Cisco and Nvidia plan faster links for AI data centres
The companies are developing an open pluggable-optics design aimed at denser AI networks.
Cisco and Nvidia have announced plans for a new pluggable optical module designed to connect additional computing capacity within AI data centres. The proposed USFP design is intended to support 400 gigabits per second per lane and an aggregate capacity of up to 12.8 terabits per second.
The Ultra-Dense Small Form Factor Pluggable, or USFP, was announced at the Open Compute Project Global Summit in San Jose on 12 October. It is a proposed hardware form factor for Ethernet networks used to connect processors, switches and other components in large AI systems.
The first planned modules would use 16 full-duplex lanes in a footprint comparable with an existing OSFP module. Cisco and Nvidia say the design could later expand to 32 lanes, allowing data-centre operators to increase capacity without adding further layers of network equipment.
The companies describe USFP as the first pluggable design intended for 400Gb/s electrical signals per lane. They claim it could provide four times the bandwidth density and about 1.3 times the power efficiency of 200Gb/s-per-lane optics. Those performance comparisons are manufacturer claims, not results from independent testing.
The proposal is aimed at a practical problem in AI infrastructure: moving data between large numbers of processors while controlling power, cooling, latency and the physical size of the network. Adding more switches and optical links can increase complexity and energy use, so higher-density connections are intended to extend existing deployment models.
Cisco and Nvidia said they were working with TE Connectivity and Amphenol on cages, connectors and modules. They plan to develop USFP as an open, community-led specification with multiple suppliers, a structure intended to reduce dependence on one manufacturer and give data-centre operators more choice.
Independent industry coverage described the announcement as a development plan rather than a finished product launch. SDxCentral reported that the initial package could support 6.4Tb/s or 12.8Tb/s depending on its lane configuration, while also noting that more details about the specification are expected later.
The announcement does not include a general availability date, customer deployments or independent measurements of power savings. Its significance therefore lies in the direction of network design and in the support the companies are giving to an open form factor, rather than in an immediate change to the performance of existing AI data centres.
Fact-check Approved · Priya Shah — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The article distinguishes clearly between the announced design and the companies’ performance claims. The technical specifications and partner list are supported by the primary announcement and independently reported trade coverage.
- confirmed Cisco and Nvidia announced USFP plans on 12 October 2026. — The Nvidia investor release gives the date and describes the announcement. source
- confirmed USFP is designed for 400Gb/s per lane and up to 12.8Tb/s aggregate capacity. — These specifications appear in the primary release. source
- confirmed The first planned modules would use 16 full-duplex lanes and could expand to 32. — The lane configuration is described in the primary release. source
- confirmed Cisco and Nvidia claim four times the bandwidth density and about 1.3 times the power efficiency of 200Gb/s-per-lane optics. — The figures are stated by the companies and are explicitly attributed as claims. source
- confirmed TE Connectivity and Amphenol are working with Cisco and Nvidia on the hardware ecosystem. — The partner list appears in the announcement. source
- confirmed The announcement is a development plan rather than a finished product launch. — The release repeatedly describes planned modules and a specification under development; SDxCentral also reports further details are forthcoming. source
Editor's note
The proposed form factor, stated capacities and named partners are supported by the companies’ announcement and independent trade coverage. Real-world performance, availability, adoption and the claimed efficiency gains remain unverified.Sources
More on this in other media
- The Guardian — „nvidia cisco”
- BBC News — „nvidia cisco”
- Sky News — „nvidia cisco”