What is USFP?

A pluggable is the small metal module that slides into the front of a switch or network card and turns electrical signals into light for the fiber. Every big jump in network speed has needed a new one, or a new version of an old one: SFP, QSFP, QSFP-DD, OSFP. USFP is the format Cisco and NVIDIA propose for the next jump, which they announced together at the OCP Global Summit on October 12, 2026 (Cisco Newsroom).

Lanes

16 to 32

Per lane

400 Gb/s

Per module

6.4T to 12.8T

Heat

100 W max

What the companies announced

The first USFP modules are meant to carry 16 full-duplex electrical lanes, each running at 400 Gb/s, in a module roughly the size of today's OSFP. That's 6.4 Tb/s in each direction. The design has room for 32 lanes, which would take one module to 12.8 Tb/s. Against the 200G-per-lane modules shipping now, Cisco and NVIDIA claim four times the front-panel bandwidth density and about 1.3 times the power efficiency.

A module may give off up to 100 W of heat without a dedicated cooling circuit inside it. That matters because the rival XPO format puts a liquid cold plate in every module; USFP is meant to work in both air-cooled and liquid-cooled systems.

The first products to use it are meant to be switches and SuperNICs (NVIDIA's name for its high-end network cards), on NVIDIA Spectrum-X Ethernet and Cisco Silicon One networks.

Why a new format at all

Switch chips get faster by running more lanes, faster lanes, or both. Today's top chips run 200 Gb/s per lane, and an OSFP module with eight of those lanes is a 1.6T port. The industry's next lane rate is 400 Gb/s. Bill Gartner, who runs optics at Cisco, said getting pluggables to 400G per lane "would require more than another incremental extension" of the existing formats (joint release).

The trouble is the front panel. A 1RU switch only has room for about 32 OSFP ports across its face, so as chips get bigger, the ports run out before the chip's bandwidth does. When that happens, network builders add another layer of switches. The USFP pitch is that 16 fast lanes per opening lets one switch connect more 1.6T endpoints directly, so large AI clusters need fewer tiers. Our fabric calculator shows the effect on a real cluster size.

One USFP module, first generation

400G400G400G400G400G400G400G400G400G400G400G400G400G400G400G400G

16 lanes × 400 Gb/s = 6.4 Tb/s each way, or four 1.6T ports

The new connector

USFP doesn't use the card-edge connector of SFP, QSFP and OSFP. It uses a new two-dimensional array of contacts with a two-step insertion: the module goes in, and the high-speed contacts only engage once it's fully seated. Cisco and NVIDIA say this means less wear on the contacts, cleaner signals and a more predictable insertion force, and that it improves 200G-per-lane performance too. TE Connectivity and Amphenol are building the cages and connectors.

A new connector also means new cages. No one has said an OSFP module will fit a USFP port, or the other way round, so plan on USFP being a new port type rather than an upgrade to existing switches (NAND Research).

What USFP is not, yet

It isn't a published standard and it isn't a product. Cisco and NVIDIA say they will publish it as an open specification through a community-led multi-source agreement, but there's no MSA website, member list or draft. There are no samples, no prices and no ship dates, and no 400G-per-lane switch chip has been announced to plug it into. Our status tracker follows each of those steps.

Who it's for

Hyperscalers and the newer GPU clouds ("neoclouds") are the first audience. Anyone planning a large AI back-end network for the 400G-per-lane generation should watch it, because the module format a next-generation switch ships with decides which optics, cages and suppliers you can use for years.

• USFP vs OSFP

  Same footprint, four times the bandwidth, a new connector. Why your OSFP cages won't take it.

• USFP vs XPO

  Air-friendly 6.4T against a liquid-cooled 12.8T. Two camps, two bets on the next switch chip.

• USFP vs co-packaged optics

  Keep swappable modules, or put the optics in the switch package for the lowest power.

• Every form factor

  OSFP to co-packaged optics in one table: lanes, capacity, heat and who backs each.