NVIDIA and Cisco Plan USFP Pluggable Optics for AI Fabrics

Executive Summary

• NVIDIA and Cisco announced plans for Ultra-Dense Small Form Factor Pluggable (USFP), a high-density optical module designed for 400Gb/s-per-lane signaling, according to NVIDIA Newsroom.
• USFP is designed to support up to 12.8Tb/s of overall capacity, letting AI factory builders scale Ethernet AI fabrics without adding network tiers, per NVIDIA Newsroom.
• The companies claim 4x the bandwidth density and approximately 1.3x the power efficiency versus 200Gb/s-per-lane pluggable modules, NVIDIA Newsroom states.
• NVIDIA and Cisco are working with TE Connectivity and Amphenol on cages, connectors and modules, and plan an open industry specification through a community-led multi-source project, according to NVIDIA Newsroom.

Key Takeaways

• USFP is a planned form factor, not a shipping product. The first transceivers are described as planned, and the specification effort is described as upcoming.
• The design target is density per switch, not raw speed alone. NVIDIA and Cisco frame the problem as connecting more 1.6Tb/s endpoints inside the same network footprint.
• Vendor performance claims — 4x bandwidth density, 1.3x power efficiency — are stated as design intent tied to USFP-based modules, not measured results from deployed fabrics.
• Ecosystem breadth is the explicit strategy. TE Connectivity and Amphenol anchor the supply side, and the companies say additional partners are expected.

NVIDIA USFP Targets Switch Capacity Limits in AI Fabrics

NVIDIA and Cisco announced plans on October 12, 2026 for Ultra-Dense Small Form Factor Pluggable, a pluggable optical module designed around 400Gb/s-per-lane electrical signals. The stated problem is structural: as AI models grow and inference demand expands, AI factories must connect more computing systems while managing network complexity, power and costs. When a switch reaches its capacity, further expansion can require additional switch layers and optical links, which the two companies say increases latency, power consumption and costs.

USFP is positioned as an alternative to adding tiers. NVIDIA and Cisco say the form factor is designed to let builders move from 200Gb/s-per-lane to 400Gb/s-per-lane Ethernet AI fabrics while extending a deployment model already proven at scale across multiple generations of data center networks. The first planned USFP-based transceivers would pack 16 full-duplex lanes in an octal small form factor (OSFP)-sized footprint supporting both switches and SuperNICs, with the ability to expand to 32 and more lanes later.

The capacity figure NVIDIA and Cisco cite is 12.8Tb/s of overall capacity. By doubling both lane count and bandwidth per lane, the companies say AI fabrics can connect more 1.6Tb/s endpoints without adding network tiers. That framing matters commercially: the pitch is not that operators need faster optics in isolation, but that they can raise compute scale inside an existing network footprint while reducing networking's share of the AI factory power budget.

NVIDIA Connector and Cooling Design Choices

The technical differentiation NVIDIA and Cisco describe sits in the connector and thermal path. USFP uses a new two-dimensional array connector with a two-step insertion design that engages high-speed contacts after the module is inserted. The companies say this reduces contact wear, improves signal integrity and makes insertion force more predictable. Because the connector is designed for 400Gb/s performance, they also claim improved performance at 200Gb/s per lane, and the architecture allows lane count to scale to 32 full-duplex lanes and beyond.

On power, NVIDIA and Cisco say USFP supports dissipation of up to 100 watts without needing a separate, dedicated embedded cooling circuit inside the transceiver, which reduces switch design complexity and improves reliability. All USFP configurations are described as supporting optical designs that reduce latency and power consumption by simplifying signal processing, with support for a broad range of optical technologies and connection distances for system designers.

Related: Global Logistics Market Size, Trends and Forecast Statistics by Country and Company 2026-2030

These are design claims attached to a planned module. The source does not report independent testing, deployed configurations or customer names, so the practical question for buyers is whether the connector, thermal and signal-integrity behavior holds at 400Gb/s per lane in volume production.

NVIDIA Ecosystem and Standards Ambitions for Pluggable Optics

NVIDIA and Cisco are collaborating with TE Connectivity and Amphenol to build USFP cages, connectors and modules, with additional partners expected. The shared form factor is intended to give suppliers a common foundation for developing high-density pluggable infrastructure for 400Gb/s-per-lane fabrics. The two companies plan to create an open industry specification for USFP through an upcoming community-led multi-source project, and Cisco's forward-looking statements reference establishment of a USFP MSA and expected benefits.

For deeper context, see our AI Chips analysis: "How Custom Silicon Is Reshaping the AI Chip Market in 2026".

The collaboration builds on prior work between the two vendors to ensure NVIDIA Spectrum-X Ethernet networking and Cisco Silicon One-based Ethernet networks interoperate, which the source says simplifies AI infrastructure deployment for enterprises, neoclouds and sovereign clouds.

Optics Signals

The source does not specify geography for any party involved, so no location detail is asserted here.

Additional coverage: Amylyx Avexitide Meets Phase 3 Hypoglycemia Target With 55% Efficacy

What This Means for Practitioners

For network architects and infrastructure buyers planning AI fabric expansions, USFP is a signal about direction rather than an available option. The decision it informs is tier planning: if 400Gb/s-per-lane pluggable optics arrive on the stated timeline, operators could connect more 1.6Tb/s endpoints per switch and defer added switch layers, latency and power draw. Until then, the practical steps are to track the planned multi-source specification effort, confirm interoperability between Spectrum-X and Cisco Silicon One-based Ethernet in current designs, and treat the density and efficiency figures as design targets pending production hardware and independent validation.

NVIDIA Implementation Risks

The source frames USFP as a plan, and the risks follow from that status. The primary risk is timing: NVIDIA and Cisco describe planned transceivers and an upcoming specification project, and Cisco's forward-looking statements explicitly cite the establishment of a USFP MSA as a matter that may differ materially from expectations. Ecosystem execution is a second risk — the form factor depends on TE Connectivity and Amphenol delivering cages, connectors and modules, and on additional partners joining. Third is operational dependency: NVIDIA notes reliance on third parties to manufacture, assemble, package and test its products, plus risks around design or manufacturing defects, changes in industry standards and interfaces, and market acceptance of partner products. Performance claims remain unverified by the source.

Editorial independence disclosure: this article was written independently from the subject companies. Source note: all facts above derive from NVIDIA Newsroom.