USFPGuide
Illustration of a data center switch front panel with dense rows of pluggable optical modules and fiber cables

Announced October 12, 2026 · not shipping yet

USFP, the 6.4T pluggable for AI networks

Cisco and NVIDIA want the next generation of optical modules to run 400 Gb/s on every electrical lane, in a module the size of today's OSFP. This site tracks what has really been published, what hasn't, and how USFP stacks up against OSFP, XPO and co-packaged optics.

Lanes
16 to 32
Per lane
400 Gb/s
Per module
6.4T to 12.8T
Heat
100 W max

What USFP is Try the fabric calculator

Four times the bandwidth in one port's space

Today's fastest pluggables, the 1.6T OSFP modules going into AI clusters now, carry eight lanes of 200 Gb/s. USFP doubles the lane rate to 400 Gb/s and doubles the lane count to 16, which is where the 4x claim comes from: 6.4T through a front-panel opening about the size of one OSFP port.

A 32-lane version takes a single module to 12.8T. The companies also claim about 1.3 times the power efficiency of 200G-per-lane optics, a figure no one outside Cisco and NVIDIA has measured yet (NAND Research).

Every published figure, and what's still missing

One USFP module, first generation

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

The first USFP design has 16 full-duplex lanes. The 32-lane version doubles that to 12.8T. Source: Cisco and NVIDIA.

Status tracker

Where USFP stands today

An announcement isn't a product. Here is every step between the press release and a module you can put on a purchase order, and which ones have happened. We re-check it at least monthly; last on October 17, 2026.

The tracker with sources

  1. Announced by Cisco and NVIDIA October 12, 2026
  2. Connector and cage partners named October 12, 2026
  3. Multi-source agreement formed Not yet
  4. Specification published Not yet
  5. 400G-per-lane switch silicon announced Not yet
  6. Module samples shown Not yet
  7. Switches or SuperNICs with USFP ports announced Not yet
  8. Modules you can order Not yet

Checked October 17, 2026. 2 of 8 steps done.

How much each format carries

USFP is one of five ways the industry is trying to get optics past 1.6T per port. They differ in where the optics sit, how they're cooled and whether a technician can pull one out of a running switch.

  • OSFP 1.6T
  • OSFP-XD 3.2T
  • USFP 6.4T to 12.8T
  • Open CPX 7.2T
  • XPO 12.8T to 25.6T
One-direction capacity per module (per socket for Open CPX) at the lane rate each format has announced. The pale bar is the announced roadmap: USFP at 32 lanes, XPO at 400G per lane. Sources: the OSFP MSA, Cisco, Open CPX MSA, Arista and NAND Research.

Does it save a tier in your fabric?

The real argument for 400G lanes isn't the module, it's the switch behind it. Double the lane rate and a switch chip of the same size has twice the 1.6T ports, so a cluster that needed three tiers of switches can fit in two. Fewer tiers means fewer switches, fewer optics and less latency.

Put in your endpoint count and port speed and the calculator shows the switches, links and pluggable ends on today's 102.4T chips and on a 204.8T, 400G-per-lane chip.

Open the fabric calculator

4,096 endpoints at 1.6T on a 102.4T switch
3 tiers
The same cluster on a 204.8T, 400G-per-lane switch
2 tiers

Non-blocking leaf-spine maths. A 102.4T chip has 64 ports of 1.6T (up to 2,048 endpoints in two tiers); a 204.8T chip has 128 (up to 8,192).