Cisco, NVIDIA Team on 400G USFP Optical Standard for AI Data Centers

The Cisco Systems and NVIDIA announcement lands as AI data centers face a wiring bottleneck potentially more constraining than compute. By targeting a new USFP standard designed to quadruple bandwidth density, the partners are addressing the physical layer that connects AI infrastructure, an area where Cisco's networking and security gear already sits.

The collaboration is framed as part of a broader Cisco story: the company has a reported market capitalization of about $453.7b and builds networking and security gear for global internet and cloud infrastructure. Working with NVIDIA, TE Connectivity and Amphenol on an open optical standard is presented as evidence for the view that Cisco is embedded in the AI networking cycle rather than just selling generic routers.

The same move carries a counterpoint. Pushing 51.2T and 102.4T systems, 1.6T optics and new 400G modules intensifies the hardware mix and memory cost headwinds that analysts already flag as a margin constraint, especially with Ethernet heavyweights like Arista and Juniper also chasing AI fabrics. The bull case rests on Cisco extending AI-focused networking and full-stack security across hyperscale, enterprise and public sector customers while absorbing those headwinds without eroding operating margins.