Nvidia and Cisco said on October 12 they plan to introduce USFP, a new pluggable optical transceiver format for AI data centers designed to raise bandwidth within the same network footprint while reducing networking’s share of overall power use.

The companies said the first planned USFP modules would use an octal small form factor pluggable (OSFP)-sized footprint and support 400 Gb/s per lane, with 16 full-duplex lanes for switches and SuperNICs.

Nvidia and Cisco are working with TE Connectivity, a Switzerland-based connectivity and sensor company, and Amphenol, a US-based maker of electronic connectors, on connectors and modules.

They also said they plan to create an open industry specification through a community-led multi-source project.

The announcement signals a move toward next-generation AI networking. Broadcom said 400G-per-lane optics are a path toward 200T-class switching capacity for future AI and cloud infrastructure.

🔗 Source: Nvidia

Recent Nvidia developments