NVIDIA and Cisco have joined forces to launch the Ultra-Dense Small Form Factor Pluggable (USFP), an innovative optical transceiver standard designed to help AI factory operators scale networking infrastructure efficiently. Designed as the industry’s first module supporting 400Gb/s-per-lane electrical signaling, the USFP specification delivers up to 12.8Tb/s of throughput per module, quadrupling bandwidth density while drastically curbing network energy overhead.
Overcoming Network Bottlenecks in AI Factories
As hyperscale artificial intelligence models expand and inference demands accelerate worldwide, network architects face severe physical and operational constraints. Traditional scaling strategies rely on adding additional switch layers and optical links when a switch reaches its maximum connection density. However, expanding physical tiers introduces higher latency, greater power consumption, and escalating infrastructure costs.
The USFP architecture tackles these scaling hurdles head-on:
- quadrupled Bandwidth Density: USFP modules house 16 full-duplex lanes within an Octal Small Form Factor (OSFP)-sized footprint, yielding 4x the front-panel bandwidth density compared to 200Gb/s-per-lane legacy optics.
- Enhanced Energy Efficiency: By delivering approximately 1.3x the power efficiency per bit of transferred data, the technology minimizes networking’s overall share of the data center power budget.
- Streamlined Network Topologies: By doubling both total lane count and per-lane throughput, operators can connect more 1.6Tb/s endpoints without needing extra intermediate network tiers.
- Extended Serviceability Model: USFP preserves the familiar plug-and-play operational simplicity and field serviceability that modern data center operations depend on.
Also Read: Absolute Security Acquires FirmGuard to Strengthen AI-Native Firmware Security and Enterprise Cyber Resilience
Advanced Hardware and Thermal Engineering
Engineered explicitly for 400Gb/s-per-lane electrical signaling, the USFP form factor incorporates a two-dimensional array connector featuring a two-step mechanical insertion mechanism. High-speed electrical contacts engage only after the module is fully seated into the cage, mitigating mechanical pin wear, improving signal integrity, and standardizing insertion force.
| Technical Metric | Legacy 200Gb/s Optics | Ultra-Dense SF Pluggable (USFP) |
| Max Capacity per Module | 1.6 Tb/s – 3.2 Tb/s | Up to 12.8 Tb/s |
| Per-Lane Electrical Rate | 200 Gb/s per lane | 400 Gb/s per lane |
| Full-Duplex Lane Count | Up to 8 lanes | 16 lanes (expandable to 32+) |
| Front-Panel Density | Standard OSFP baseline | 4x higher bandwidth density |
| Power Efficiency Improvement | Standard baseline | ~1.3x energy efficiency gain |
| Power Dissipation Limit | Varies | Up to 100W (standard cooling) |
To support demanding workloads without driving up chassis complexity, USFP accommodates up to 100 watts of power dissipation using conventional airflow and thermal management. Eliminating the requirement for dedicated internal liquid cooling loops within the transceiver simplifies switch engineering and enhances system reliability. The physical connector design also scales to 32 full-duplex lanes in future iterations while providing superior performance when operating backward at 200Gb/s per lane.
Executive Perspectives
“Building the world’s largest AI factories starts with rethinking the smallest of connections,” said Gilad Shainer, senior vice president NVIDIA. “With USFP, NVIDIA and Cisco are building an ecosystem to reinvent pluggable optics — connecting more computing 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 a more complex operating model,” said Bill Gartner, senior vice president and general manager of optical systems and optics at Cisco. “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. Through our collaboration, USFP is designed to deliver a new level of bandwidth density while preserving the serviceability, deployment flexibility and operational simplicity of pluggable optics. By taking a full-stack approach from silicon and systems to optics and the network operating system we can optimize every layer to deliver the performance and scale the next generation of AI networks will require.”
Building an Open, Multi-Source Ecosystem
For the smooth and seamless adoption of the technology within global data centers, both NVIDIA and Cisco are working in partnership with other interconnect vendors such as TE Connectivity and Amphenol to develop USFP cages, connectors, and transceivers.
This vendor neutral development goes hand-in-hand with the ongoing technological integration of the NVIDIA Spectrum-X™ Ethernet platforms with Cisco’s Silicon One based networking infrastructure. In addition, both the firms plan to initiate a multi-source agreement (MSA) project for an industry specification of the USFP form factor.