USFP optical transceiver promises 12.8Tb/s AI data center scaling
Nvidia and Cisco have unveiled the USFP optical transceiver, a pluggable interface supporting 400Gb/s per lane and a path to 12.8Tb/s total bandwidth to scale AI infrastructure.

Nvidia and Cisco have introduced the Ultra-Dense Small Form Factor Pluggable (USFP) optical transceiver, designed to increase bandwidth in data centers for AI workloads. The new interface supports 400Gb/s per lane, with an initial 16-lane design delivering 6.4Tb/s aggregate capacity and a clear scaling path to 32 lanes for 12.8Tb/s. This total capacity is four times that of optical interfaces offering 200Gb/s per lane. USFP uses 400G SerDes technology, offering approximately 1.3x greater power efficiency than previous solutions.
Several development steps remain before USFP reaches its full 12.8Tb/s capacity. The first USFP-based optical transceivers are expected to feature 16 full-duplex lanes in an octal small form factor footprint. Cisco expects the complete module to eventually achieve the 12.8Tb/s overall bandwidth target. The design aims to help customers scale AI infrastructure efficiently without redesigning networks or abandoning established operational models.
Pluggable design preserves existing data center operations
USFP maintains a pluggable form factor compatible with existing switches and Nvidia's SuperNIC network adapters. This preserves operational simplicity and ecosystem flexibility for data center operators. The interface supports independently field-replaceable modules, mature supply chains, and interoperability across suppliers.
The pluggable design also supports power dissipation of up to 100 watts without requiring a dedicated cooling circuit inside the transceiver. It features a two-step insertion design intended to improve signal integrity and reduce contact wear. Cisco acknowledged that pluggable modules such as USFP will form part of a broader ecosystem including near-packaged, co-packaged, and other emerging optical architectures.
Nvidia and Cisco collaboration enables next-gen optical networking
The collaboration between GPU maker Nvidia and networking specialist Cisco marks a joint effort on AI networking innovation. Cisco positions USFP as access to next-generation optical technology for data centers. "USFP is designed to give data center customers access to next-generation optical networking technology," Cisco said. The companies report that USFP is the industry’s first 400G-per-lane capable pluggable form factor.
After developing the USFP specification, Nvidia and Cisco began working with TE Connectivity and Amphenol to create remaining components. Third-party companies are now making cages, connectors, and modules for the interface, with more partners expected to join the effort. This extends the benefits of the pluggable ecosystem to higher capacity AI fabrics and denser switch and compute equipment.
Technical specifications and scaling path
The technical foundation of USFP supports both 200G and 400G SerDes based systems and ASICs. Its design provides a path to 12.8T of overall pluggable capacity, building on existing 1.6T fabrics. The interface supports SuperNIC network adapters, which are specialized accelerator cards designed by Nvidia and its partners to move data directly between GPUs using RDMA over Converged Ethernet (RoCE).
| Feature | Specification |
|---|---|
| Per-lane bandwidth | 400Gb/s |
| Initial lane count | 16 full-duplex lanes |
| Initial aggregate capacity | 6.4Tb/s |
| Scaling target | 32 lanes |
| Target aggregate capacity | 12.8Tb/s |
| Power dissipation support | Up to 100W |
| Efficiency gain | ~1.3x greater power efficiency |
| Form factor | Pluggable, octal small form factor |
The next phase will expand the number of lanes to 32 or more. This expansion could allow AI networking systems to connect additional 1.6Tb/s endpoints without adding more network tiers.





