The dominant narrative shared across the data center industry today undoubtedly revolves around power: how much is available, where it’s coming from, how quickly it can be delivered, and whether traditional power infrastructure can continue to cope with rising demand. But while power is central in navigating the roadmap to sustainable and consistent growth, it’s far from the whole picture.

Investment in data centers themselves and new cycles of next-generation GPUs is at an impressive high. Yet too often, networking – the backbone of operations – is treated as something to be bolted on later, rather than designed in from day one.

When connectivity is an afterthought, it becomes the bottleneck that slows deployment, performance, and ultimately limits the value of compute and power investment. Securing the best returns rests on re-envisioning networking infrastructure globally for the AI era.

Against this backdrop, in a recent DCD>Broadcast, industry experts from Corning, Vultr, and the Telecom Industry Association (TIA) share why next-gen networks and connectivity are critical to realising the bigger picture.

There is no AI without broadband

For David Stehlin, CEO of TIA, the value of connectivity can be illustrated clearly: “There is no AI without broadband. It’s just that simple. So you need broadband everywhere.”

Having witnessed multiple technology waves over four decades in the industry – from the rise of fiber in long-haul networks, to fiber to the home (FTTH), wireless, and the internet itself – Stehlin explains how with each advancement, the industry assumed it had reached its peak, only to be proven wrong by the next leap forward.

AI is the latest inflection point in this series of technological advancements, and it’s exposing how interdependent power, data centers, and connectivity really are. In the US alone, data centers are expected to grow from 25GW of demand in 2024 to more than 80GW in 2030.

Meeting this demand will require significant upgrades not just in power generation, but in distribution and physical infrastructure. Yet none of this matters if the data can’t move quickly, reliably, and securely between internal systems and across facilities.

From local links to global reach

Modern AI clusters aren’t just bigger versions of traditional workloads – they’re fundamentally different and endlessly more complex.

“We’re putting together large supercompute clusters,” says Kevin Cochrane, chief marketing officer at Vultr. “And to these clusters, we’re bringing a lot of data.”

This shift has altered the way in which we think about and design networks. AI training generates massive data flows inside the data center, while inference workloads demand high-speed access across racks, facilities, and ultimately across the world.

Data centers are no longer isolated entities. Operators are building facilities in key regions and connecting them to hubs elsewhere, making long-distance connectivity a core operational requirement. This is where fiber becomes non-negotiable.

“We need to make a substantial move to fiber optics,” explains Cochrane. “It’s essential for delivering higher rates of transmission and better connectivity across long distances.”

Connectivity challenges don’t end at the data center doorstep. AI is inherently and increasingly global, meaning the infrastructure supporting it must match up. Delivering AI services to a global population requires strong and reliable inter-regional and international connectivity via resilient routes.

“We need to have multiple access points where the cables actually terminate,” adds Cochrane. “That gives us greater redundancy and availability when we’re piping traffic through to our data center locations.”

Undersea cables, alternate terrestrial routes, and geographically diverse landing points are all part of this picture. As is the reality that connectivity is uneven across the globe. In regions such as Sub-Saharan Africa or parts of Latin America, bandwidth constraints remain a major barrier to growth.

“If you want to reach 90 percent of the world’s population within a couple of milliseconds, there’s still a lot of work to do in connectivity,” says Cochrane.

Fiber growth in every direction

In today’s fast-paced, high-stakes landscape, innovation is occurring all at once across the board – and for connectivity, it’s no different. “There’s no one place where fiber is growing,” says Stehlin. “It’s multi-dimensional.”

Advancements are happening at the systematic and operational level in how networks are designed, built, and maintained. But speed of deployment is becoming just as important as technical capability and performance.

“We can’t do these things in series,” explains Stehlin. “They have to be done in conjunction with one another in order to keep up.”

From Corning’s perspective, AI infrastructure has to scale in three distinct ways: within the rack, across racks in the data center, and out across the globe.

“Each of these architectures requires different networking,” says Cochrane. “And a different kind of innovation.”

Inside the rack, optical scale-up is emerging as a critical goal. Moving optics closer to compute is central to delivering higher performance and better power efficiency, but it also introduces new design and deployment challenges.

Beyond the rack, scale-out architectures demand relentless improvements in bandwidth, latency, and resilience. And as inference workloads grow, so does the need to distribute large trained models and real-time data flows across geographically distributed clusters.

“The amount of bandwidth required is through the roof,” adds Cochrane. “Latency tolerance is shrinking, and users expect real-time responses.”

As infrastructure scales, non-technical constraints are becoming just as limiting as bandwidth or power. Permitting, for instance, has become a key bottleneck. Whether it’s approvals for new data centers or undersea cable landing points, these delays are generating friction.

Similarly, workforce challenges are mounting. Skilled labor is in short supply and experienced technicians are aging out of the industry. Building networks at this scale requires not just better products, but solutions that can be installed faster, with fewer specialist skills. That’s where the physical layer comes into frame.

Into the physical

“Relatively speaking, it’s easy to design the logical network,” says Josh Smith, market development manager for data centers at Corning. “But then physics and the real world get in the way.”

As AI clusters grow denser, fiber counts rise dramatically. Traditional approaches that rely heavily on jumper cables and field splicing are no longer viable at scale. Install times are too long, and labor constraints compound the issue.

Decisions about cabling architecture, connectorization, and physical layout are now having a dramatic and decisive impact. Left too late, even well-established projects can stall while teams race against the clock to adapt designs that were never meant for this level of density.

To keep pace, fiber is being pushed to new limits. Smaller diameters, higher fiber counts, and more compact connectors are becoming standard. In some cases, what once held eight fibers now supports 32 in a fraction of the space.

But density alone isn’t enough. Installation methods also need to evolve. Smith expands: “We’re seeing a drive toward new connector technologies, including lens-based connectivity.”

By using lenses and mirrors to bridge air gaps, these systems reduce dust sensitivity and eliminate physical contact. As a result, hundreds of fibers can be connected at once – and even entire plug-and-play units that deliver power, cooling, and optical connectivity all at the same time.

“That’s the dream state,” adds Smith. “Install speed, density, and bandwidth all improving and working together.”

Watch out for the next bottleneck

As fiber counts climb into the thousands, traditional splicing methods simply can’t keep up. Pre-terminated solutions are becoming essential, but they require far more upfront planning.

Cable lengths, routing, and measurement accuracy all matter more than ever before. Even seemingly small details can have major consequences at scale.

This is where digital tools, such as BIM models and digital twins, become increasingly mission critical. Used correctly, they help teams identify problems before they hit the site, rather than reacting once schedules are already under pressure.

Corning’s approach focuses on building resilience into the physical layer itself. Bend-insensitive fibers, clearer installation guidance, and fully connectorized solutions act as insurance policies against the risk of human error.

“We try to think through all the ways this can go wrong and make sure it goes right,” says Smith.

Looking ahead, Smith posits that technologies such as augmented reality training and further automation could help bridge the skills gap and support faster, more consistent builds.

Connectivity as a core design principle

At the advent of the internet, connectivity was rebuilt to support a new way of working. AI represents a similar shift – and it demands the same level of rework.

The message from industry experts from across the data center ecosystem is that power and compute matters, but neither are enough to unlock the full potential of the AI era in isolation. Without connectivity designed as a foundational element, the puzzle is not complete.

Networks are no longer back-office supporters – they’re active enablers of performance, resilience, and global reach. Treating them as an afterthought risks leaving facilities a generation behind before they have the chance to hit the ground running. Delivering AI infrastructure that endures must begin with the network.

To hear more about current trends and future outlooks for networks and connectivity, watch the full DCD>Broadcast episode, here. And to find out more about navigating the network design challenges posed by artificial intelligence, check out ‘Advancing the AI revolution’ a new eBook from DCD and Corning.