Visitors to a data center are shown the generators, the cooling plant and the halls of servers. They are rarely shown the fiber. Yet a facility with perfect internal redundancy is still only as available as the network routes that connect it to the outside world.
Power and cooling get careful engineering because everyone understands what happens when they fail. Fiber deserves the same rigor. It is buried, hidden in conduit and spread across miles of public right-of-way, which makes it easy to overlook until a cut takes a facility offline. This perspective covers what real diversity means, how to verify it, how to plan for growth and why documentation matters as much as the glass itself.
Diversity is physical
Two fiber circuits are not diverse just because they come from two carriers. If both run through the same conduit, the same bridge or the same street, one backhoe can take them both out. Real diversity means separate building entries, separate paths and, ideally, separate routes all the way to different network hubs. Proving it requires route maps, not assurances.
The problem is common because carriers lease capacity from each other and share infrastructure more often than buyers realize. A circuit sold by one carrier may ride on another's fiber for part of its route. Two carriers may both use the same bridge crossing because it is the only practical way across a river. On paper, the facility has two providers. In the ground, it may have one path. A facility with perfect internal redundancy is only as available as the routes that connect it.
A facility with perfect internal redundancy is only as available as the routes that connect it.
How to verify diversity
Verification is detailed work, but it is not mysterious. A facility operator should insist on seeing where the fiber actually goes, end to end, and should check for the places where routes tend to converge:
- Building entries: at least two, physically separated, ideally on opposite sides of the facility.
- Pathways inside the building: separate routes from each entry to the meet-me rooms, not one shared tray.
- The last mile: separate streets and conduits from the property line to the first splice points.
- Shared structures: bridges, rail crossings, highway rights-of-way and utility corridors where routes may merge.
- Network hubs: connections that terminate in different aggregation points, not two circuits into the same hub.
Route maps should be geographic, detailed enough to show street-level paths, and refreshed when carriers reroute or regroom circuits. Diversity that existed at commissioning can erode over time as networks change, so it should be rechecked periodically rather than assumed.
Inside the fence line
Diversity does not stop at the property line. Many facilities that have two separate outside routes bring both into the same corner of the site, run them through the same duct bank to the building and land them in a single entrance room. A fire, a flood or a construction accident in that one area can undo the diversity that was so carefully arranged outside.
Good campus design treats fiber entries the way it treats utility power feeds: as independent paths that should not share a failure point. That means separate vaults at the property line, separate duct banks across the site, separate building entries and, inside the building, separate pathways to at least two meet-me rooms. It also means protecting those pathways during later construction, when a new building or utility trench can cut through what used to be a clean route.
Carrier-neutral meet-me rooms matter here as well. A facility that welcomes many carriers into well-designed, secure interconnection space gives its tenants choice and makes true diversity easier to achieve, because more providers bring more potential routes.
Capacity grows faster than expected
Data center connectivity needs tend to grow faster than planned, especially as workloads such as AI training and inference move large volumes of data between sites. Laying conduit with spare capacity, and fiber with spare strands, costs far less during construction than adding it later.
Most of the cost of a fiber build is in the civil work: trenching, boring, permitting and restoring the surface. The fiber itself is a smaller share. Once a trench is open, adding extra conduits and higher strand counts adds relatively little. Reopening the street years later to add capacity means paying for the civil work again, with new permits and new disruption.
Consider a hypothetical campus planned in phases. The first building needs modest connectivity. By the third phase, the campus may need connections to several other facilities, multiple cloud on-ramps and dedicated links for workloads that move data between sites. If the original duct bank was sized only for phase one, every later phase starts with a construction project. If it was sized for the full campus, later phases start with pulling fiber through conduit that is already there.
Workload patterns reinforce the point. Training large models can spread across multiple buildings or sites, and the links between them carry heavy, sustained traffic. Inference workloads push for low latency to users and to other facilities. Both trends mean that the network between buildings, and between campuses, is becoming as important as the network to the internet at large.
Laying conduit with spare capacity, and fiber with spare strands, costs far less during construction than adding it later.
Lit, dark and owned
Operators have several ways to get connectivity. Lit services are bought from carriers as managed circuits at a defined bandwidth. They are simple to procure, but the buyer depends on the carrier's routes and equipment. Dark fiber is leased as unlit strands that the buyer lights with its own equipment, giving more control over capacity and upgrades. Owned fiber, built or acquired outright, offers the most control and the most responsibility.
The right mix depends on the facility's scale, its tenants and how long the operator expects to hold it. Many facilities combine all three: carrier-neutral meet-me rooms with lit services from several providers, dark fiber to key interconnection points and owned laterals that connect the building to nearby routes. What matters is that the mix is chosen deliberately, with diversity and growth in mind.
Documentation and monitoring
Fiber is long-lived infrastructure, and the people who installed it are often long gone when it fails. Accurate as-built records, splice documentation and continuous monitoring make it possible to locate and repair a fault quickly. Without them, a cut that should take hours to fix can take days.
Good records show where each cable runs, where each splice is located, which strands are in use and what they serve. Monitoring tools can measure the distance along a fiber to a break or a degraded splice, but that measurement is only useful if it can be matched to a physical location. A crew that knows exactly where to dig restores service far faster than one that has to search.
Records also have to be kept current. Every reroute, splice and new lateral changes the picture. An operator that treats documentation as a one-time deliverable at the end of construction will find, a few years later, that its records no longer match the ground.
Questions to ask a fiber provider
Whether buying lit services, leasing dark fiber or contracting for a build, an operator should expect clear answers to a few basic questions:
- Can you provide a geographic route map, end to end, and update it when the route changes?
- Where does this route share conduit, structures or rights-of-way with other carriers?
- How many spare strands and spare conduits are available along the route?
- How is the route monitored, and what are the response and repair commitments?
- Who holds the as-built and splice records, and how often are they updated?
A provider that cannot answer these questions clearly is telling the buyer something important about how the route will be supported when it fails.
Common mistakes
The same errors appear again and again in data center connectivity. Accepting a carrier's assurance of diversity without a route map. Bringing two circuits into the building through a single entry. Sizing conduit for opening day rather than for the life of the campus. Leaving documentation to the installer and never updating it. Treating fiber as a procurement line item rather than as infrastructure that deserves engineering attention.
None of these mistakes is expensive to avoid at the start. All of them are expensive to fix later, usually at the worst possible moment.
Where Span and Keystone meet
UpManagement Span builds and operates fiber serving data centers, commercial buildings and residential communities. Because Span sits in the same group as UpManagement Keystone, data center projects can plan fiber routes, building entries and capacity together with power and cooling, from the first site study. That coordination is one of the practical benefits of owning both kinds of infrastructure under one roof.
In practice, that means diversity is designed in, not verified after the fact. Conduit is sized for the campus, not the first building. Records are maintained as part of operations, and Span monitors its network around the clock. Both companies report against the same UpManagement standard each month, including uptime, so connectivity is held to the same expectations as the power and cooling it supports.
Published by UpManagement. This article is general information about the sectors in which the UpManagement companies operate. It is not legal, tax, investment or compliance advice, and it is not an offer to sell or a solicitation of an offer to buy any security. Where a company is described as a platform in formation, its approvals are not yet in place.
