산업 분석에 따르면 광섬유 구축이 AI 데이터센터 간 연결 확장에서 핵심 병목 제약으로 부각되고 있다.

A decade ago, DC Blox might have prioritized land near existing fiber from AT&T, Verizon, Zayo, or another established network provider. That calculation has shifted. As data center campuses have expanded and hyperscalers now represent a larger share of the customer base, connectivity has increasingly become another construction package tied directly to the project itself, according to Wabik.

“If Zayo or Verizon or AT&T just happens to be there close by, that's a good thing,” he said. “But fiber construction is inherently anymore just part of the construction component.”

This shifts the site-selection question from “Is there fiber nearby?” to “Can fiber be built here at the scale and diversity the customer requires?” For DC Blox, Wabik noted, this often means assessing whether sufficient public right-of-way exists to establish three, or sometimes four, diverse fiber paths into a facility.

This distinction matters because AI workloads are pushing infrastructure into markets where power, land, and energy options may be more abundant than established carrier density. Hyperscalers themselves have also emerged as major network builders.

Wabik provocatively characterized these companies as today’s telecom providers, citing the massive scale of terrestrial fiber they commission and the growing role of firms like Amazon, Google, and Meta in subsea cable development. The point is not that traditional carriers have vanished, but rather that hyperscalers increasingly design, commission, and control enormous portions of the connectivity required to support their own infrastructure.

DC Blox now treats requests for 864-count fiber as routine, and in some cases, 1,728-count cable. Such volumes would have been unthinkable in an earlier era, when a handful of fibers from an established carrier could easily satisfy a data center’s connectivity needs.

AI-Scale Fiber Gets Physical

The scale becomes tangible when the conversation shifts from abstract network capacity to what must actually be buried. Wabik noted that 864-count fiber is now roughly the smallest deployment DC Blox routinely handles. Consequently, the company no longer designs conduit systems around the assumption that a modest amount of spare capacity will suffice for decades.

A few years ago, installing a handful of ducts during an initial build might have seemed sufficient for the next two decades. Today, Wabik’s minimum expectation sits closer to 10 ducts, sometimes 14. “I'll put an 864 in it, I've got nine ducts left,” he said. “Who knows what's coming and when it's coming?” Sometimes, the answer arrives almost immediately.

Customers may request another 864-count cable before crews finish installing the first. The second and third ducts can begin filling while the original construction is still underway. Density is equally pronounced inside hyperscale network facilities, which Wabik described as 10,000- to 20,000-square-foot nodes containing hundreds of thousands of fiber splices, with equipment physically interconnected through vast webs of cabling.

This scale mirrors broader trends across AI infrastructure. Wabik recalled hyperscaler power requirements shifting rapidly from 25 kW per cabinet to 35 kW, and then to 50 kW, over mere weeks. Precise specifications often continue evolving even as the underlying facility is still in the planning stages.

DC Blox’s role, he noted, is to deliver the building, power, cooling, and connectivity infrastructure around customer equipment whose final configuration may continue shifting as deployment nears. That inherent uncertainty drives aggressive fiber over-provisioning. Because it is difficult to predict exact future requirements, the strategy is to build enough physical pathways to accommodate whatever comes next.

“It's just a massive buildout,” Wabik said. His shorthand for the moment is “beautiful insanity.”

Building a Connectivity Ecosystem

The result extends far beyond a collection of point-to-point fiber routes. AI-scale campuses are evolving into nodes within increasingly dense connectivity ecosystems. DC Blox may lay fiber between hyperscale buildings, across campuses, or into major network hubs. Customers can then request that this newly constructed network interconnect with other fiber providers at multiple points along the corridor.

These interconnections often require a series of intercepts and splices, allowing a customer’s network to meet others at six or seven locations along a single corridor. The resulting architecture resembles a mesh more than the traditional dual-route entry into a building. For hyperscalers, the ability to engineer this mesh is becoming a foundational element of the infrastructure equation. Yet, there is a critical wrinkle when discussing fiber as a potential bottleneck.

Wabik does not view raw fiber deployment capacity as the primary choke point. The industry is meeting demand by building at extraordinary scale. The more significant hurdle typically arises earlier: before fiber can be laid, permission must be granted to place it.

The Hidden Constraint Is Permitting

When asked whether connectivity can keep pace once a developer secures ample power, Wabik described data center development as a complex choreography. Land must be graded, substations constructed, underground conduit installed, buildings erected, and metro and long-haul connectivity extended to the campus. If any critical component falls behind schedule, the facility cannot operate as designed.

On the fiber side, Wabik noted that the greatest obstacle is rarely trenching or horizontal directional drilling—it is permitting. DC Blox recently developed a roughly 475-mile route between Myrtle Beach, South Carolina, and Atlanta. Over that distance, fiber traverses numerous municipalities and counties, each potentially demanding its own permit-equivalent approvals. While construction productivity is measurable—crews can bore a set number of feet per day and install a defined volume of cable—permits do not adhere to the same timeline. “Getting those permits is really the challenge,” Wabik said.

This reality has become significant enough to dictate routing decisions. If DC Blox learns that a specific county historically took a year to approve permits, effectively becoming the longest pole in a project’s schedule, Wabik said the company may simply route subsequent builds around that jurisdiction. This marks a consequential shift for an industry that traditionally confines permitting discussions to the data center campus itself. Permitting history now actively influences the geographic routing of the network connecting that campus.

A recent DC Blox project in metro Atlanta illustrates the friction. The company constructed roughly 55 miles of fiber to form a ring around the market. Most permits cleared during the first two months of a 12-month construction window. Others lagged. Several arrived at what Wabik described as “right at the wire,” following months of persistent follow-up with local jurisdictions.

Additional infrastructure crossings further complicate the process. Attaching fiber to a bridge brings a highway department into the approval chain. Crossing a river may require U.S. Army Corps of Engineers clearance. Traversing a railroad introduces yet another third party with distinct requirements and timelines. Wabik noted that DC Blox has encountered individual segment permits taking up to a year, or in some cases, 18 months.

The construction sequence then devolves into maximizing progress while awaiting final approvals. Demobilizing crews introduces additional complications. “If they have to go home, we have to remobilize them,” Wabik said. “That's just a huge cost we really try to avoid.”

Fiber Supply Tightens Again

Securing permission is only half the challenge. The physical supply chain has tightened once more. Wabik noted that equipment lead times ballooned during pandemic-era disruptions, only to ease somewhat throughout 2023 and 2024. Certain fiber-related components briefly returned to lead times measured in weeks.

That reprieve proved short-lived. Sourcing 864-count or 1,728-count fiber today can entail lead times of 70 to 80 weeks at standard market pricing, Wabik said. For an 864-count microfiber with an outside diameter of roughly 10 or 11 millimeters, he estimated pricing around $7 to $7.50 per foot.

Faster access to inventory is available for buyers willing to pay a premium. Wabik noted that fiber held in stock can trade for $10, $12, or even $14 per foot when projects face urgent material shortages. This supply pressure extends beyond connectivity; large backup generators for secondary data center power also face lead times ranging from 12 to 18 months.

The challenge, therefore, is not simply obtaining one scarce component. Developers are coordinating a collection of long-lead systems whose delivery schedule