Particle.news
Get it on Google Play
Download on the App Store
5 ARTICLES
·
5h ago

NVIDIA and Cisco Announce USFP Pluggable to Quadruple Front‑Panel Bandwidth

The 400 Gb/s‑per‑lane USFP design aims to boost per‑module capacity within existing rack and service models so data centers can add AI fabric bandwidth without redesigning hardware.

A close-up view of the Cisco and NVIDIA Ultra-Dense Small Form-Factor Pluggable (USFP) module, featuring a sleek, metallic design against an abstract, high-tech background representing AI data center networks.
A look at the new USFP optical module designed for next-generation data center networks.
Image

Overview

  • The companies announced design plans on Monday, October 12, 2026, for USFP, a new pluggable transceiver targeting 400 Gb/s per lane and an initial 16‑lane (6.4 Tb/s) implementation with a path to 32 lanes (12.8 Tb/s).
  • USFP is sized to fit an OSFP footprint while offering about four times the bandwidth density of today’s 200 Gb/s‑per‑lane pluggables and roughly 1.3 times better per‑bit power efficiency versus 200 Gb/s designs.
  • Engineered features include a new two‑dimensional connector with a two‑step insertion to protect high‑speed contacts and support for roughly 100 watts of dissipation so transceivers do not need embedded cooling.
  • NVIDIA and Cisco are working with TE Connectivity and Amphenol and plan an open, community‑led multi‑source specification, but the project is still in ecosystem and specification development with no broad product shipments announced.
  • If adopted, USFP could let operators increase switch front‑panel capacity and keep field‑replaceable optics, while the industry still evaluates tradeoffs with alternatives such as near‑ or co‑packaged optics and awaits multi‑vendor validation and volumes.