Structured cabling: the layer everything else runs on
It’s the cheapest part of a data centre to install and the most expensive to get wrong. The cabling plant outlives three or more generations of the kit plugged into it — so the choices made at install decide, for a decade, how fast you can go, how easily you can change, and how much of a network fault you’ll spend your life chasing. This is a technical guide to the physical layer, with interactive tools to explore reach and the speed roadmap.
“Just Cables”? The Layer Everything Depends On
Structured cabling is a standards-based, hierarchical system of cables, connectors and pathways — designed once, as a whole, so that any device can be connected to any other through a predictable patching scheme. The opposite is point-to-point sprawl: a new run for every need, no documentation, no headroom. The difference doesn’t show on day one. It shows on every change, every fault and every upgrade for the next ten years.
It outlives the kit
Servers and switches are refreshed every 3–5 years; the cabling and pathways behind them often last 15–25. You’re not cabling for today’s hardware — you’re cabling for three generations you haven’t bought yet.
It’s where faults hide
A large share of network problems are physical: a marginal connector, a kinked fibre, an untested run, the wrong patch lead. They’re intermittent, hard to find, and they masquerade as “the network being slow.”
Cheap to lay, costly to redo
Cabling is a small fraction of build cost. Ripping it out of a live hall — or working around a mess for years — is not. The economics overwhelmingly favour doing it properly, once.
It sets your speed ceiling
The media you install fixes how fast you can run and how far. Put the wrong fibre or copper in, and the upgrade you wanted in year four needs a re-cable, not a transceiver swap.
Inside a Structured Cabling System
The standards — TIA-942 in North America, ISO/IEC 11801-5 internationally — define a hierarchy of spaces and the cabling that links them. Traffic flows from the carriers at the edge of the building, through distribution areas, down to the equipment in the cabinet. Two cabling types do the work: backbone (between distribution areas, almost always fibre) and horizontal (the final run to the equipment).
Backbone vs horizontal
Backbone links the distribution areas (MDA↔HDA) and carries aggregated traffic — high-count fibre, future-proofed for the next speed. Horizontal is the last hop to the equipment, where copper or fibre meets the server.
Patch panels & the cross-connect
Permanent cabling terminates on patch panels; short patch leads make the actual connections. Moves and changes happen on the patching, never by re-pulling permanent cable — that’s the whole point of “structured.”
Pathways & containment
Tray, basket, ladder and underfloor/overhead routes keep cable supported, separated (power vs data) and accessible. Bad pathways are where bend-radius and congestion problems are born.
Copper, Fibre & How Far They Reach
Three media do almost all the work in a modern hall: balanced copper (twisted-pair, RJ45), twinax DAC (direct-attach copper, for in-rack), and optical fibre — multimode (MMF) for short reach, single-mode (SMF) for everything longer. The right choice is a trade of distance, speed, power and cost.
| Media | Typical use | Reach | Notes |
|---|---|---|---|
| Cat6A copper | Horizontal to server, 10GBASE-T | 100 m | Cheap, RJ45, no transceiver — but power-hungry and capped at 10G in practice |
| Cat8 copper | Top-of-rack, 25/40GBASE-T | ~30 m | Short-reach only; largely superseded by DAC in the rack |
| Twinax DAC | Server↔ToR switch, in-rack | 1–7 m | Lowest cost and power for short hops; passive or active |
| OM4 multimode | Intra-row / short backbone | ~100 m @ 100G+ | Cost-effective optics (VCSEL); reach shrinks as speed rises |
| OM5 multimode | Short reach, SWDM | ~100–150 m | Wideband MMF for shortwave WDM; modest gain over OM4 |
| OS2 single-mode | Backbone, inter-room, campus | 500 m – 10 km+ | The future-proof choice — the fibre rarely limits the speed |
The headline rule: multimode is cheaper per link but limited and shrinking in reach as speeds climb; single-mode costs a little more in optics but is effectively speed-proof. Many operators now default to single-mode for backbone precisely so the cabling never becomes the reason they can’t upgrade.
Pick a link speed, then drag the distance. The four media light up where they can carry that speed — and grey out where they can’t.
How the Racks Are Cabled
Where you put the switches decides how much cable you pull, how you patch, and how you cool. The three classic patterns each trade cabling volume against flexibility and switch count.
Top of Rack
A switch (or pair) in every cabinet; servers connect with short DAC. Minimal copper, simple per-rack containment — but more switches to manage and more uplink fibre. The default for dense, modern halls.
End of Row
Aggregation switches at the row end; every server homes back to them. Fewer, larger switches — but long horizontal runs and heavy cable bundles down the row.
Middle of Row
The EoR idea with the switch mid-row, halving the worst-case cable length. A pragmatic compromise between cable volume and switch count.
AI and high-density compute have pushed the industry firmly toward ToR with single-mode backbone: keep the high-speed copper hops tiny (server↔ToR over DAC), and run fibre — lots of it — back to the spine. Which is exactly where the speed roadmap gets interesting.
From 10G to 1.6T
Ethernet speed has climbed by roughly 10× a decade — and the way it gets there has changed. Early jumps added lanes (four 10G lanes made 40G). Modern jumps come from faster lanes: NRZ gave way to PAM4 (two bits per symbol), lane rates went 10→25→50→100→200G, and reach over copper and multimode shrank every step. Drag through it:
Drag the slider through two decades of Ethernet. Watch the lane structure, optics and how far each medium reaches.
Notice the pattern as you climb: copper drops out entirely above 40G, multimode reach collapses toward the length of a single row, and single-mode quietly carries on. That’s the structural reason high-speed halls are going single-mode — the fibre you lay today still works at the next two speeds.
Faster, Denser, Closer to the Chip
The drivers are AI clusters and the insatiable east-west traffic between GPUs. 800G is shipping in volume; 1.6T (IEEE 802.3dj) is being standardised now, and 3.2T is on the roadmap. But you can’t just keep adding lanes — power and density force the optics themselves to change. These are the shifts a cabling plant laid today should anticipate.
200G-per-lane (224G SerDes)
The leap behind 1.6T: doubling the per-lane rate to ~200G (224G raw, PAM4). Fewer fibres for the same bandwidth — but far tighter signal-integrity and loss budgets, so cleaner, better-tested cabling matters more.
Linear-drive optics (LPO)
Stripping the DSP out of the transceiver to cut power and latency. A big deal for AI fabrics where tens of thousands of links make every watt and nanosecond count.
Co-packaged optics (CPO)
Moving the optics onto the same package as the switch ASIC, replacing power-hungry electrical traces. It pushes fibre right up to the silicon — and changes how the front of the rack is cabled.
Hollow-core fibre (HCF)
Light travels ~50% faster through air than glass. Hollow-core cuts latency and is moving from lab to early deployment — latency-sensitive AI and trading workloads are the first adopters.
Single-mode everywhere
As multimode reach shrinks with each speed, the industry is standardising on parallel single-mode (DR) for the data hall — the most future-proof bet you can make at install time.
Higher-count, pre-terminated
Base-8/base-16 MPO trunks, pre-tested and ready to plug, are replacing field termination — faster to deploy, denser, and cleaner to certify at AI-fabric scale.
The AI angle — coming next in this series
AI/GPU data halls turn all of this up to eleven: 800G–1.6T fabrics, racks drawing 40–130kW, and cable volumes that make containment and airflow a first-class design problem. Our next guide, Planning an AI/GPU-ready data hall, goes deep on 800GbE networking and the physical design that supports it — building directly on the fundamentals here.
What Separates a Good Install From a Liability
The media choice is the easy part. What makes a cabling plant a pleasure to operate — or a decade-long headache — is the discipline around it. Score an existing install honestly against the things that actually bite:
Labelling and certification sit at the top for a reason: an unlabelled, untested plant is one where every future change is a guess. A proper install is certified — every link tested to standard with a saved result — and documented, so the person who works on it in five years isn’t reverse-engineering your decisions in the dark. Certification is also what unlocks a manufacturer’s long-term system warranty — no certified test results, no warranty.
How Optronix Helps
Structured cabling is the physical foundation under everything else we do — and it starts with design. We design, install, test, certify and warrant cabling systems to TIA/ISO standard, across the UK, Europe and worldwide — from a single row expansion to a greenfield hall, or untangling a plant that grew without a plan. Because the design is right and the install is certified, the result can be backed for 25 years.
Design pedigree at hyperscale
Our team has designed structured cabling systems for some of the largest hyperscale data centres in the world — campuses where a single design has to scale to hundreds of thousands of links and survive multiple speed generations. That same discipline goes into every project, whatever the size: media and topology chosen for where you’re heading, counts and pathways sized for growth, and a documented scheme that’s a pleasure to operate.
Backed by a 25-year manufacturer warranty
Optronix is a certified installation partner of the major cabling manufacturers — so the systems we design and install can carry their 25-year manufacturer system warranty, covering components, performance and applications assurance. That cover is only available through accredited installers, on properly designed, installed and certified plant — exactly what we deliver.
Design & specify
Hyperscale-grade design at any scale: media, topology, counts and pathways sized for the speeds you’ll run — not just today’s — with headroom to upgrade on a transceiver swap, not a re-cable.
Install & terminate
Copper and fibre, pre-terminated trunks, patch fields and containment — installed cleanly, to standard, with the cable management that keeps a hall workable.
Test & certify
Every link tested and certified (Fluke), with results documented — so the plant is provably to spec on day one and audit-ready after.
Document & manage
Labelling, as-built records and DCIM-ready documentation, plus ongoing moves, adds and changes as part of operations.
Building, expanding, or untangling a hall?
Tell us what you’re running now and where you want to get to, and we’ll design and install a cabling plant that gets you there — tested, certified, documented and backed by a 25-year manufacturer warranty.
Sources & notes
- TIA-942 (Telecommunications Infrastructure Standard for Data Centers) and ISO/IEC 11801-5 — the structured-cabling hierarchy, spaces and media classes
- IEEE 802.3 Ethernet standards — including 802.3df (800GbE, 2024) and 802.3dj (1.6TbE, in development); media reaches are the standardised PMD figures
- Ethernet Alliance Technology Roadmap — speed timeline and the path to 1.6T / 3.2T and 200G-per-lane signalling
- General industry practice for topologies (ToR/EoR/MoR), DAC/optics selection, and emerging technologies (LPO, CPO, hollow-core fibre)
General guidance, not a design specification. Reaches are the standardised maxima for the relevant Ethernet PMD and cabling class; achievable distance on a real link depends on insertion loss, connectors and installation quality, and standards evolve. Confirm current figures against the relevant standard and a tested channel. Interactive figures are rounded for clarity.