse fiber routes are more useful to data-center developers. It shapes capital decisions because network limitations can delay revenue or require expensive route construction. It depends on policy because rights-of-way, permitting and infrastructure security determine how quickly networks can expand.

It also creates workforce demand for fiber technicians, network engineers, construction crews and cybersecurity specialists.

Connectivity can even provide more flexibility around power, water and sustainability. When facilities are linked by high-capacity networks, workloads can be distributed among locations with different energy availability, cooling conditions and resource constraints.

The eight pillars of the Infrastructure of Intelligence™ are separate systems, but they function as one.

A weakness in the network can limit the entire stack.

What Leaders Should Start Asking

Communities and developers should move beyond asking whether a property has access to fiber.

What they should ask:

• Where do the available routes physically run?

• Who owns and operates them?

• How much capacity is available today?

• What would be required to expand it?

• Are there multiple carriers?

• Are the backup routes genuinely diverse?

• Where are the nearest cloud on-ramps and internet exchanges?

• What latency can the network provide to priority markets?

• Can nearby facilities be interconnected directly?

• Who is responsible for permitting new routes?

• What are the major physical and cybersecurity risks?

• What investments are carriers planning over the next five years?

These questions separate a marketable location from an infrastructure-ready one.

The Golden Nugget

Fiber on a map is not the same as connectivity a customer can buy.

The real asset is usable capacity: available at the right location, accessible through the right providers, connected to the right markets and protected by genuine redundancy.

That distinction will become increasingly important as more communities attempt to position themselves for AI investment.

The IOI Take

Connectivity is the circulatory system of the Intelligence Economy.

Processors create intelligence. Power keeps those processors operating. Networks allow the resulting intelligence to move between machines, facilities, markets and people.

This is why the next generation of AI infrastructure will not be built as a collection of isolated data centers. It will be built as a connected system of compute clusters, fiber corridors, interconnection hubs, cloud regions and edge facilities.

The strongest regions will be the ones that understand both sides of the equation.

They will know where the power is.

And they will know where the intelligence can go.

One Last Thought

Networks are easy to overlook because their best work is invisible.

When connectivity performs well, users rarely think about the fiber beneath the street, the network exchange across town or the subsea cable crossing an ocean. They simply expect the answer to arrive.

But that invisible journey is one of the most important physical processes in the AI economy.

The chip may create intelligence.

The network is what makes that intelligence useful.

Workforce

Policy

Capital

Sustainability