400ZR Explained: 400 Gigabits in a Pluggable, No Transponder Shelf

The short answer

What is 400ZR in simple terms? 400ZR is a way to send 400 gigabits per second between buildings — typically two data centers — on a single colour of light, using a small pluggable module that sits directly in a router or switch port. The module contains the same kind of sophisticated long-distance optical technology that used to need its own separate shelf of equipment, so on the links it fits, that shelf is simply gone. If you remember one sentence, make it that one: 400ZR did not invent a faster signal so much as it shrank a room-sized idea into a port.

Think of the old way as air freight. Your router packs a parcel and hands it to a separate freight company in its own building — the transponder shelf — which repacks it into a special long-haul container, flies it across the country, and unpacks it at the other end into a parcel the far router understands. The freight company is excellent at its job, but you pay for the building, the power, the staff and the repacking, on every route, forever. 400ZR is the moment the long-haul container shrinks to a kit that bolts onto your own delivery van. The van drives the same highway the freight plane flew — the same fiber, the same amplifiers along the way — but the separate building, and the repacking, disappear. What did not disappear is the highway itself: the amplifiers and wavelength equipment along the fiber are still there, and this article will be careful about exactly which box 400ZR removes and which boxes it does not.

The problem it solved: the shelf between the router and the fiber

Ordinary data-center optics are sprinters. The grey (single-colour, short-reach) modules inside a building flash light on and off — bright means one, dark means zero — and a receiver simply decides which it saw. That is cheap and beautifully simple, and it works until distance, or a crowded fiber carrying many colours at once, smears and fades the flashes until the receiver can no longer tell them apart. Long-haul networks solved that problem decades ago with coherent transmission, which this article explains next — but coherent equipment was big, power-hungry and sold as dedicated transport shelves. So a data-center interconnect (a link between two data centers, often in the same metro area) traditionally looked like this: router, short grey link, transponder shelf, long coherent line, transponder shelf, short grey link, router. Two translations, two extra boxes to buy, power, cool and manage — per link.

Meanwhile the routers themselves grew coherent-sized appetites. Once switch and router ports reached 400 gigabits, the obvious question was why the coherent engine could not live in the port, the way every other optic does. The honest answer was heat and size: a coherent DSP plus a tunable laser is a lot to fit in a pluggable's power envelope. By the late 2010s the chips had shrunk enough — and the industry did the thing that turns a trick into a market: it wrote one fixed, open recipe so every vendor's module would talk to every other vendor's module. That recipe is 400ZR.

Coherent, in plain English: describing the wave, not flashing it

Here is the one idea the whole article hangs on. A sprinter optic sends information the way you would signal with a torch: on, off, on, off. A coherent optic sends it the way a skilled musician plays a note: not just loud or silent, but loud and with a precise timing character — and it changes both, many billions of times per second, in combinations that each stand for several bits at once. Light has two independent directions of wiggle (called polarizations), and coherent sending uses both at once, doubling the cargo again. The receiver does not just detect brightness; it mixes the incoming light with its own laser and reconstructs the full shape of the wave, which is why coherent receivers can hear signals far too faint and smeared for a torch-style receiver to read.

The smearing matters as much as the faintness. Glass slows different colours by slightly different amounts, so over tens of kilometres a sharp pulse spreads into its neighbours (engineers call this chromatic dispersion), and the two polarizations drift apart as well. A torch receiver cannot un-spread a pulse. A coherent receiver hands the whole wave to its DSP, which knows the maths of how fiber spreads it and runs that maths backwards — un-smearing the signal in arithmetic instead of optics. That is the real trick inside every 400ZR module: not a stronger torch, but a receiver that reconstructs what was sent from a waveform that no longer looks like it. The price is power and heat — that maths, at 400 gigabits, is why these modules draw roughly 15–18 watts where a grey optic draws a fraction of that — and the reward is reach measured in tens of kilometres per wavelength instead of hundreds of metres.

What the standard actually fixes

400ZR is an Implementation Agreement from the Optical Internetworking Forum (OIF), released in 2020 and mirrored by the IEEE as 802.3ct. Its power is not any single number — it is that the numbers are fixed. One rate, one modulation, one error-correction scheme, one power class: a module built to it by one vendor interoperates with another vendor's module, in a third vendor's router. The recipe, in plain terms:

Notice what the table is really selling: boring. Metro interconnect is a volume business, and volume follows interoperability. A fixed recipe means a buyer can price three vendors' modules against each other, stock one spare for any link, and never run a translation shelf whose configuration is a snowflake. The same fixedness is also 400ZR's boundary — the moment a link needs something the recipe does not allow, you have left 400ZR's country, and the honest options are its extended-reach relatives or a traditional line system, not a heroic tweak.

Reach is a budget: OSNR and dispersion, without fear

Every 400ZR datasheet leads with distance, and distance is the least honest number on it — not because vendors lie, but because kilometres do not damage signals; noise and smearing do, and they merely correlate with kilometres. Two budgets decide whether a real link works, and both fit in a paragraph each.

Budget one: OSNR — is the signal still above the hiss? Every amplifier on a fiber line boosts the signal and adds a little hiss of its own (spontaneous emission, if you want the term). Boost, hiss, boost, hiss: the signal stays the same height while the floor rises under it, span by span. OSNR is the height of the signal above that floor. 16QAM's sixteen shapes sit close together, so telling them apart needs a clear margin above the noise — the 400ZR preset in this site's reach planner requires 26 dB of OSNR (measured in the standard 0.1 nm reference slice). Amplifiers combine in a way that punishes long chains: ten identical spans that each deliver a healthy 35 dB on their own cascade to just 25 dB end to end — a fail, by 1 dB, before any other penalty is counted. That single piece of arithmetic is most of optical engineering: per-span numbers always look fine; the cascade is where links die.

Budget two: dispersion — can the DSP still un-smear it? The DSP can reverse chromatic dispersion only up to a rated total, expressed in picoseconds per nanometre. Ordinary G.652 fiber adds about 17 of those units per kilometre, so the budget is spent by distance directly: 80 km spends about 1,360 of the 2,400 ps/nm the planner's 400ZR preset allows — comfortable. Stretch the same design to 400 km and dispersion alone demands 6,800 — nearly three times the budget — on a route where you might still be admiring the OSNR. A second, smaller smearing effect (polarization-mode dispersion) grows with the square root of distance and rarely binds first on modern fiber; the planner checks it anyway, because budgets you do not check are the ones that bind.

One worked metro link, with the planner's numbers

Two data centers, 80 km apart, one amplified span of G.652 fiber between them — the classic 400ZR use case. Run the three checks the free planner on this site runs. Dispersion: 17 × 80 km ≈ 1,360 ps/nm against a 2,400 tolerance — pass, with room to spare. Polarization smearing: about 0.9 ps against a 10 ps tolerance — pass easily. OSNR: a healthy single amplified span delivering 29 dB against the 26 dB requirement — pass, with exactly the +3 dB margin the planner treats as a sound design target. Verdict: plain 400ZR, interoperable modules from any vendor, no shelf. This is the link the standard was drawn around, and it is genuinely this straightforward.

Now stretch the same recipe to a 400 km regional route — five 80 km spans — and watch two budgets fail independently. Dispersion reaches 6,800 ps/nm, far past 2,400. And even if each span is individually strong, cascaded OSNR erodes span by span toward the requirement line and below it once real penalties land. Nothing is broken; the link has simply left the fixed recipe's envelope. The step up is OpenZR+: its representative 400G mode in the same planner tolerates 12,000 ps/nm of dispersion and needs 24 dB of OSNR — both budgets pass on this route. Same pluggable idea, stronger correction maths and a louder launch, in exchange for more watts and mode-matching homework. That two-line comparison — which budget binds, and which module class clears it — is the whole of ZR selection, and it never appears in the kilometre column.

400ZR vs ZR+ vs OpenZR+: three names, one family argument

The decision rule that falls out of the table: inside ~120 km, point to point, on a line you control — plain 400ZR, and enjoy the boring. Beyond that distance, through wavelength-switching (ROADM) networks, or where one module must serve several rates, you are shopping in OpenZR+ territory and the datasheet pair, not the brochure family name, is what must match. And one rung further out, traditional transponder and line-system builds still own the longest, densest routes — 400ZR's family moved the boundary; it did not abolish it. What is next is already standardised, too: the OIF's 800ZR doubles the rate per wavelength against the same ~120 km target, which is why faceplates and power budgets bought today are worth checking against it.

Five mistakes buyers actually make

1. Reading "120 km" as a promise. It is the size of an envelope defined by OSNR, dispersion and line penalties together. An 80 km link on old, high-loss fiber with tired amplifiers can fail where a clean 110 km link passes. Budget the link; do not measure it with a map.

2. Forgetting the line system still exists. 400ZR removes the transponder, not the amplifiers, multiplexers or filters that put many wavelengths on one fiber. A single unamplified wavelength (about 40 km, loss-budget permitting) is the only truly shelf-free case — everything longer still rides a line system someone must design and power.

3. Filling a line card without a power and heat check. Fifteen-plus watts per port, times a full faceplate, is a real thermal design point that older chassis were never rated for. The module fitting the cage is the beginning of the question, not the end.

4. Mixing family names at the two ends. A 400ZR module and an OpenZR+ module set to a different mode are two radios speaking different dialects on the same frequency. Standard ZR-to-ZR interworks by design; anything fancier must be verified mode by mode, ideally in writing from both vendors.

5. Buying ZR for a link that never leaves the building. Across a hall or a campus, grey optics and their fraction-of-the-power, fraction-of-the-price economics win, and a coherent DSP is a heater solving a problem you do not have. Reach classes exist to be matched to links, not maximised.

Try it yourself

Do not take the worked example's word for any of this — break it. The free coherent DWDM / 400ZR reach planner runs exactly the budgets above on your own route: set the spans, fiber type, launch power and amplifier noise figure, and it cascades the OSNR span by span, draws the margin waterfall, checks dispersion and PMD against the selected module, and gives the straight verdict — 400ZR works, step up to OpenZR+, or redesign the route. Try the 80 km link, then stretch it span by span and watch which budget binds first; that moment, not the kilometre count, is the lesson. For the short grey links at either end of the coherent span, the optical link budget calculator does the simpler launch-minus-losses arithmetic, and if the ports at either end need breaking out or consolidating, the 400G/800G port breakout planner covers the faceplate side of the same design.