USFP and the Road to 400G Per Lane: What Changes and What Doesn't

On October 12, 2026, NVIDIA and Cisco announced plans for USFP, short for Ultra-Dense Small Form Factor Pluggable. It is a new optical module built for 400Gb/s per lane, aimed at the next generation of AI networks.

It is a design plan, not a product. The announcement came with no ship dates and no pricing. Even so, it shows where high-speed networking is heading, and it is worth understanding before it shows up in your next design review.

What was announced

The first planned USFP modules fit 16 full-duplex lanes into a footprint the size of today’s OSFP module. At 400Gb/s per lane, that is 6.4Tb/s of aggregate capacity, which Cisco describes as four times the bandwidth density of today’s 1.6T pluggables. The design can scale to 32 lanes, which is where the 12.8Tb/s figure comes from. NVIDIA and Cisco also cite roughly 1.3x better power efficiency, tied to 400G SerDes technology.

Cisco adds that USFP is designed to support both 200G and 400G SerDes based systems and ASICs, giving customers a path from today’s 1.6T fabrics.

Two engineering details stand out. NVIDIA says USFP uses a new two-dimensional array connector with a two-step insertion design, meant to reduce contact wear and improve signal integrity. It also supports up to 100 watts of power dissipation without a dedicated cooling circuit inside the transceiver, which simplifies switch design.

TE Connectivity and Amphenol are working with NVIDIA and Cisco on cages, connectors, and modules, and more partners are expected to join.

Where this takes the industry

Three shifts are worth watching.

Fewer tiers for the same compute. When a switch runs out of ports, networks add layers, and every layer adds latency, power draw, and cost. More bandwidth per port and more ports per switch let operators connect more systems before they need another tier.

Power becomes the real constraint. NVIDIA and Cisco frame the goal as shrinking networking’s share of the total power budget in an AI facility. Power spent moving data is power that is not running compute, so efficiency per bit matters as much as raw speed.

An open, multi-source ecosystem. NVIDIA and Cisco plan to publish an open industry specification through a community-led multi-source project. Cisco’s own announcement lists interoperability across suppliers, independently replaceable modules, and mature supply chains among the advantages operators value, and says USFP is meant to extend them. That is Cisco describing its own vision for the ecosystem, not an endorsement of any particular supplier. Still, multiple suppliers building to a shared spec is how buyers get options, and how a new form factor reaches real volume.

Pluggable optics are also not the only path. Cisco notes that the industry continues to innovate across pluggable, near-packaged, and co-packaged approaches, each with different tradeoffs.

The phased view

Nobody has published a timeline, so treat this as a sequence, not a schedule.

• Now: 1.6T pluggables, built on 200Gb/s lanes, are the baseline USFP is measured against. Nothing in the announcement changes what ships today.
• Next: The multi-source project forms, additional partners join, and the spec takes shape. Sample availability and host support are the milestones to watch. No dates have been announced.
• Later: The design is meant to scale to 32 lanes and 12.8T per module. No dates have been announced.

What it demands of the supply chain

Faster lanes generally mean tighter margins. Signal integrity, power, and thermal behavior all get less forgiving as speeds climb, and a module that meets a spec on paper still has to work in the specific switch it is plugged into.

That is why validation matters more with each generation. At Dromeda, every product goes through 100% application testing across more than 300 replicated OEM environments, and our production failure rate is under 0.02%. Whatever generation you are buying, the test that counts is how the module performs in your switch.

Questions to ask any optics supplier about USFP

If a supplier brings up USFP in your next conversation, these are worth asking:

• Timeline: When will samples be available, and to whom? Be skeptical of firm dates before the spec is published.
• Specification: Is the multi-source specification public, and who is participating?
• Host support: Which switches and network adapters will support USFP, and when?
• Compatibility: Does the module work with both 200G and 400G SerDes hosts?
• Validation: How are modules tested in my specific switch, and what happens when a link does not come up?
• Today’s build: What are the lead times and compatibility guarantees on what I am buying now?

Where Dromeda stands

What to watch

• Spec and membership: who joins the multi-source project, and when the open specification is published.
• Samples and host support: the first availability of USFP modules, and which switches and network adapters support them.
• Partner expansion: additional cage, connector, and module suppliers beyond the two announced.
• OCP: Cisco tied the announcement to its presence at the Open Compute Project event, so more detail may surface in sessions there.
• Other architectures: how pluggable, near-packaged, and co-packaged approaches develop side by side.
• Dates: none have been announced, so be cautious with anyone quoting them.

Planning a build or refresh?

The useful conversation right now is about what ships today: compatibility, lead times, and how your optics will be validated in your environment. Talk to our team about what you are planning.