A second row of electrical contacts in a cage that is otherwise compatible with QSFP - eight lanes instead of four, from a module slightly deeper than a QSFP28.

The connector adds a row of contacts behind the existing one. A QSFP28 module inserted into a QSFP-DD cage engages only the first row and operates normally at 100 Gb; a QSFP-DD module engages both and uses all eight lanes.

The cage is a little deeper to accommodate the second row and the additional electronics, but the faceplate opening is the same, so port density on the front panel is unchanged.

That is the whole point of the design: a switch built with QSFP-DD ports offers 400 Gb where 100 Gb ports were, at the same density, while remaining able to use every module already owned.

The consequence is entirely thermal - twice the lanes in nearly the same volume - which is why power class and cooling dominate the specification conversation for this format.

Because eight lanes dissipate far more than four in almost the same volume, and an overheating module throttles or shuts down rather than degrading gracefully.

A 400 Gb module can dissipate several times what a 10 Gb module does, in a package only modestly larger. The heat has to leave through the cage and into the switch's airflow.

Modules are therefore classified into power classes, and switch manufacturers state how many modules of each class may be fitted and in which ports. Exceeding that is not a warning - the modules run hot, error rates rise, and the switch may disable ports to protect itself.

Cage design matters: heatsinked cages, riding heatsinks that contact the module, and engineered airflow are all part of current switch design rather than optional refinements.

So when planning a 400 Gb deployment, read the switch's population rules, confirm the cabinet's airflow and inlet temperature, and monitor module temperatures once in service. This is the constraint that most often limits what can actually be installed.

Yes - that backwards compatibility is the format's main commercial advantage.

A QSFP28 module fits and runs at 100 Gb. A QSFP+ module generally fits and runs at 40 Gb. The port detects what is present and configures itself.

That allows a switch refresh to happen before an optics refresh. New switches with QSFP-DD ports can be installed and cabled with the existing 100 Gb optics, and individual links moved to 400 Gb afterwards as capacity demands and budget allow.

It also means 100 Gb optics purchased now are not stranded by the next generation of switches, which is a meaningful consideration when optics represent a large share of a fabric's cost.

Check the specific platform for any restrictions on which ports support which speeds and whether breakout is available in each mode - these details vary and are usually documented in a compatibility matrix rather than the main datasheet.

Into four 100 Gb links, or eight 50 Gb links, depending on the module and the switch - and breakout is a large part of why 400 Gb ports are bought.

Eight lanes at 50 Gb can be presented as eight separate 50 Gb interfaces, or paired into four 100 Gb interfaces. A splitter cable or a parallel fibre fan-out connects them to the corresponding ports at the far end.

That makes a 400 Gb spine port an efficient way to feed four leaf switches at 100 Gb, using one port, one module and one cable assembly.

The considerations are familiar from lower speeds: not every port supports every breakout mode, configuration usually requires the port group to be reset, and all the sub-links share one physical path, so resilient pairs should not both terminate on one breakout.

Also confirm the lane rate compatibility - 100 Gb links formed from two 50 Gb lanes are not the same as QSFP28's four 25 Gb lanes, and the far end must support the mode being used.

QSFP-DD where backwards compatibility with existing QSFP optics matters, OSFP where thermal headroom is the priority - and in practice the switch platform decides.

Both formats carry eight lanes and both support 400 Gb and beyond. QSFP-DD keeps the QSFP mechanical envelope and accepts existing QSFP modules directly, which is a substantial practical advantage in an estate that already owns 100 Gb optics.

OSFP is slightly larger with an integrated heatsink design, giving better thermal capacity and more headroom for higher power modules - which matters as lane rates continue to rise. It needs an adaptor to accept QSFP modules.

Switch vendors choose one, so the decision is usually made by the platform selected rather than independently.

Where there is a choice, the question is whether the existing optics investment or the future thermal headroom matters more. For most enterprise deployments the compatibility argument wins; for the largest and densest fabrics, the thermal argument does.

Parallel single-mode or multimode ribbon for most reaches, with duplex single-mode options using wavelength multiplexing - and the fibre plant matters more than ever.

Parallel modules use eight or sixteen fibres through an MPO connector, carrying the lanes separately. This is the common arrangement inside a data centre and is what supports breakout.

Duplex single-mode variants multiplex the lanes onto one fibre pair, which allows 400 Gb over existing duplex plant - valuable in buildings and between sites where installing new ribbon is impractical.

Multimode is increasingly constrained at these rates: supported distances are short, and older grades may not work at all. New builds that expect to reach 400 Gb and beyond should be installing single-mode.

End face cleanliness matters more at higher lane rates because the loss budgets are tighter, so MPO inspection and cleaning tooling is not optional.

Direct attach and active optical cable assemblies are available for in-rack and in-row connections and avoid the connector question entirely.

Rarely today, but the port type is worth having for compatibility and headroom rather than for the speed itself.

400 Gb links are driven by spine layers in large fabrics, by service provider aggregation and by data centre interconnect. An enterprise campus or a single-hall data centre generally has no traffic that requires it.

What is worth having is the port. A switch with QSFP-DD ports accepts existing 100 Gb optics today, offers breakout flexibility, and provides an upgrade path that does not require replacing the switch when demand grows.

The practical advice for most organisations is to build at 100 Gb with equipment whose ports can go further, and to install single-mode fibre wherever new cabling is laid - because the fibre is the part that is expensive to change, and single-mode does not need changing.

Buying 400 Gb optics before there is traffic for them is rarely justified; buying ports that can take them usually is.