NVIDIA and Cisco unveil USFP, a new pluggable optics form factor: 400Gb/s per lane and up to 12.8Tb/s

Start with the facts in the official press release. It is dated October 12, 2026, datelined Santa Clara, California, at the OCP Global Summit, and issued by NVIDIA and Cisco together. USFP stands for Ultra-Dense Small Form Factor Pluggable, described as the first module designed to support 400Gb/s-per-lane electrical signals, with capacity up to 12.8Tb/s. Two direct benefits for AI factory operators are spelled out: letting Ethernet AI fabrics scale from 200Gb/s per lane to 400Gb/s per lane while extending a deployment model already proven at scale across generations of data center networks, and letting customers connect more computing systems without adding network tiers.

The key numbers in the spec. Cisco states that USFP delivers 4x the bandwidth and front-panel bandwidth density, with approximately 1.3x the power efficiency, versus 200Gb/s-per-lane pluggable modules, and reaches 12.8Tb/s of overall capacity in a comparable OSFP-sized module. The first planned USFP transceivers pack 16 full-duplex lanes into an OSFP-sized footprint, supporting both switches and SuperNICs, with the ability to expand to 32 and more full-duplex lanes in future. The connector is worth calling out on its own: Cisco describes a new two-dimensional array connector with a two-step insertion design that engages high-speed contacts only after the module is inserted, reducing contact wear, improving signal integrity and making insertion force more predictable — and that connector technology is what leaves room to keep scaling lane count. On thermals, USFP supports power dissipation up to 100 watts without needing a separate dedicated embedded cooling circuit inside the transceiver, which reduces switch design complexity.

The ecosystem and the open-spec angle may matter more over time. The release says NVIDIA and Cisco are collaborating with TE Connectivity and Amphenol to build USFP cages, connectors and modules, with additional partners expected to join; a shared form factor gives suppliers a common foundation for high-density pluggable infrastructure for 400Gb/s-per-lane fabrics. The two companies also say they plan to create an open industry specification for USFP through an upcoming community-led multi-source project. The context is existing collaboration to make NVIDIA Spectrum-X Ethernet networking and Cisco Silicon One-based Ethernet networks work together seamlessly, simplifying AI infrastructure deployment for enterprises, neoclouds and sovereign clouds. In other words, this module is not just a two-vendor product; it is meant to become a multi-source public interface.

How to read it. First, this is two competitors collaborating on a single form factor rather than launching rival products; the quotes come from Gilad Shainer, senior vice president at NVIDIA, and Bill Gartner, senior vice president and general manager of optical systems and optics at Cisco, and both stress the same theme: AI factories need to connect more compute while cutting networking's share of the power budget. Second, watch the tense in the wording — 'announced plans for', 'first planned transceivers', 'plan to create an open industry specification' — meaning the spec and the products are still in progress, with no volume shipping details and no pricing. Third, for anyone following AI infrastructure, the value of this kind of announcement is that it locates the bottleneck: when the industry starts seriously solving problems at the granularity of 400Gb/s-per-lane electrical signals and 100-watt module thermals, the network layer has become one of the real constraints on scaling AI clusters.