A BlackForge Ventures company
BlackForge Data Centers

Water, gas & fiber

How Long-Haul and Subsea Fiber Routes Affect Data Center Site Selection

Long-haul fiber connects regions and subsea cables connect continents, and a data center site’s value to large users depends partly on how directly it can reach both. Long-haul routes closely follow road and rail corridors and often share conduit, so proximity is common but usable access points and truly diverse paths are not.1 Subsea cables carry the great majority of intercontinental traffic, but most inland campuses reach them only through terrestrial backhaul to the hubs where that capacity is handed off.2

Last reviewed · 10 min read · BlackForge Data Centers

Key takeaways

  • Long-haul routes run along roads, railroads and other linear corridors, but you can only connect where a provider has, or will build, an access point.1
  • Route distance, not map distance, sets latency: light in fiber adds about 5 microseconds per kilometer each way, or roughly 1 millisecond of round-trip time per 100 km.3
  • Diversity at regional scale means routes that do not share a conduit, river crossing or corridor, and many providers do share them.1
  • Subsea cables land at coastal landing stations that need an FCC cable landing license and federal and state coastal approvals; inland sites reach them through backhaul.45
  • AI demand is driving new long-haul construction, but remote campuses often still need a new build, which is a right-of-way project.67

01What long-haul fiber is and where it runs

Long-haul fiber carries traffic between metro areas and regions, often over hundreds of kilometers. These are high-strand-count cables, usually in buried conduit, built along continuous rights-of-way such as railroads, highways, pipelines and electric transmission lines. A peer-reviewed mapping study of the U.S. long-haul network found a clear correspondence between long-haul fiber and the roadway and railway systems, and found that providers commonly place fiber in jointly used, previously installed conduit because building new conduit costs far more.1

Because light weakens over distance, long-haul routes need in-line amplifier sites along the path, typically placed about 80–100 km apart depending on the system design.8 These small buildings or huts hold power and optical equipment. Along with splice points, they are the natural places where a route can be accessed, which is one reason a cable passing a site does not mean the site can connect to it.

Our fiber connectivity guide covers the basics of route types, diversity and laterals, and dark fiber vs. lit services covers how capacity on these routes is bought. This guide focuses on the regional and global layer.

02Why long-haul routes matter in site selection

For latency-sensitive colocation and cloud sites, metro fiber and proximity to interconnection hubs usually matter most. Long-haul routes matter most for sites outside major metros, which is where many large power-driven campuses end up. A remote campus needs high-capacity, diverse paths back to the hubs and regions it serves, or it cannot be used regardless of how much power it has.

How long-haul fiber weighs by site type
Site typeMain fiber needLong-haul role
Metro colocationDense metro fiber and many carriersPresent through the metro hubs; rarely a site-level question
Cloud region facilityLow-latency links among nearby facilities and hubsImportant for paths to other regions
Remote hyperscale or AI training campusHigh-capacity diverse paths to distant hubsCentral; may require new construction
Inference or user-facing AIReasonable latency to population centersImportant; distance limits how remote the site can be
EdgeLocal access networksUsually minor

Provider data points the same way. Zayo’s annual bandwidth report, drawn from its own sales, found that long-haul dark fiber demand from its largest customers doubled between 2024 and 2025, with hyperscalers and carriers accounting for about 95% of long-haul purchases.9 Most of the roughly $5 billion in AI-related contracts Lumen announced in 2024 was described by analysts as hyperscalers buying 20-year dark fiber rights to connect new AI training data centers.10

The shift toward large, remote AI campuses has made long-haul access a real screening criterion rather than an afterthought. See AI data center site requirements, what a gigawatt-scale AI campus requires and the tradeoffs in near-metro vs. remote power-rich sites.

03Latency, distance and route geometry

Light in fiber travels at roughly two-thirds of its speed in a vacuum, which adds about 5 microseconds of delay per kilometer each way, or roughly 1 millisecond of round-trip time for every 100 km of route.3 Long-haul routes rarely follow a straight line. A rail corridor may bend around terrain, and a route between two cities may detour through a third. The useful number is route distance from the site to each target hub, plus the lateral, and equipment along the path adds a little more.

Fig. 1Minimum round-trip delay by fiber route length

  • 100 km route1
  • 500 km route5
  • 1,000 km route10
  • 2,000 km route20
  • 4,000 km route40

milliseconds

Propagation delay only, at about 5 microseconds per kilometer each way. Equipment, routing and indirect paths add more.3

For training workloads, a few extra milliseconds is usually acceptable. For inference, interconnection and cloud regions that pair facilities, it may not be. When a user gives a latency budget, convert it into a maximum route distance and test each candidate site against actual routes, not radius circles. The InterTubes researchers also found that new links along previously unused transportation corridors could cut both risk and propagation delay, which is part of the case for new routes to remote campuses.1 Our guides to latency requirements by workload and AI training vs. inference site needs cover the budgets.

04Diversity at regional scale

Diversity at the site boundary is only half the story. Two long-haul routes that leave a site in opposite directions can still converge on the same bridge, mountain pass or city. Natural chokepoints concentrate routes: major river crossings, narrow valleys, and the approach to large metros. A single event at one of those points can cut several routes at once. The InterTubes study found that many network paths depend on a relatively small set of heavily shared long-haul links, and that hardening or bypassing those links does the most to improve resilience.1

  • Trace each route from the site to its destination hub, not just to the nearest junction.
  • Identify shared crossings and corridors, including routes owned by different providers that use the same rail line, conduit or bridge.
  • Prefer destinations in different metros, or at least different buildings, so one hub outage does not isolate the site.
  • Check hazard exposure along each path, such as floodplains, wildfire areas and seismic zones.

The same logic applies at global scale. On September 6, 2025, several subsea cables were damaged in the Red Sea near Jeddah, including SMW-4 and IMEWE.11 Microsoft warned Azure customers of higher latency on traffic passing through the Middle East while it rerouted that traffic over other paths.12 The traffic kept flowing because diverse routes existed, but it flowed more slowly.

05Subsea cables and landing stations

Subsea cables carry the great majority of intercontinental traffic. TeleGeography has confirmed the commonly cited figure of more than 99%, while noting that a precise calculation is not possible without current data on satellite traffic.2 Each cable comes ashore at a cable landing station, a facility near the coast where the marine cable meets terrestrial networks. From there, terrestrial backhaul routes carry traffic inland to network hubs and data centers.

Fig. 2From subsea cable to inland campus

  1. 01

    Subsea cable

    Marine segment between continents

  2. 02

    Landing station

    Where the marine cable meets terrestrial networks

  3. 03

    Terrestrial backhaul

    Fiber from the coast to inland hubs

  4. 04

    Network hub

    Carrier hotels, exchanges and cloud on-ramps

  5. 05

    Campus

    Diverse long-haul or metro paths from the hub

A simplified path. Where capacity is handed off, and to whom, varies by cable system and carrier.

Ownership has shifted toward the largest cloud and content companies. They now account for roughly three-quarters of international bandwidth and take part in more than two-thirds of planned cable deployments, according to TeleGeography data.13 Meta’s Project Waterworth, announced in February 2025, is planned at more than 50,000 km across five continents, with landings in the United States, India, Brazil and South Africa.14

Fig. 3Subsea cables at a glance

of intercontinental traffic
>99%
of international bandwidth: hyperscalers
~75%
planned length of Meta’s Waterworth
50,000+ km
Traffic share per TeleGeography;2 hyperscaler share per TeleGeography data as reported in 2026;13 Waterworth per Meta’s 2025 announcement.14

Siting near a landing station has a few specific considerations:

  • Landing stations cluster on stretches of coast with suitable seabed, shore conditions and permitting history, so they are not evenly spread.
  • Coastal sites face storm surge, coastal flooding, salt air and, in some areas, hurricane exposure. Our guide to natural hazard risk covers how to screen these.
  • Coastal land is often expensive, constrained and subject to additional coastal zone review.
  • A new landing needs an FCC cable landing license, a U.S. Army Corps of Engineers permit for work in U.S. waters and, in coastal states, a federal consistency review under the state’s coastal program.45

Licensing itself has been changing. The FCC adopted new submarine cable landing license rules in August 2025 to speed review and address national security risks, and the last of those rules took effect on July 8, 2026.4 Industry groups had complained that licensing timelines had stretched from about nine months to more than three years in some cases.15 A new landing is a different project from a data center, with its own lead times.

For most inland campuses, the practical question is how directly the site connects to the hubs where subsea capacity is handed off, and whether the backhaul from those hubs is diverse. A site does not need to be near the coast to use international capacity.

06When the route has to be built

Large remote campuses sometimes sit far enough from existing routes, or need enough capacity, that a provider builds new long-haul cable to them. That turns fiber into a linear infrastructure project with its own path, schedule and approvals.

Providers are building more of these routes. In 2025 Zayo broke ground on AI-focused routes including a 385-mile low-latency Chicago–Columbus route, a 521-mile Chicago–Minneapolis dark fiber route and a 123-mile Phoenix–Tucson route.6 In November 2025 it completed a 622-mile route from Umatilla, Oregon, through Prineville to Reno, Nevada, partly funded by a federal middle mile grant.7

Fig. 4Selected new Zayo long-haul routes, 2025

  • Umatilla–Prineville–Reno622
  • Chicago–Minneapolis521
  • Chicago–Columbus385
  • Phoenix–Tucson123

route miles

Routes started or completed in 2025, as announced by the provider.67
  • Right-of-way: easements or permits along roads, rail lines, pipelines or transmission corridors, each with its own owner and terms. Placement in highway right-of-way follows the state DOT’s utility accommodation policy, which a federal rule effective in 2022 required states to update with broadband coordination requirements.16
  • Crossings: railroads, highways, rivers and wetlands, which can require separate agreements and permits.
  • Environmental review: a long route can cross species habitat or cultural resource areas, and a federal permit, federal funding or federal land can trigger NEPA review.
  • Amplifier sites: small parcels along the route for equipment huts with power, spaced to suit the optical design.8
  • Commitment: providers usually want a committed anchor customer before building.

Projects that pair power and fiber along a shared corridor, such as a new transmission line, can sometimes simplify this, but the agreements are separate and the timing rarely lines up on its own. Landowners asked to grant a route should read our guides to mineral rights and easements and transmission line easements.

07How to screen a site for long-haul access

  1. 01List the user’s target hubs or regions and its latency budget, if any.
  2. 02Map long-haul routes within a reasonable distance and identify the owner of each.
  3. 03Find known access points and amplifier sites on each route, and ask owners where new access is possible.
  4. 04Measure route distance to each target hub and convert it to round-trip latency, at roughly 1 millisecond per 100 km.
  5. 05Trace each path for shared conduit, chokepoints and hazard exposure.
  6. 06If new construction is needed, sketch the corridor, crossings and landowners, and ask which providers would build.

When we screen remote sites, long-haul access is checked alongside power, because the sites with the most power available are often the ones farthest from existing high-capacity fiber. If you want that check on a specific property, you can get a site reviewed.

CalculatorCooling water use →CalculatorFiber latency & distance →

Common questions

What is long-haul fiber?

Long-haul fiber is high-capacity optical cable that connects cities and regions over long distances, often hundreds of kilometers. It usually runs in buried conduit along railroads, highways and other corridors and needs amplifier sites roughly every 80–100 km. For data centers, long-haul routes provide the paths that connect remote campuses to network hubs and to other regions.

How far can a data center be from a network hub?

It depends on the workload’s latency tolerance. Round-trip latency grows by roughly 1 millisecond per 100 km of fiber route. Training campuses can often be hundreds of kilometers from hubs, while interconnection-heavy colocation and paired cloud facilities need to be much closer. Measure actual route distance, which is usually longer than straight-line distance, and compare it with the user’s budget.

Does a data center need to be near a subsea cable landing station?

Usually not. Most data centers reach subsea capacity through terrestrial backhaul to the network hubs where that capacity is handed off. Being near a landing station can help for some international or content workloads, but coastal sites bring storm surge, flooding, land cost and coastal permitting concerns.5 For most inland campuses, diverse paths to those hubs matter more than proximity to the coast.

Who builds long-haul fiber to new data center campuses?

Mostly established fiber providers, often after a large customer commits to capacity. Zayo began several new AI-focused long-haul routes in 2025, and Lumen’s 2024 AI contracts with hyperscalers included new routes and long-term dark fiber rights.610 Very large operators also build or co-build their own networks, including subsea cables.14

Can I connect to a long-haul fiber line that crosses my property?

Not necessarily. Long-haul cables can only be accessed where the owner has splice points or amplifier sites, or agrees to create a new one, and the strands may already be allocated. A route crossing a property is a useful signal and a reason to contact the owner, but connection depends on the provider’s capacity, design and commercial interest.

Notes

  1. 1.R. Durairajan, P. Barford, J. Sommers and W. Willinger (ACM SIGCOMM), “InterTubes: A Study of the US Long-haul Fiber-optic Infrastructure,” 2015. pages.cs.wisc.edu
  2. 2.TeleGeography, “Do Submarine Cables Account For Over 99% of Intercontinental Data Traffic?,” 2023. resources.telegeography.com
  3. 3.CommScope, “Latency in Optical Fiber Systems,” n.d. commscope.com
  4. 4.Federal Communications Commission, Office of International Affairs, “DA 26-684 (Public Notice),” 2026. docs.fcc.gov
  5. 5.Oregon Department of Land Conservation and Development, “Coastal Project Notice for Federal Permit NWP2018-179-1: Jupiter Submarine Cable System,” 2019. omls.oregon.gov
  6. 6.Zayo, “Zayo Levels Up Its AI Ready Infrastructure with New Routes and 100/400G Enablement,” 2025. zayo.com
  7. 7.Zayo (Business Wire), “Zayo Builds the Backbone for Western U.S. AI Growth with Completion of the Umatilla–Prineville–Reno Route,” 2025. businesswire.com
  8. 8.Juniper Networks, “PTX Inline Amplifier (ILA) Description,” n.d. juniper.net
  9. 9.ISE Magazine, “AI Infrastructure Expansion Drives Record Demand for Fiber Networks and Bandwidth Capacity,” 2026. isemag.com
  10. 10.Fierce Network, “Lumen landed $5 billion in contracts to help hyperscalers and the like meet AI demands,” 2024. fierce-network.com
  11. 11.Kentik, “Subsea Cables Parted in Red Sea Again,” 2025. kentik.com
  12. 12.TechRadar, “Microsoft Azure services see major disruption after Red Sea cables cut,” 2025. techradar.com
  13. 13.IEEE Communications Society Technology Blog, “Hyperscalers Dominance of Subsea Cable Capacity to Increase in the AI Era,” 2026. techblog.comsoc.org
  14. 14.IT Brew, “Meta unveils ambitious five-continent-spanning subsea cable project,” 2025. itbrew.com
  15. 15.INCOMPAS, “INCOMPAS Applauds FCC on Adopting Submarine Cable Item,” 2025. incompas.org
  16. 16.U.S. Department of Transportation, Federal Highway Administration, “Broadband Infrastructure Deployment (final rule, 2021-26231),” 2021. transportation.gov

Have a site in mind?

Get a straight answer on your land.

Send a parcel number, an address, a map pin or a target load. We’ll tell you what it can support and what it would take.

Start a conversation →

This guide is general information about data center site selection. It is not engineering, legal, tax or investment advice. Requirements vary by state, utility and county, so confirm the specifics for any site with the relevant authorities and advisors.

Terms in this guide

Related guides

More in Water, gas & fiber

Guide 8 of 18 in Water, gas & fiber · all 242 guides · glossary · calculators