800G vs 400G Port Economics: Optics Selection Guide

What port economics actually covers

A port decision is a cost decision before it is a bandwidth decision. Moving a fabric from 400G to 800G per port changes what the operator spends on switch interfaces, on optics, on fibre and patch panels, on rack power and cooling, and on the spares held for every link type. Both rates usually live in the same build, because the answer is rarely the same on every tier.

This guide sets out that cost structure and the optics on each side of it, for planners comparing a 400G refresh against an 800G refresh.

Why the rate decision is not only about bandwidth

Doubling the rate per port cuts the number of switch ports needed for the same capacity, and with them the count of optics, cages and patch cords. The counterweights are equally real: a higher-rate module dissipates more power into the same rack unit, the optical content per port rises, and the cabling becomes more demanding, because a parallel 800G port usually terminates on an MPO or MTP assembly rather than a duplex pair. A 400G tier that is already cabled, powered and documented is often the cheaper half of a mixed build.

The rule most refreshes settle on: keep 400G as the volume tier while the port count per rack is not the binding limit, and move to 800G where panel density, rack unit count or spine capacity forces the change. Both families and their reach classes sit on the 800G transceiver page and the 400G QSFP-DD and OSFP page.

Recommended optics behind each rate

The reach column lists the standard classes behind each code; the working limit is the measured loss on the route. Grey parts are listed on the 100G QSFP28 page and among the rate families under optical communication products.

Where the money goes: ports, fibre, power, racks

Four line items move when the rate changes, and they rarely move in the same direction. Ports: fewer interfaces for the same capacity means fewer optics, cages and patch cords, so the per-bit cost of the interface falls. Fibre: a duplex pair behind a 400G LR4-class port cannot be pushed to 800G without parallel fibre or a coherent module, so on a built-out duct the cable becomes the limit. Power and cooling: a higher-rate module dissipates more per port, and a rack near its thermal limit may have to spread the same capacity over more line cards. Racks and floor space: fewer ports means fewer chassis and cable management rows, which is where the visible savings land. Three numbers decide the comparison: the capacity target, the ports and radix per rack unit, and the measured loss on the longest route.

Breakout, radix and panel density

Breakout lets both rates live in one build: an 800G cage can feed two 400G links, so a spine can be refreshed while the access tier stays at 400G or 100G until its own cycle. Two checks decide whether that is economic: the host must support the breakout mode and the split ratio, and the panel must have the ports and the polarity for the assembly. Compare radix per rack unit rather than the headline rate, since a rack with no room for another patch panel gains nothing from a faster port. A staged move of this kind is described in the 400G to 800G migration guide.

Lifecycle, sparing and coding

Every link type carries a spare and a documentation entry, at a cost per type rather than per link. Consolidating an estate onto fewer 800G types lowers the spares count; adding an 800G tier beside a mature 400G tier raises it. Optics supplied for branded switches are compatible parts coded for the platform you name, tested against the port, and are not manufactured by the brand of the host. Platforms common here include Mellanox, Arista, Cisco and Juniper. Where several links share one pair see the WDM transceivers page, and where capacity is aggregated onto wavelengths see the 800G DCO transponder card.

How to choose, in order

1. Fix the capacity target and the refresh window, then the rate per tier.
2. Count ports and radix per rack unit; if neither is the limit, 400G stays the volume tier.
3. Measure the longest route after patch panels and check the installed fibre type against the rate.
4. Check the port power budget and airflow of each rack before adding high-rate ports.
5. Set the breakout plan, and count link types rather than links for the spares pool.

Frequently asked questions

Is 800G always cheaper per bit than 400G?
On the interface line, yes: fewer ports carry the same capacity. Overall it is cheaper only where port count per rack unit is the binding limit, since a rack already out of power or airflow gains little.

Can 400G and 800G live in the same fabric?
Yes, and most builds do exactly that. The spine is refreshed first, and breakout assemblies let an 800G cage feed 400G access links until that tier is replaced.

What decides between DR8 and 2xFR4 at 800G?
Reach and cabling. DR8 is the short parallel class on MPO or MTP cabling; 2xFR4 is the 2 km class on a duplex pair that already exists.

Does a higher port rate reduce fibre cost?
Sometimes. Fewer links can mean fewer pairs, but a higher-rate link may need more strands or a coherent module. Where fibre is scarce, coherent on the existing pair usually beats new cable.

What should be in the request for a rate comparison?
The capacity target, ports and radix per rack unit, measured route length and loss by tier, the fibre and cabling in place, the rack power and airflow limits, and the switch platform and firmware.

Send the capacity target, the ports and radix per rack unit, the measured routes and the switch platform, and we confirm the rate per tier, the reach code and the coding for every link type. Related reading: 800G DCO transponder card for wavelength level capacity, 100G QSFP28 where a site is not yet at 400G, and all optics selection guides.