Why Fiber Is the New Power in AI Data Centers

Published July 21, 2026

By Jennifer Lleras | A1 Data Center

The AI data center story has been dominated by power, and for good reason. Large AI campuses need extraordinary amounts of dependable electricity, and grid limits are already shaping where new projects can be built.

But electricity is only half of the infrastructure equation.

A site can have land, power, cooling, permits, and the latest compute hardware. It can still fall short if it cannot move massive volumes of data quickly, reliably, and securely. As artificial intelligence demand grows, fiber connectivity is becoming one of the most important factors in data center development.

Google’s newest transatlantic cable makes that point hard to ignore.

On July 21, Google announced the successful connection of Nuvem, a subsea cable linking Myrtle Beach, South Carolina, with Sines, Portugal, through Bermuda and the Azores. According to Reuters, the roughly 7,000-kilometer system includes 16 fiber pairs and has a total design capacity of about 384 terabits per second.

That is more than a cable milestone. It is a signal that the AI infrastructure race is also a connectivity race.

AI infrastructure depends on the routes that connect continents, not just the buildings that house servers.

AI campuses need more than megawatts

Power remains the first question in many AI data center conversations. Can the grid support the load? Is there enough generation nearby? Can the operator secure long-term power at a price that works? Can the site handle backup, cooling, and future expansion?

Those questions matter. AI clusters place heavy and sustained demand on electrical systems. Locations with available power have an advantage.

Yet compute only creates value when data can reach it and leave it.

AI systems depend on constant movement of information. Training large models requires huge data flows between storage, compute clusters, and cloud regions. Inference depends on low-latency connections between users, applications, and models. Enterprise AI workloads often move across private networks, public cloud platforms, edge locations, and global customer bases.

That means the value of a data center is no longer defined by its electrical capacity alone. It is also defined by how well it connects to the wider digital world.

A poorly connected site can face several limits:

• Higher latency for users and applications

• Fewer options for cloud and carrier interconnection

• More dependence on a small number of fiber routes

• Greater exposure to outages or congestion

• Higher network costs over time

  
  

For AI, those limits matter. Delays of milliseconds can affect real-time applications. Network congestion can slow data movement between regions. Route concentration can create operational risk.

Fiber is becoming a site selection issue, not an afterthought.

Subsea cables are the hidden highways of AI

Subsea cables may seem distant from a data center campus in Virginia, Texas, Georgia, Arizona, or the Carolinas. In reality, they form the backbone of global cloud and AI infrastructure.

More than 95% of global data traffic travels through subsea cables. These systems connect continents, cloud regions, financial markets, governments, research institutions, content platforms, and enterprise networks.

Nuvem matters because it adds a major new route between the United States and Europe. It connects the southeastern United States with Portugal through island waypoints in the Atlantic. That geography is meaningful because route diversity is becoming as valuable as raw capacity.

Capacity helps meet demand. Diversity helps keep systems resilient.

A new transatlantic route can support:

• More traffic between North America and Europe

• Better path options for cloud providers and carriers

• Reduced pressure on older or more crowded routes

• Stronger service continuity during maintenance or disruption

• More flexibility for companies placing workloads across regions

  
  

For data center developers, this changes the way coastal and inland markets are viewed. A cable landing point does not need to sit next door to every hyperscale campus, but the distance from a campus to dense fiber routes can shape performance, cost, and resiliency.

The strongest data center markets tend to combine several ingredients: power, land, cooling options, permitting feasibility, skilled construction capacity, and deep fiber access. As AI workloads grow, that last ingredient will carry more weight.

The physical strands of fiber carry the data flows that make AI workloads useful.

Bandwidth is becoming a form of capacity

Data center capacity is usually discussed in megawatts. That makes sense because power determines how much compute can run on a site.

But bandwidth is also a form of capacity.

A high-density AI campus needs enough network capacity to move data between servers inside the facility, between buildings across the campus, between cloud regions, and between end users around the world. If that movement is constrained, the compute investment cannot perform at its full potential.

This is especially true for AI because workloads are data-heavy at every stage.

Training can require large amounts of information to be gathered, cleaned, stored, copied, and moved to specialized compute clusters. Once a model is trained, deployments may need to support millions of prompts, responses, embeddings, searches, transactions, and data retrieval operations. Enterprises using AI may also need secure paths between private data stores and public or private AI systems.

The result is a shift in how infrastructure teams think about network design.

A strong AI data center strategy now needs:

Carrier diversity

Multiple network providers reduce dependence on a single carrier and create better options for pricing, routing, and uptime.

Route diversity

Separate physical paths lower the risk that one fiber cut or regional issue disrupts service.

Low-latency access

Shorter and better-designed routes improve response times for latency-sensitive workloads.

Scalable interconnection

Campuses need room to add cross-connects, cloud connections, exchanges, and private networking services.

Secure transport

AI workloads often involve sensitive company data, regulated information, intellectual property, or customer records. Secure, private, and well-managed network paths matter.

When these elements are missing, a data center may still function. It may not compete well for the most demanding AI workloads.

Fiber can influence where AI data centers get built

The next wave of AI data centers will not be placed only where electricity is available. They will also follow fiber-rich corridors.

That does not mean every AI campus must sit in the most famous data center market. In fact, power constraints are already pushing developers to look at new regions. But newer markets still need strong network foundations.

A location with abundant power but weak connectivity may need major fiber investment before it can support hyperscale AI. A location with fiber depth but limited power may also struggle. The best sites will solve both problems together.

This creates new strategic value for places near cable landing stations, long-haul fiber routes, internet exchanges, and cloud on-ramps. It also gives inland markets a reason to build stronger links to coastal landing points and major network hubs.

Nuvem’s Myrtle Beach connection is a useful example. A cable landing does not automatically turn a city into a top-tier data center hub. But it can strengthen the region’s digital infrastructure and create new options for carriers, cloud providers, and enterprises. Over time, those options can influence where workloads are routed and where supporting facilities make sense.

For developers, the lesson is practical: fiber due diligence needs to happen early.

That means asking questions such as:

• Which long-haul fiber routes serve the site?

• How many carriers are present or willing to build?

• Are there physically diverse paths out of the campus?

• What is the distance to major network hubs?

• Is the site exposed to single points of failure?

• Can the local permitting process support new fiber construction?

• Are there nearby rights-of-way, rail corridors, utility corridors, or highway paths?

  
  

These questions belong beside power, water, zoning, and tax considerations. They are no longer secondary details.

New fiber routes can determine whether emerging data center markets can support AI growth.

Resiliency is now a competitive advantage

AI infrastructure has a low tolerance for downtime. Cloud platforms, financial systems, logistics tools, health systems, industrial operations, and consumer applications increasingly rely on continuous access to compute and data.

That makes network resiliency a competitive advantage.

A single fiber route may be fast. It may even be inexpensive. But if it becomes a single point of failure, it creates business risk. Physical damage, severe weather, equipment issues, construction mistakes, and maintenance events can all affect network service.

Subsea systems face their own risks, including damage from ships, fishing activity, seismic events, and geopolitical tensions. Terrestrial routes face construction cuts, storms, wildfires, floods, and local infrastructure failures.

The answer is not fear. It is design.

Strong AI infrastructure uses layers of redundancy. That can include multiple carriers, multiple subsea systems, multiple terrestrial paths, redundant meet-me rooms, diverse entrances into a facility, and traffic management systems that can shift workloads when a path is degraded.

For data center owners and tenants, this is where connectivity becomes part of risk management. A facility with excellent power but limited network diversity may not meet the reliability expectations of mission-critical AI users.

The most valuable sites will be able to show clear answers to two questions:

1. What happens if one route fails?

2. What happens if an entire region has a connectivity issue?

   
   

The stronger the answer, the more attractive the site becomes.

Security depends on the network as much as the building

Data center security often brings to mind fences, cameras, access controls, guards, and hardened buildings. Those measures still matter. But AI raises the importance of network security as well.

AI systems often connect to sensitive datasets. They may process proprietary code, legal documents, medical records, industrial data, customer information, or financial records. The path that information travels is part of the security model.

Fiber routes, interconnection points, encryption practices, carrier relationships, and network monitoring all affect risk.

This is one reason major cloud and technology companies invest directly in subsea cables. Owning or participating in cable systems can give companies more control over capacity, routing, performance, and security. It can also reduce reliance on third-party routes in markets where demand is rising quickly.

For enterprise users, the takeaway is clear. Evaluating an AI data center should include more than rack density and power availability. It should include the network architecture that connects the facility to the rest of the world.

Key security questions include:

• Are private network options available?

• How are routes monitored?

• What encryption options are supported?

• Where are major interconnection points located?

• Which carriers and cloud platforms can be reached directly?

• How are redundant paths separated physically?

  
  

The answers can affect compliance, uptime, and customer trust.

Inside the facility, network design is as critical as power distribution for AI workloads.

The AI buildout will reward connected markets

The AI infrastructure race is often described as a contest for chips, land, and electricity. Those are still critical. But fiber is now part of the same strategic conversation.

A data center that cannot move information well cannot fully serve AI demand. A region that cannot offer diverse, scalable connectivity may lose projects to places that can. A campus that treats fiber as a late-stage utility connection may face costly delays or performance limits.

Google’s Nuvem cable is one example of a broader shift. Cloud and AI growth are driving investment in the physical networks that carry digital services. The next generation of data centers will depend on both electrons and photons, on power lines and fiber strands, on local campuses and global routes.

That is why Why Fiber Is the New Power in AI Data Centers is more than a headline. It reflects a practical reality for developers, operators, investors, and communities competing for the next wave of AI infrastructure.

Megawatts may determine where compute can run. Fiber will determine how well that compute can connect, scale, and serve the world.