NVIDIA and Cisco Announce USFP 400Gb/s-per-lane Optical Modules

NVIDIA and Cisco announced on 12 October 2026 plans for Ultra-Dense Small Form Factor Pluggable (USFP) optical modules, which support 400Gb/s-per-lane electrical signals to enable up to 12.8Tb/s of total capacity. (400Gb/s-per-lane: the amount of data transmitted per individual path in a multi-lane electrical signal connection.) (Ultra-Dense Small Form Factor Pluggable (USFP): A high-density optical module designed to provide high-speed data connections for next-generation artificial intelligence factories.) These modules target AI factory builders who need to scale Ethernet AI fabrics without adding network tiers. If you are currently deploying 200Gb/s-per-lane pluggable optics, these modules offer a higher-density path for future-proofing your infrastructure.

USFP introduces 400Gb/s-per-lane connectivity for AI factories

Ultra-Dense Small Form Factor Pluggable (USFP) is a high-density optical module designed to equip the hardware ecosystem for next-generation AI factories. The first planned USFP-based transceivers will pack 16 full-duplex lanes into an octal small form factor (OSFP)-sized footprint. (full-duplex lanes: communication paths that allow data to be sent and received simultaneously in both directions.) The design allows for expansion to 32 and more lanes in the future. The capacity reaches 12.8Tb/s.

Performance gains over 200Gb/s-per-lane optics

NVIDIA and Cisco say the USFP module delivers 4x the bandwidth density and approximately 1.3x the power efficiency compared to 200Gb/s-per-lane pluggable modules. (bandwidth density: the amount of data capacity that can be transmitted within a specific physical area or footprint.) These gains allow AI builders to increase compute scale within the same physical network footprint while decreasing the power share of the networking infrastructure. This happens because AI models are growing and inference demand is expanding, requiring more data movement per watt. (inference demand: the requirement for an artificial intelligence model to provide outputs based on new input data.)

Simplified cooling and signal integrity

The USFP module uses a new two-dimensional array connector to handle 400Gb/s-per-lane electrical signals. This connector features a two-step insertion design that engages high-speed contacts after the module is inserted. The design reduces contact wear and makes insertion force more predictable. Furthermore, the USFP supports power dissipation of up to 100 watts without requiring a separate, dedicated embedded cooling circuit within the transceiver.

Multi-source ecosystem and open standards

NVIDIA and Cisco are collaborating with TE Connectivity and Amphenol to build a multi-source hardware ecosystem around the USFP form factor. This shared foundation provides a common path for suppliers to develop high-density pluggable infrastructure for 400Gb/s-per-lane fabrics. To prevent vendor lock-in, NVIDIA and Cisco plan to create an open industry specification for USFP through an upcoming community-led project. (vendor lock-in: a situation where a customer becomes dependent on a single supplier for specialized hardware or software.)

Scalability from 16 to 32+ lanes

The initial USFP configuration supports 16 full-duplex lanes, but the connector technology is designed for further scaling. The architecture supports expanding to 32 full-duplex lanes or more in the future. This scalability helps ensure the OSFP-sized footprint remains viable as bandwidth requirements increase.

Data center network architects scaling AI fabrics

Data center network architects scaling AI fabrics should use USFP to reach 12.8Tb/s capacity per connection. By using these modules, you can scale your Ethernet AI fabrics without adding additional switch layers, which typically increase latency and power consumption. The USFP enables you to connect more 1.6Tb/s endpoints while maintaining a lower network complexity.

Hardware procurement managers seeking multi-source availability

Hardware procurement managers seeking multi-source availability should track the progress of the open industry specification for USFP. Because NVIDIA and Cisco are collaborating with TE Connectivity and Amphenol, you can identify a broader range of suppliers to ensure long-term component availability. This multi-source strategy helps avoid the risks associated with single-vendor dependencies in the AI factory supply chain.

Electrical engineers designing high-density switch ports

Electrical engineers designing high-density switch ports can use the USFP’s 100-watt power limit to simplify switch hardware design. Because the module does not require a dedicated internal cooling circuit, you can reduce the complexity of the switch housing. You should verify that your switch housing can accommodate the 100 watts power dissipation limit to ensure reliability.

Consumer PC gamers and mobile developers

Consumer PC gamers can ignore this because the USFP is a data center infrastructure component and not a component for consumer PCs; read our general GPU networking news for home use instead. Mobile device developers should skip this as the focus is on AI factory interconnects; read our mobile hardware and interconnect specifications instead.

A data center capacity expansion scenario for Colette

Colette runs a data center and currently operates an AI factory that is reaching the limits of its 200Gb/s-per-lane switch layers. To expand, Colette evaluates whether to add more switch layers or upgrade to USFP modules. She chooses USFP because it provides 12.8Tb/s of capacity and 1.3x better power efficiency per bit. By choosing USFP, Colette scales her compute without increasing network complexity. If Colette had chosen to stick with 200Gb/s-per-lane modules, she would have been forced to add more switch layers, which would have increased her latency and costs.

Applications for infrastructure and system design

What this replaces: The USFP replaces 200Gb/s-per-lane pluggable modules

The USFP replaces 200Gb/s-per-lane pluggable modules and the need for additional switch layers. By providing 4x the bandwidth density, it allows builders to increase compute scale within the same physical network footprint. This reduces the share of the overall AI factory power budget dedicated to networking.

Unconfirmed details and market competition

NVIDIA and Cisco did not state the specific completion date for the open industry specification for USFP. Additionally, the company did not confirm the current status of NVIDIA Spectrum-X Ethernet networking support for the new modules. While NVIDIA and Cisco are leading this development, Broadcom remains a primary competitor in the Ethernet switching and interconnect market. The ultimate success of USFP depends on the community-led project establishing a widely adopted standard for the 400Gb/s-per-lane connector.

Reported by NVIDIA on 12 October 2026: NVIDIA and Cisco Introduce New Pluggable Optics Form Factor to Scale AI Factories.