Inside a data-centre hall, most links are short — a few metres of copper or fiber between a server and its switch. Three technologies compete for exactly those links: DAC assemblies, AOC assemblies, and the classic pairing of pluggable transceivers with patch cords. They deliver the same bits; they differ in almost everything else. This comparison is the one our engineers work through with buyers, and it mirrors how the GRT catalogue itself is organised across DAC cables, AOC cables and the transceiver families.
What each one physically is
A DAC (direct attach copper) is a fixed twinaxial copper cable with transceiver-style connectors factory-fitted at both ends. An AOC (active optical cable) is the same idea with fiber in the middle and the optoelectronics sealed into the connector shells. The transceiver route splits the assembly into parts: two pluggable modules plus a separate patch cord, each independently replaceable, the cord re-routable to any length your stock covers. DAC and AOC are single factory units; the transceiver link is a small system.
Cost: the honest arithmetic
Per link, DAC is cheapest — one assembly, no modules. AOC costs more per link than DAC and less than a two-module optical link at the same short distance. But per-link price is not project cost. Transceiver links reuse their modules across re-cabling generations and absorb length changes from stock; DAC and AOC are bought per exact length and become scrap when the layout changes. A one-off row build favours assemblies; a hall that re-cables quarterly may find the "expensive" optical route cheaper across three years. Price the churn, not just the link.
Power, bulk and reach
DAC draws the least power (the link is passive copper) but its cable is the thickest and its practical reach the shortest — a few metres before the bundle becomes an airflow problem. AOC draws a little power at each end, is fiber-thin, and reaches comfortably across a hall. Pluggable optics draw the most power per end and reach kilometres — grotesque overkill at 3 metres, and the only option at 3 kilometres. The reach ladder is the cleanest summary: DAC owns the rack, AOC owns the row and the hall, optics own everything beyond.
Failure and operations
When a DAC or AOC fails, you replace the whole assembly — diagnosis is "swap it". When an optical link fails, you can isolate module, cord or host port independently, which is faster to diagnose but needs spares in three categories instead of one. Operational tempo matters too: assemblies install fastest (one part, no cleaning), while optical links demand connector hygiene discipline. None of this is decisive alone; together with distance it usually is.
The decision in one table of rules
- Same rack or adjacent racks, fixed layout → DAC.
- Across a row or hall, up to a few tens of metres, high EMI or tight trays → AOC.
- Anything longer, anything that must re-route, anything mixing media → transceivers plus patch cords.
- Migration in progress (rates changing per port) → transceivers; assemblies freeze the rate at purchase.
Most real designs mix all three by layer: DAC inside racks, AOC to aggregation, optics upward. Send us the rack layout and we will return the per-link schedule with the mix costed — that schedule, not a unit price, is the document worth comparing between suppliers.
Are DAC/AOC coded like modules?
Yes — the connector carries the same identity memory, and hosts can be selective. Order assemblies against your platform list.
Which is best for 25G server links?
The same ladder applies at 25G: DAC in-rack, AOC across the row, SFP28 optics beyond. See the 25G transceiver page for the optical side.
The sparing and spares question
Each technology fails differently, and the spares shelf should reflect that. A failed DAC or AOC is replaced whole: stock complete assemblies at your deployed lengths, in a ratio that matches your link count and your tolerance for a dark port awaiting delivery. A failed optical link decomposes: module at end A, module at end B, or the cord between — the spares shelf holds a few modules per type and cords per length, and a technician can restore the link by swapping halves until it lights. Organisations with strong field discipline often find the optical route’s diagnosability worth its premium at surprisingly short distances; organisations that prize install speed and zero cleaning ritual choose assemblies and accept the whole-unit swap. Neither is wrong. The error is choosing without noticing you have chosen.
Which option is best for a one-rack build that will never change?
DAC, almost without argument: fixed layout, shortest distances, lowest cost and power. The flexibility you are not buying is flexibility you will never use.
Length discipline: the unglamorous decider
Walk a mature data hall and read its cabling like a text. The DAC rows are uniform: racks built as units, switches at a fixed position, lengths ordered to a schedule and dressed identically — the technology rewards sameness. The optical rows are various: lengths from stock bins, routes that wander as layouts evolved, the archaeology of a hall that changed its mind repeatedly and could afford to. That is the deepest practical difference between the families. Assemblies freeze a decision at purchase; modular optics keep the decision open at a per-link premium. Facilities with stable, repeatable rack designs should absolutely harvest the assembly discount. Facilities whose layouts churn — cloud builders, research halls, colocation with rotating tenants — should price the openness honestly and will usually find optics cheaper across the churn. The technology question was never really about copper or glass. It is about how often your building changes its mind.
A decision in one paragraph
Inside one rack with a settled layout, buy DAC and spend the savings elsewhere. Between racks or between rooms, buy optical modules and cords, whose reach and flexibility match the geography. Across any path that may be re-routed, lengthened or re-purposed, AOC earns consideration where its fixed assembly is acceptable and its electronics-free run is an asset. Then standardise: one supplier discipline, stocked spares per deployed type, recorded pairings. The decision per link takes minutes; the discipline behind it is what the next decade of maintenance will thank you for.
A final practical note on stocking: assemblies should be stocked by exact length and connector family; optical modules by type and coding platform; cords by discipline and length family. Three small shelves, three labelling conventions — and a spares audit once a year against the as-built records. The shelves are cheap. Finding them empty at the wrong moment is not.