Nvidia and Cisco introduce USFP optical interface for AI data centers
Nvidia and Cisco have jointly unveiled Ultra-Dense Small Form Factor Pluggable (USFP), a new optical transceiver standard supporting 400Gb/s per lane, the first interface to do so. Cisco says the module is designed to eventually reach total transfer rates of 12.8Tb/s, offering four times the bandwidth of existing 200Gb/s optics and 1.3x better power efficiency. Initial USFP transceivers will have 16 full-duplex lanes and work with both switches and Nvidia's SuperNIC adapters, which move data between GPUs using the RoCE protocol.
GoKawiil's interpretation of the reporting above, not reported fact.
The move suggests Nvidia and Cisco are positioning networking infrastructure as a key bottleneck to solve as AI training clusters scale up and demand ever more data throughput between GPUs. By building the standard around an established deployment model, the companies may be aiming to ease adoption for data center operators already using similar form factors. Future expansion to 32 or more lanes could let AI networks support more high-speed endpoints without adding extra network tiers, though full 12.8Tb/s capacity remains a longer-term goal.
- USFP is the first optical interface to support 400Gb/s per lane, per Cisco.
- The standard aims for up to 12.8Tb/s total bandwidth with better power efficiency than current optics.
- Initial 16-lane transceivers will support switches and Nvidia's GPU-focused SuperNIC adapters, with 32-plus lane versions planned next.
Source: techspot.com — Alfonso Maruccia, 2026-10-15
Published there as: “Nvidia and Cisco unveil next-gen network interface built for the AI data center boom”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.