NVIDIA and Cisco Unveil USFP Optics: 5 Facts About the 400Gb/s-Per-Lane Module

AI data centres run on networks as much as chips, and NVIDIA and Cisco have just proposed a new way to wire them. On October 12, 2026, the two companies…

October 14, 2026
4 min read

AI data centres run on networks as much as chips, and NVIDIA and Cisco have just proposed a new way to wire them. On October 12, 2026, the two companies announced plans for the Ultra-Dense Small Form Factor Pluggable (USFP), a new optical module designed to scale what they call AI factories. The press release gives no shipping date, so this is a roadmap, not a product you can order.

What USFP Is?

USFP is a high-density pluggable optical module built around 400Gb/s per lane, described as the first module designed to support 400Gb/s-per-lane electrical signals. It reaches up to 12.8Tb/s of capacity. Compared with today’s 200Gb/s-per-lane optics, the companies claim 4x the bandwidth density and about 1.3x the power efficiency. Background on both firms is on NVIDIA’s Wikipedia page and Cisco’s; the original release is on NVIDIA’s newsroom.

USFP at a Glance

FeatureDetail
Lane speed400Gb/s per lane
CapacityUp to 12.8Tb/s
Density vs 200G optics4x bandwidth density
Power efficiencyAbout 1.3x better
First modules16 full-duplex lanes, OSFP-sized footprint
FutureCan expand to 32 or more lanes
ConnectorNew two-dimensional array, two-step insertion
Power handlingUp to 100 watts without separate embedded cooling circuit
PartnersTE Connectivity, Amphenol; more expected

Why It Matters for AI Factories

Training large AI models means thousands of chips talking at once, and the network can become a bottleneck and a power hog. NVIDIA and Cisco say USFP helps operators add compute without adding network tiers and lowers networking’s share of the power budget. The module targets both switches and SuperNICs, and is meant to work with NVIDIA Spectrum-X Ethernet and Cisco Silicon One-based networks. Our NVIDIA page tracks the company’s other AI hardware news.

In plain terms, every optical module is a small box that converts electrical signals to light so data can travel between racks. Packing more lanes into the same footprint means more data through the same switch, which is why density and power efficiency are the headline numbers.

Design Details

The new connector uses a two-dimensional array with a two-step insertion design. The companies say this reduces contact wear, improves signal integrity and makes insertion force more predictable. It also improves performance at 200Gb/s per lane, so it is not useful only in a future 400G world. The module can support up to 100 watts of power dissipation without a dedicated cooling circuit and works across a range of optical technologies and distances.

An Open Standard Is Planned

NVIDIA and Cisco say they are working with TE Connectivity and Amphenol on cages, connectors and modules, and plan an open industry specification through a community-led multi-source agreement, referred to as a USFP MSA. Executives quoted include Gilad Shainer of NVIDIA and Bill Gartner of Cisco.

What Is Not Confirmed

The release leaves several things not confirmed:

  • A shipping date or when systems using USFP will be available.
  • Pricing.
  • Which other companies will join the multi-source project.
  • Independent testing of the 4x density and 1.3x efficiency claims, which are the companies’ own figures.

Quick Answers

What does USFP stand for? Ultra-Dense Small Form Factor Pluggable.

Who announced it? NVIDIA and Cisco, on October 12, 2026.

How fast is it? 400Gb/s per lane, up to 12.8Tb/s.

Can I buy it? No availability has been announced.

Source: NVIDIA newsroom press release, October 12, 2026. Performance figures are company claims.

It is worth noting how this fits the wider industry move. Hyperscalers are building ever larger clusters, and each generation of GPUs needs faster links to avoid sitting idle while waiting for data. A shared, open module standard could also lower costs by letting several suppliers make compatible parts, which is the usual reason companies publish a multi-source agreement. Operators will still want to see real-world power figures, reliability data and interoperability tests before committing, so the next milestones to watch are the formal specification, first partner samples and any named customers.

Follow us on Google News Get real-time updates & exclusive tech coverage
Follow

Leave a Reply

Your email address will not be published. Required fields are marked *