NVIDIA and Cisco Announce Plans for USFP Optical Modules for AI Ethernet Fabrics

NVIDIA and Cisco have outlined plans for Ultra-Dense Small Form Factor Pluggable (USFP), a high-density optical-module form factor aimed at next-generation Ethernet networks for AI factories. The companies are also working with TE Connectivity and Amphenol on a multi-supplier hardware ecosystem.

USFP is designed for 400Gb/s electrical signals per lane. NVIDIA said the first planned USFP transceivers will package 16 full-duplex lanes in an OSFP-sized footprint, for up to 12.8Tb/s of aggregate capacity. The modules are intended for deployment on both switch and SuperNIC sides of the network, with a planned path to 32 or more lanes in the future.

The proposal targets a central constraint in large AI fabrics: expanding bandwidth and connection density without increasing the number of network layers or the physical footprint of the network. NVIDIA claims USFP can provide four times the bandwidth and front-panel bandwidth density, and approximately 1.3 times the power efficiency, relative to 200Gb/s-per-lane pluggable optics.

For operators, the relevant architectural proposition is denser Ethernet connectivity within an OSFP-sized envelope. This could support fabric growth while reducing incremental networking infrastructure and potentially lowering the networking share of an AI factory’s energy budget. It could also intensify competition across high-speed optics and AI networking hardware, where port density, network-tier count and data-transfer power consumption are becoming material design considerations.

The announcement remains a plan rather than a commercial product launch. Its impact on procurement and data-center network architectures will depend on commercialization, interoperability, cost, standards status and operator adoption.