Kirkwood IG CEO Scott Bergs discusses fiber connectivity as critical enabler for AI data center buildout and interconnect density.

For years, fiber was a forgiving element of data center site selection—developers could secure land, line up power, plan the facility, and then work with carriers on connectivity. At AI scale, that model no longer works.

The architecture of traditional communications service provider networks fails to meet the latency and capacity needs of high-density compute environments, according to Scott Bergs, CEO of Kirkwood Infrastructure Group. As a result, network infrastructure can no longer be treated as an afterthought. For hyperscalers and neo-cloud providers, fiber route diversity, latency, physical security, and future capacity must now be considered alongside power and land from the start.

As data center campuses expand farther from established digital hubs in pursuit of available power, the challenge scales accordingly. Inter-campus connections that once extended two or thirty miles now stretch 250 miles or more—distances that were traditionally considered long-haul fiber routes are increasingly becoming routine inter-campus infrastructure.

Bergs previously led DF&I, a dark-fiber platform in Northern Virginia and Maryland. Kirkwood Infrastructure Group represents a second phase in a broader infrastructure investment strategy developed through IPI Partners, an investment platform focused on digital infrastructure. When IPI was acquired by Blue Owl Capital, the investment funds and portfolio companies remained intact while accessing a broader capital base. Kirkwood emerged as the team and operating platform separated from the underlying network assets, creating flexibility as individual markets and funds mature. The company now operates across Florida, Georgia, Mississippi, Alabama, and Louisiana while evaluating projects in Kansas and Missouri.

The challenge of timing is critical. For a traditional 5 MW data center, developers could build first and create an ecosystem of carriers afterward. Hyperscale and neo-cloud campuses require a fundamentally different approach. Existing carrier networks often lack the capacity and latency engineering for dense AI compute. Even when capacity exists nearby, physical route diversity requirements—typically four diverse network paths for high-density facilities—can force substantial new construction.

An area with ample capacity on one or two routes may still require two or three new physical builds to meet connectivity requirements. Waiting until late in development to discover this problem creates serious consequences. Large parcels can be physically landlocked. Developers may require private easements. Public rights-of-way may not reach the property usefully. Natural barriers can eliminate routes entirely. Compounding this, power infrastructure and road improvements frequently disturb public rights-of-way, potentially triggering construction moratoriums that prevent fiber providers from accessing the same corridors. Late-stage connectivity planning can extend project schedules or render necessary routes impossible to build. Connectivity planning must now begin before the ultimate tenant is even identified, allowing fiber construction to coordinate with power, transportation, and other infrastructure while protecting the target service date.

Power availability remains a dominant force shaping data center geography. Developers increasingly examine sites not only for available utility capacity but also for their ability to support behind-the-meter generation—requiring evaluation of proximity to natural gas pipelines, available capacity, zoning permits for onsite generation, and whether required equipment can arrive on schedule. A site with theoretical natural gas access is not necessarily executable for power, and a location with executable power may lack adequate communications infrastructure. This is creating a new class of potential markets: locations with sufficient power and land but considerably less communications infrastructure than established hubs like Northern Virginia. Network developers must now follow compute into these areas with extended, earlier planning.

Over three decades, Bergs estimates that his capacity projections have been consistently wrong—and never overestimated. This reflects the progression in network consumption. Enterprise and carrier networks initially relied on lit telecommunications services. As bandwidth increased, some users moved to dark fiber. High-density compute has pushed requirements further: from individual fiber strands to dedicated cables to dedicated conduit.

Kirkwood increasingly sees demand—in some cases mandates—for conduit capacity rather than a specified fiber count. Dedicated conduit provides hyperscalers with greater physical-layer security, allows operators to control routing and latency, and creates room to expand without disturbing existing traffic. Fiber counts within these routes are rising rapidly. Kirkwood sees conventional single-mode cables ranging from roughly 864 to as many as 6,912 fibers depending on application, making the conduit system itself a strategic asset.

Emerging technologies like hollow-core fiber complicate this picture. Hollow-core fiber guides light through an air-filled core rather than glass, offering potential latency and performance advantages. However, current implementations have a larger physical form factor than conventional single-mode fiber—a practical concern when maximizing capacity within existing conduit. An inch-and-a-half conduit may not accommodate enough hollow-core fiber to create a compelling capacity advantage over a conventional 6,912-count cable in the same space. Installation also presents challenges, requiring different splicing expertise than conventional single-mode fiber.