What happened
NVIDIA and Cisco announced the development of USFP, a new optical transceiver form factor intended to support 400 Gb/s-per-lane connectivity for AI infrastructure. The design integrates 16 full-duplex lanes into an OSFP-sized footprint, offering four times the bandwidth density of current 200 Gb/s modules and improved power efficiency. The companies are collaborating with TE Connectivity and Amphenol to develop the necessary connectors and modules, with plans to establish an open industry specification through a multi-source project.
NVIDIA and Cisco have jointly unveiled the Ultra-Dense Small Form Factor Pluggable (USFP) standard, a new optical transceiver form factor specifically designed for next-generation AI factories. According to eeNews Europe, the technology is engineered to support 400 Gb/s-per-lane connectivity, which represents a significant leap from the current industry standard of 200 Gb/s-per-lane pluggable optics.
The initial USFP design integrates 16 full-duplex lanes into a footprint the size of an existing OSFP module. This configuration allows for a total capacity of up to 12.8 Tb/s per module. The companies claim this design offers four times the bandwidth and front-panel bandwidth density of current 200 Gb/s modules, while also providing approximately 1.3 times better power efficiency.
A key technical innovation in the USFP design is a two-dimensional array connector with a two-step insertion mechanism. This mechanism ensures that high-speed electrical contacts engage only after the module is fully inserted, which helps reduce contact wear, improve signal integrity, and provide more predictable insertion force. The modules are designed to dissipate up to 100W without requiring dedicated embedded cooling circuits, potentially simplifying switch designs.
NVIDIA and Cisco are collaborating with TE Connectivity and Amphenol to develop the physical connectors, cages, and modules for USFP. The companies plan to establish an open industry specification through a community-led multi-source project, aiming to create a common platform for interoperable high-density optical hardware. This initiative builds on existing collaborations between NVIDIA Spectrum-X Ethernet networking and Cisco Silicon One-based networks.
Source details: eenewseurope.com ↗
Why it matters
This development addresses the critical bottleneck of network connectivity in large-scale AI training clusters. By increasing bandwidth density and reducing power consumption per bit, USFP allows operators to connect more computing systems within the same physical footprint without adding -inducing switching layers. This is particularly relevant for European data center operators and network equipment makers seeking to scale AI infrastructure efficiently. The move toward an open specification also aims to prevent vendor lock-in and foster interoperability among component suppliers, potentially lowering costs and accelerating the deployment of high-density Ethernet networks required for modern AI factories.
The primary significance of USFP lies in its ability to scale AI infrastructure without proportionally increasing network complexity or power consumption. By allowing more computing systems to be connected within the same network footprint, the technology helps avoid the addition of extra switching layers, which can otherwise increase , costs, and energy usage.
For European data center operators and network equipment manufacturers, this development offers a concrete path to higher-bandwidth networking while preserving the serviceability advantages of pluggable optics. It addresses the growing demand for high-density Ethernet networks that are essential for training large AI models.
The push for an open industry specification is also notable. By involving multiple component suppliers and aiming for a multi-source project, NVIDIA and Cisco are attempting to create a standardized ecosystem that reduces vendor lock-in. This could lead to more competitive pricing and wider adoption of high-density optical hardware across the industry.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
Monitor the formation of the open industry specification and the entry of additional partners beyond TE Connectivity and Amphenol. Watch for the release of initial transceiver hardware and real-world performance benchmarks in AI factory deployments. Additionally, track how this standard impacts the competitive landscape for existing 200 Gb/s optics providers and whether it influences broader Ethernet networking standards in the industry.
The next major milestone will be the formalization of the open industry specification. Observers should watch for the announcement of additional partners joining the multi-source project and the publication of the technical standards that will define USFP interoperability.
Hardware availability is a key factor to monitor. While the design is announced, the actual release of USFP transceivers, switches, and SuperNICs from NVIDIA, Cisco, and their partners will determine the practical impact of the technology. Early adopters in AI factory deployments will provide the first real-world performance data.
The competitive response from other optical component manufacturers and networking vendors will also be important. The success of USFP may depend on how quickly the broader industry adopts the standard and whether it becomes the de facto requirement for next-generation AI infrastructure.