Scaling connectivity up and out across the data hall has largely defined modern network architectures. Yet the complexity reaches new heights as ultra-low-latency, reliable connections aren’t just critical inside individual facilities, but increasingly required between physically separate data centers.

When stretching across buildings, campuses, and beyond, connectivity requirements and the accompanying challenges take on a distinctly different shape. While the fundamental objective of connecting AI resources efficiently remains intact, the required solutions can vary depending on the distance involved. Choosing the right connectivity approach starts with understanding the unique constraints of each environment.

So, what happens when AI infrastructure must extend beyond one facility? In a recent DCD>Broadcast episode, Corning’s Nilson Gabela unpacks this key question and shares how the company is helping to connect modern facilities more efficiently and reliably beyond the constraints of the data center walls.

Understanding scale across

Setting the context of the conversation, Gabela offers a definition of scale across networks: “I think the most useful way to think about it is that ‘scale across’ is a category that happens when AI infrastructure expands beyond a single site.”

Under the broader scale-across umbrella, Gabela identifies three distinct segments: campus data center interconnect (DCI), regional DCI, and long-haul DCI – each presenting its own combination of physical and operational considerations.

Campus DCI typically covers distances below 10km and is generally used to connect multiple buildings within the same campus. Regional DCI extends from around 10km to 100km, connecting multiple sites, meaning latency and network resilience become increasingly important.

Beyond 100km, long-haul DCI connects sites over significantly greater distances, introducing additional concerns around attenuation, amplification, and signal regeneration.

“There are different distances, different physics, and the operational piece is extremely different as well,” adds Gabela.

While all three sit within the same scale across category, the challenges evolve considerably as the distance increases, meaning infrastructure designed for one environment can’t necessarily be applied directly to another.

Managing density and distance

For campus DCI, the primary pressure is physical density. Connecting multiple buildings requires large volumes of fiber, while existing pathways and space can quickly become constrained as the volume of connectivity required increases.

“AI data centers can require roughly ten times more connectivity than traditional environments, placing additional pressure on already constrained infrastructure,” explains Gabela. “As fiber counts increase, operators must also consider how large cables can be routed efficiently around and between buildings.”

This makes the physical design of the network an increasingly important consideration, particularly for operators working within existing infrastructure where expanding current capacity isn’t always straightforward.

Regional DCI introduces another layer of complexity. Operators must connect multiple sites while maintaining consistent performance across the wider network. Here, latency and resilience become particularly important, with connectivity needing to support reliable communication between facilities over significant distances.

“Once you go to the regional DCI piece, the problem expands a little bit more. Now, you have to link those sites, and by doing so, latency has to be very, very precise,” adds Gabela.

Capacity remains important, but sufficient redundancy must also be built into the network to ensure that individual connections can be added or replaced as requirements evolve.

Microduct and micro-cable solutions can help address this challenge by providing high fiber counts within constrained pathways while leaving additional capacity available for future expansion.

For example, Gabela highlights a seven-way microduct installation that can initially accommodate one cable carrying hundreds of fibers, while leaving the remaining pathways free for future connectivity requirements.

Long-haul DCI presents a different set of challenges again. Once distances exceed 100km, attenuation becomes increasingly important because signal degradation can limit how far information can travel without regeneration.

“Capacity still matters. We want to get as much fiber as we can into those ducts, but latency is also very important here,” says Gabela. “Attenuation is critical over these long distances.”

One of the less obvious challenges that comes with scaling across is power and land availability. Long-haul networks may require huts containing repeaters to amplify signals along the route, meaning operators need access to both suitable sites and reliable power infrastructure.

Solving today's connectivity pressures

Across all three segments, sits the same broader challenge: how to expand AI capacity while maintaining performance and deploying infrastructure quickly enough to keep pace with demand. The specific priorities, however, vary according to the environment.

For campus DCI, this means maximizing the amount of fiber that can fit within existing pathways while accelerating installation. Pre-terminated solutions can help address the deployment challenge by delivering ready-to-connect cable assemblies that reduce the amount of work required on site.

For regional DCI, the focus shifts towards connecting multiple sites consistently and reliably, with latency, resilience, and future capacity all playing a role in network design. Operators need infrastructure capable of supporting current requirements while leaving flexibility for future expansion.

Long-haul environments place greater emphasis on reducing infrastructure and power costs while maximizing capacity within existing ducts and routes. Gabela highlights Corning's work with Lumen as an example of how cable design can help operators increase capacity without necessarily rebuilding the underlying infrastructure.

Corning redesigned a traditional 432-fiber cable to double the fiber capacity within the same conduit, allowing the customer to make greater use of infrastructure that was already in place.

This example illustrates how improvements in cable design can help address capacity constraints within existing networks, particularly where installing additional pathways would be expensive or impractical.

Preparing for the next generation

While today's technologies are helping to address immediate density and deployment challenges, the scale of AI infrastructure means operators also need to consider what their networks will require several years from now.

Emerging technologies such as multi-core fiber, pre-terminated multi-core solutions, and hollow-core fiber are expected to play different roles within this future landscape.

For multi-core fiber, the primary opportunity is density, with the technology offering the potential to significantly increase capacity within constrained pathways. Pre-terminated multi-core solutions combine this density advantage with faster deployment, helping reduce some of the practical challenges associated with handling and splicing higher-density fiber.

By reducing latency and attenuation, hollow-core fiber could allow operators to extend connectivity over greater distances while maintaining the performance required by AI workloads.

Gabela describes the three technologies as addressing distinct challenges within scale across:

“Multi-core fiber is for density. Pre-term multi-core is more a deployability solution. Hollow core is more concerned with performance within the scale across as we look towards future requirements.”

However, widespread adoption will depend on reaching the limits of the solutions available today. Until existing approaches can no longer provide the required combination of capacity, performance and deployment speed, emerging technologies will continue to develop alongside current infrastructure rather than replacing it altogether.

Designing for flexibility

“When it comes to preparing for the future, the number one thing you have to pay attention to is infrastructure,” says Gabela. “Leave yourself the option to grow.”

In campus environments, this could mean installing additional conduit capacity, while regional networks could incorporate multi-way microduct systems that allow additional cables to be installed as requirements increase.

The same principle applies to cable selection. Operators may begin with traditional cable designs and subsequently move towards micro-cables, pre-terminated solutions or emerging technologies as their requirements evolve.

Ultimately, scale across demonstrates how the geographical expansion of AI infrastructure is creating a new set of connectivity challenges, with priorities changing significantly depending on whether the goal is to connect buildings, sites, or cities.

For Gabela, understanding these distinctions is the first step towards building the right infrastructure:

“Start by identifying where you are on the scale-across spectrum, and which of these constraints is most important to your application.”

As he puts it, operators should think of these elements together “as a hyperscale data center solution category” – infrastructure that connects people, data, and information across increasingly complex AI environments.

To hear more about scale-across networks from Nilson Gabela, watch the full DCD>Broadcast episode, here.