AnswersPublished22 sources
Cisco and Nvidia Plan 400G-per-Lane USFP Optics for AI Data Centers
Cisco and Nvidia plan USFP, a pluggable format designed for 400 Gb/s per lane: the first planned 16 lane modules target 6.4 Tb/s, with a roadmap to 32 lanes and 12.8 Tb/s.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Cisco and Nvidia plan USFP, a pluggable format designed for 400 Gb/s per lane: the first planned 16 lane modules target 6.4 Tb/s, with a roadmap to 32 lanes and 12.8 Tb/s.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Cisco and Nvidia announce at the October 12–15, 2026, Open Compute Project Global Summit in San Jose about scaling AI data-center n. Article summary: Cisco and Nvidia announced plans for USFP, a higher-density, 400Gb/s-per-lane pluggable-optics format intended to scale AI Ethernet fabrics while retaining replaceable modules and customer choice; Cisco also highlighted . Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
At the October 2026 Open Compute Project Global Summit, Cisco and Nvidia outlined USFP, a planned high-density pluggable-optics format for AI Ethernet networks. The design targets 400 Gb/s-per-lane signaling and a path to 12.8 Tb/s, while keeping the serviceability of pluggable modules. The announcement describes a roadmap, not a product with a stated shipping date.2
9
11
USFP’s planned capacity and design
USFP stands for Ultra-Dense Small Form-Factor Pluggable. The first planned transceivers are described as having 16 full-duplex lanes in an approximately OSFP-sized footprint, with support intended for switches and SuperNICs. Cisco and Nvidia describe an initial 6.4 Tb/s configuration and a path to 12.8 Tb/s by expanding to 32 lanes.9
14
20
The companies claim that USFP can provide four times the bandwidth density and about 1.3 times the power efficiency of pluggable optics built around 200 Gb/s-per-lane signaling. Those are vendor-reported comparisons; the available sources do not provide independent benchmark results.8
11
A key part of the proposal is to increase capacity without abandoning replaceable optics. Cisco and Nvidia say the design is intended to preserve serviceability and customer choice, and they plan to develop an open industry specification with ecosystem partners including TE Connectivity and Amphenol.2
8
11
What is still a plan—not a deployment
The sources describe design plans and a proposed open specification, but do not give a USFP sampling or availability schedule.11 They also do not establish the claimed 100-watt thermal capability or provide enough detail to assess the proposed connector’s wear and signal-integrity benefits. The announcement materials cited here do not offer a direct technical comparison with co-packaged optics or explain how USFP relates to Cisco–Nvidia Spectrum-X integration.
That distinction matters: USFP is being presented as a pluggable form-factor roadmap. The available evidence does not show that the modules, or all of the associated ecosystem, are already shipping.2
9
11
Cisco’s Silicon One milestones
Cisco also outlined different stages of its Silicon One product roadmap:
- G302: Cisco introduced a 51.2 Tb/s switching chip with 200G SerDes aimed at top-of-rack deployments. The company said customer shipments were planned before the end of 2026.
4
- G300: The 102.4 Tb/s chip was available for customer sampling, while the first G300-powered systems were already shipping, according to the report.
- P200: Cisco said its P200-powered 51.2 Tb/s systems were in production at hyperscalers including Microsoft and Alibaba Cloud.
Alibaba-specific reporting says its P200-powered Quantum Cat routers had been deployed at scale in the company’s eCORE backbone.29 Cisco also said Live Protect support for Cisco SONiC customers was planned to begin in mid-2027.
30
Other AI-networking announcements at OCP
The summit included several related infrastructure announcements. Nvidia said it was contributing Vera Rubin rack and compute-tray designs to OCP through its MGX architecture, and announced Falcon Transport support on future ConnectX-10 SuperNICs and BlueField DPUs with Google.15
Meta described two IceCube switch variants—one using Nvidia Spectrum-6 and another using Broadcom Tomahawk6C—and said both run Meta’s FBOSS network operating system.36 Separately, Lumentum, Astera Labs, and Corning announced a rack-scale PCIe-over-optics demonstration combining Astera Labs’ Scorpio Smart Fabric Switch, Lumentum’s 1060 nm VCSEL technology, and Corning fiber.
34
Together, these announcements show companies pursuing higher-capacity links, open hardware designs, and different approaches to scaling AI networks. For USFP specifically, the firmest conclusion is narrower: Cisco and Nvidia have proposed a higher-density, serviceable pluggable format, but its future specifications and availability remain to be established.2
9
11
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Cisco and Nvidia Plan 400G-per-Lane USFP Optics for AI Data Centers"?
Cisco and Nvidia plan USFP, a pluggable format designed for 400 Gb/s per lane: the first planned 16 lane modules target 6.4 Tb/s, with a roadmap to 32 lanes and 12.8 Tb/s.
Sources
blogs.cisco.comCisco and NVIDIA Announce USFP: Expanding the Path to ...
newsroom.cisco.comNVIDIA and Cisco Introduce New Pluggable Optics Form Factor to Scale AI Factories
storagereview.comNVIDIA and Cisco USFP Optics Target 400Gb/s per Lane, With MGX ...
convergedigest.comNVIDIA and Cisco Unveil 12.8T USFP Pluggable Optics