Cisco, NVIDIA Release Compact Networking Part for AI Data Centers
Cisco and NVIDIA have introduced a new optical module called USFP that supports 400 Gb/s per lane and up to 12.8 Tb/s total capacity. The companies say it can help data centers connect more AI computing systems while using less power and keeping a small footprint. The article frames the technology as a way for smaller businesses to expand AI infrastructure without major overhauls.
Cisco and NVIDIA’s USFP is an optical pluggable aimed at dense AI data-center links. It carries 400 Gb/s on each lane and can reach 12.8 Tb/s overall, twice the per-lane rate of existing 200 Gb/s modules, while taking up less space.
The design is said to use about 1.3 times the power efficiency of 200 Gb/s alternatives and to fit established deployment practices, which may reduce retraining needs. Cisco and NVIDIA are also working with TE Connectivity and Amphenol on ecosystem support; adopters may still need to check for network bottlenecks and component compatibility.
The USFP could affect small businesses and data-center operators by making denser AI networking more attainable, potentially lowering energy and space costs. Workers who maintain networks may see simpler module replacement, while suppliers and integrators may face pressure to ensure compatibility. Broader access to AI computing could influence how smaller organizations adopt AI tools, though benefits may depend on upgrading surrounding infrastructure and managing costs.