NVIDIA and Cisco are planning a major networking upgrade that could help AI data centers handle more traffic, improve power efficiency, and scale without rebuilding their existing infrastructure.

As artificial intelligence requirements continue growing, USFP (Ultra-Dense Small Form Factor Pluggable), a new optical module developed by NVIDIA and Cisco, will help to boost network capacity in order to build AI factories of the future. The module will be able to transfer data at a rate of 400 gigabits per second (Gbps) per lane.

The announcement reveals an increasing problem in the data analytics and AI industry, which is that increased data processing needs faster interconnections between processors, storage, and applications.

AI Data Centers Need More Bandwidth Without Higher Costs

The AI manufacturing process depends on massive networks for transferring data through thousands of processing devices. In the course of time, the network capacity may become insufficient due to increased workload. Adding new switches and optics will be expensive and time-consuming.

USFP aims to address this problem by increasing bandwidth within a familiar, pluggable design. This allows data center operators to upgrade networking capabilities without completely changing their existing infrastructure.

“With USFP, NVIDIA and Cisco are building an ecosystem to reinvent pluggable optics,” said Gilad Shainer, senior vice president at NVIDIA. He added that the technology is designed to connect more computing resources efficiently and support the next generation of AI factories.

How Much More Capacity Can USFP Deliver?

The planned technology will initially feature 16 full-duplex lanes, each capable of handling 400 Gbps. Together, these lanes provide an aggregate capacity of 6.4 terabits per second (Tbps). Future designs could expand to 32 lanes, reaching 12.8 Tbps.

According to the companies, USFP offers four times the bandwidth density of existing 1.6T pluggable modules and approximately 1.3 times greater power efficiency compared with 200 Gbps-per-lane modules.

In addition, the design allows for power dissipation up to 100 watts without having to install a dedicated cooling circuit within the transceiver. The new connector was designed in order to enhance signal quality, reduce friction and enable future bandwidth upgrades. This could help in scaling the AI computing capabilities within data centers as well as deal with the increased power and cooling needs.

Why Does This Matter for the Future of AI?

Cisco and NVIDIA are working with TE Connectivity and Amphenol to develop the cages, connectors, and modules needed for the USFP ecosystem. More partners may join as the technology develops. Cisco says the approach preserves key benefits of pluggable optics, including replaceable components, supplier choice, established manufacturing processes, and easier maintenance.

They also plan to create an open industry standard specification through a community-driven and multi-source project. The current collaborations among these firms include attempts at making compatible NVIDIA Spectrum-X Ethernet networking and Cisco Silicon One Ethernet networks.

As Business Fortune observes,  USFP could help data centers support larger AI models, faster inference, and increasingly complex AI agents. Its wider adoption will depend on industry participation, system compatibility, and implementation. If successful, the technology could provide a practical route to higher-capacity AI networks without forcing operators to abandon established infrastructure and operational practices.

FAQs

What is USFP technology?

USFP stands for Ultra-Dense Small Form Factor Pluggable. It is a planned high-density optical module designed to improve network bandwidth for AI data centers.

Why are NVIDIA and Cisco developing USFP?

They aim to help AI data centers increase network capacity and efficiency while preserving familiar deployment, maintenance, and infrastructure practices.

What is the maximum planned capacity of USFP?

The initial 16-lane design provides 6.4 Tbps of aggregate capacity. A future 32-lane configuration could reach 12.8 Tbps.

How could USFP benefit AI data analytics?

Faster network connections can help computing systems exchange large volumes of data more efficiently, supporting demanding AI training, inference, and analytics workloads.

When will USFP become available?

The announcement outlines design plans and ecosystem development rather than a confirmed commercial launch date. Availability will depend on further development, industry collaboration, and deployment readiness.

Sources: https://www.hpcwire.com/bigdatawire/this-just-in/digitalocean-launches-agent-learning-and-data-services-for-recursively-self-improving-ai-agents/ https://datacentremagazine.com/news/inside-cisco-and-nvidias-pluggable-optics-for-ai-factories https://blogs.cisco.com/news/cisco-and-nvidia-announce-usfp-expanding-the-path-to-ai-connectivity