Data Center Cabling: The Complete Guide

Data center cabling determines more than which port plugs into which port. It sets the ceiling on how fast a facility can move data, how much it costs to run, and how easily it scales when the next generation of switches and servers arrives.

A cabling decision made today, copper or fiber, structured or point-to-point, still has to work when 800G and 1.6T equipment lands on the same racks.

This guide covers what data center cabling actually is, how structured cabling differs from the unstructured approach it replaces, when copper makes more sense than fiber, and the best practices that keep a cabling system reliable for the long haul.

What is data center cabling?

Data center cabling is the physical layer connecting servers, switches, storage arrays, and other network equipment inside a facility. It includes the cables themselves along with the connectors, patch panels, racks, and pathways that hold them in place.

Every other layer of a data center's network depends on this layer working correctly. A switch with the right firmware and a server with the right NIC still can't communicate if the cable between them is the wrong category, run past its maximum length, or bent tighter than its minimum radius allows.

Cabling decisions get made early in a build-out or refresh, and they're expensive to unwind once racks are populated and cable trays are full.

Compute and storage tend to get most of the attention in a data center budget. A new server or switch is a visible line item; cabling rarely is. But cabling is the layer that determines whether that new equipment can actually run at the speed it was purchased for, which is why the design decisions covered in this guide are worth the same scrutiny as the hardware they connect.

Structured vs. unstructured cabling: what's the difference?

Structured cabling organizes a data center's connections into a standardized system with three subsystems.

These systems involve backbone (or vertical) cabling connecting the main distribution area to horizontal distribution areas, horizontal cabling running from those distribution areas to the racks and equipment in each zone, and work area cabling connecting equipment to the nearest outlet or patch panel. Each subsystem uses defined pathways, labeling conventions, and patch panels rather than ad hoc runs.

Unstructured cabling connects devices directly to each other, port to port, with no defined pathway or intermediate patch point. It's faster to deploy for a handful of connections and requires less upfront planning.

Past a few dozen connections, unstructured cabling becomes difficult to trace, document, or change without disrupting adjacent links, a pattern commonly called spaghetti cabling.

Structured cabling costs more to design and install upfront. In a facility with more than a few racks, that cost is usually recovered through lower maintenance time, faster fault isolation, and the ability to add or move equipment without touching unrelated connections.

Copper vs. fiber: choosing the right cabling media

Copper cabling, most commonly twinaxial direct attach cable (DAC) or Category 6A and Category 8 twisted pair, carries an electrical signal end to end with no optical conversion. That keeps cost and power draw low and adds virtually no latency, but it limits reach.

Passive DAC is generally reliable up to about 2.5 meters, which is why it's the default choice for connections within a single rack.

Fiber cabling converts the signal to light, which extends reach well beyond copper and removes the electromagnetic interference issues copper is prone to in dense racks. Multimode fiber (OM3, OM4, OM5) is the common choice for shorter runs within a data center.

OM4, for example, supports 100G transmission over distances up to roughly 150 meters. Single-mode fiber (OS2) supports the longest runs, extending to multiple kilometers, and is the standard choice for connections between buildings or data center campuses.

The trade-off is straightforward. Copper is cheaper and lower-power for short runs. Fiber becomes necessary once distance, density, or interference make copper impractical. Most data centers use both, matching the media to the specific run rather than standardizing on one.

Data center cabling standards you need to know

Several standards bodies define how data center cabling should be designed, installed, and tested. Knowing which standard applies and which edition is current matters more than most cabling guides acknowledge. Several of these standards were updated within the last two years specifically to account for the density and power demands of AI infrastructure.

ANSI/TIA-942-C, published in May 2024, is the current edition of the Telecommunications Infrastructure Standard for Data Centers. It replaced the July 2017 -B edition and added provisions covering micro edge data centers, immersion cooling systems, and balanced single twisted-pair cabling as a recognized horizontal media type.

ANSI standards are reviewed on a five-year cycle, and the -C revision was shaped in part by the same shift toward higher-density, higher-power racks that is driving AI and GPU cluster deployments.

The standard also defines four infrastructure ratings, Rating 1 through Rating 4, describing the level of redundancy built into a facility's power, cooling, and cabling pathways. Rating 4 requires fully fault-tolerant, concurrently maintainable infrastructure, including redundant cabling pathways that can each independently support the full facility load without an outage during maintenance.

Most enterprise data centers target Rating 2 or Rating 3, reserving Rating 4 for facilities where downtime carries a direct financial or safety cost.

ANSI/TIA-568 governs commercial building telecommunications cabling generally, including the category ratings (Cat 6A, Cat 8) and performance requirements that TIA-942-C references for horizontal cabling.

ISO/IEC 11801 is the international equivalent of TIA-568, used outside North America and by multinational organizations that need a single specification across regions.

ANSI/BICSI-002, currently in its 2024 edition, covers data center design and implementation more broadly, including the pathways, spaces, bonding, and grounding requirements that sit alongside the cabling-specific standards above.

Data center cabling best practices

• Plan pathways before racks are populated. Cable trays, conduits, and patch panel locations are far cheaper to change on paper than after equipment is installed.

• Respect minimum bend radius. Most fiber cabling shouldn't be bent tighter than roughly ten times its outer diameter. Tighter bends increase signal loss and can cause failures that don't show up until months later.

• Separate power and data runs. Running network cabling parallel to and in contact with power cabling introduces electromagnetic interference, particularly on unshielded copper runs.

• Label every run at both ends. Consistent labeling, tied to the naming convention in ANSI/TIA-942-C, turns troubleshooting from a multi-hour search into a five-minute lookup.

• Test and certify before go-live. Field testing against the applicable standard's transmission requirements catches marginal links before they become production outages. AddOn verifies cabling and transceivers in its own in-environment testing lab before they ship.

• Bond and ground cabling infrastructure correctly. Racks, cable trays, and patch panels should all tie into the facility's bonding and grounding system. Skipping this step is a common source of intermittent, hard-to-diagnose signal issues.

• Document as you go. A cabling record updated at install time stays accurate. One reconstructed after the fact rarely does.

• Size for the next refresh, not just current load. Cabling infrastructure typically outlasts two or three generations of switches and servers. Provisioning pathway capacity for growth avoids a second disruptive install cycle.

How AI and high-density workloads are changing cabling requirements

AI training and inference clusters generate a different traffic pattern than a typical enterprise network. GPU clusters produce heavy east-west traffic between nodes, and a single failed or high-latency link can stall an entire training run rather than affect one isolated workload. That has pushed data center operators toward choosing interconnects deliberately, rather than defaulting to whatever was used in the last build-out.

AddOn's guide to building for AI infrastructure covers the transceiver side of this shift in more depth.

The pattern that has emerged across large deployments is copper inside the rack and optics outside it. Passive DAC remains the preferred choice for intra-rack GPU-to-switch connections because of its reliability and near-zero added latency. Active electrical cables (AEC) and active optical cables (AOC) handle the 3 to 7-meter rack-to-rack distances where passive copper's reach runs out.

This shift is accelerating the move to 800G optics now, with 1.6T deployments expected to follow at a faster generational pace than previous upgrades. For teams planning data center infrastructure today, the practical implication is ordering cabling and optical components well ahead of compute hardware arrival, since lead times on both have grown alongside demand.

How to choose the right data center cabling solution

Start with two numbers: the bandwidth each connection needs today and the bandwidth it's likely to need at the next refresh. Cabling infrastructure is expensive to replace and cheap to over-provision at install time, which makes it one of the few places in a data center build where slight over-specification pays for itself.

AddOn's network cabling range spans fiber and copper patch cables, DACs, and dual-OEM cables built to cover these scenarios.

From there, match reach to media type. If every run stays within a rack, copper DAC is usually the lower-cost, lower-power answer. Once runs cross racks or rows, fiber removes the interference and distance limitations that copper runs into.

Density matters too: high-density switching environments benefit from MPO-based fiber that reduces the number of individual runs needed for a given number of connections.

If the right choice isn't obvious from reach and bandwidth alone, that's normal. A multi-tenant colocation facility, a single-tenant enterprise data center, and an edge deployment each carry different pathway and redundancy requirements under ANSI/TIA-942-C, on top of the media decision itself.

AddOn's cable finder tool and connector tool, narrow, the options based on the equipment already in a rack, and AddOn's networking specialists are available to confirm a specification before it's ordered rather than after it's installed.

Why AddOn Networks?

• AddOn Networks has served the fiber and copper connectivity industries, including data centers, enterprises, service providers, and governments, for more than two decades.

• AddOn operates a world-class in-environment testing lab, verifying every single transceiver and cable under real network conditions before they ship.

• AddOn Networks ISO certifications provide independent verification that products and processes meet recognized quality benchmarks.

• A deep US and UK-based inventory means cabling and optics ship fast, without waiting on overseas lead times.

• Free expert support is available to confirm a cabling specification before it's ordered, reducing the risk of a mismatched or undersized deployment.