NVIDIA and Cisco Jointly Define New USFP Pluggable Optical Module

C114 News, October 13 (Shui Yi) — NVIDIA and Cisco jointly announced on October 12 local time a plan for a new type of pluggable optical module, USFP (Ultra-Dense Small Form-Factor Pluggable), a new high-density, single-lane 400Gb/s optical module targeting the scaling needs of next-generation AI factory networks.

As AI models grow in size and inference demand expands, AI factories must connect more compute systems while managing network complexity, power consumption and cost. Increasing data throughput per watt and per switch helps operators make better use of their infrastructure investments.

The official press release notes that continuing to scale AI factories requires higher bandwidth per connection and more connections per switch. When switches reach their capacity limits, further network scaling may require additional switch tiers and optical links, adding latency, power consumption and cost.

USFP is designed to let AI factory builders scale Ethernet AI architectures from single-lane 200Gb/s to single-lane 400Gb/s while carrying forward deployment models proven at scale across multiple generations of data center networks. With USFP, customers can adopt next-generation optical technology to meet the performance demands of AI workloads without overhauling their data center infrastructure.

The two companies said the first USFP-based optical modules will pack 16 full-duplex lanes into an OSFP-equivalent form factor, supporting switches and SuperNICs, with future expansion to 32 or more lanes. By doubling both the lane count and bandwidth per lane, USFP enables AI architectures to connect more 1.6Tb/s ports without adding network tiers, and customers will be able to achieve 12.8Tb/s of total capacity with USFP.

Matching the size of an OSFP module, USFP delivers 4x the bandwidth and front-panel bandwidth density of single-lane 200Gb/s pluggable modules, along with about 1.3x the energy efficiency. This lets AI factory builders scale compute within the same network footprint while reducing the network's share of the AI factory's total power budget.

"Building the world's largest AI factories starts with rethinking the smallest connections," said Gilad Shainer, senior vice president at NVIDIA. "With USFP, NVIDIA and Cisco are building an ecosystem to reshape pluggable optical technology — connecting more compute with greater efficiency to power the next generation of AI factories."

"AI builders need an interconnect architecture that can scale performance without locking them into more complex operating models," said Bill Gartner, senior vice president and general manager of Cisco Optical Systems and Optical Components. "Cisco and NVIDIA recognize that scaling pluggable optics to single-lane 400Gb/s is more than an incremental extension of existing module designs. Through our collaboration, USFP aims to deliver an entirely new level of bandwidth density while preserving the serviceability, deployment flexibility and operational simplicity of pluggable optics. By taking a full-stack approach spanning silicon, systems, optical components and network operating systems, we can optimize every layer to deliver the performance and scale required for next-generation AI networks."

USFP is designed for single-lane 400Gb/s electrical signals and uses a new two-dimensional array connector with a two-step insertion design that engages the high-speed contacts after the module is inserted, reducing contact wear, improving signal integrity and making insertion force more predictable. The connector technology supports future expansion to 32 full-duplex lanes or more. Because the connector is purpose-built for 400Gb/s performance, it also delivers superior performance at single-lane 200Gb/s.

USFP supports power consumption of up to 100 watts without requiring dedicated embedded cooling circuits inside the optical module, reducing switch design complexity and improving reliability. Together, these features enable higher-bandwidth architectures while retaining the proven deployment, serviceability and manufacturability of existing pluggable optical modules.

All USFP configurations support optical designs that reduce latency and power consumption through simplified signal processing. Support for a broad range of optical technologies and link distances gives system designers the flexibility to meet the unique deployment needs of AI factories.

NVIDIA and Cisco are working with TE Connectivity and Amphenol to develop USFP cages, connectors and modules, with more partners expected to join. The shared form factor provides vendors with a common foundation for developing high-density pluggable infrastructure for single-lane 400Gb/s network architectures, building a strong ecosystem for AI factory builders.

The collaboration builds on NVIDIA and Cisco's existing work to ensure NVIDIA Spectrum-X™ Ethernet networks and Ethernet networks based on Cisco Silicon One work seamlessly together, simplifying AI infrastructure deployment for enterprises, emerging cloud service providers and sovereign clouds.

NVIDIA and Cisco plan to create an open industry specification for USFP through an upcoming community-led multi-source agreement (MSA).