Nvidia and Cisco Unveil 400 Gb/s per Lane USFP Optical Module Project for AI
Nvidia and Cisco announced on October 12, 2026 their USFP (Ultra-Dense Small Form Factor Pluggable) project, a pluggable optical module format designed to be the first to support electrical signals of 400 Gb/s per lane.
The objective stated by both groups is to enable AI factory operators to connect more computing systems without adding additional network layers and consuming less energy per bit transmitted.
12.8 Tb/s Capacity in OSFP Form Factor
The first transceivers planned based on USFP will combine 16 full-duplex lanes in a module the size of an OSFP (octal small form factor). They are intended for both switches and SuperNICs, with the possibility of scaling to 32 lanes and beyond in the future. According to the statement, customers will be able to achieve overall capacity of 12.8 Tb/s.
By doubling both the number of lanes and bandwidth per lane, the format should enable connection of more terminals at 1.6 Tb/s without adding network layers. Compared to pluggable modules at 200 Gb/s per lane, Nvidia and Cisco claim bandwidth and front-panel bandwidth density multiplied by 4, as well as energy efficiency approximately 1.3 times higher.
The economic mechanism described by both groups is as follows: when a switch reaches its capacity, network expansion may require additional switching layers and optical links, which increases latency, power consumption and costs. USFP aims to reduce the network's share of the overall energy budget of an AI factory, without redesigning existing data center infrastructure.
TE Connectivity and Amphenol Involved, Open Specification Planned
Nvidia and Cisco indicated they are collaborating with TE Connectivity and Amphenol to design USFP cages, connectors and modules, with other partners expected. The common format should provide suppliers with a shared basis for developing high-density pluggable infrastructures intended for 400 Gb/s per lane networks.
Both groups also plan to create an open industry specification for USFP, as part of a future multi-source project led by the community. This collaboration extends work already undertaken to ensure interoperability between Nvidia Spectrum-X Ethernet networks and those based on Cisco Silicon One, targeting enterprises, "neoclouds" and sovereign clouds.
Redesigned Connector and Thermal Dissipation Up to 100 Watts
From a technical standpoint, USFP is based on a new two-dimensional matrix connector with two-stage insertion: high-speed contacts engage only once the module is inserted. According to the statement, this device reduces contact wear, improves signal integrity and makes insertion force more predictable. Designed for 400 Gb/s, the connector also offers better performance at 200 Gb/s per lane.
The module supports thermal dissipation of up to 100 watts without a dedicated integrated cooling circuit, which simplifies switch design. "Cisco and NVIDIA recognized that scaling pluggable optics to 400Gb/s per lane would require more than another incremental extension of today's module designs", said Bill Gartner, senior vice president and general manager of optical systems and optics at Cisco.
Gilad Shainer, senior vice president at Nvidia, for his part indicated that both groups were building an ecosystem aimed at "reinventing pluggable optics".
