Cisco targets denser pluggable optics alongside G302 switch silicon
Illustrative image: Unsplash contributor on Pexels
Cisco is pursuing two changes to AI-network switches: a 51.2Tb/s switching chip with 200G SerDes, and a proposed pluggable optical interface running at 400Gb/s per lane. They address different parts of the bandwidth path and are at different stages of development.
Silicon One G302 is intended for top-of-rack systems. Cisco describes support for 1.6T optics and direct-to-chip liquid-cooling integration, with shipments planned before the end of 2026.
Ultra-Dense Small Form Factor Pluggable, or USFP, is being developed with NVIDIA as an open optical-module format. Unlike G302, it is a proposed interface rather than silicon with a stated shipping plan.
More bandwidth through the same frontage
Cisco and NVIDIA say USFP could deliver four times the bandwidth density and 1.3 times the power efficiency of 200Gb/s-per-lane optics. Those are projected comparisons, not measurements from a generally available module range. The companies intend to develop an open specification through a community-led multi-source project.
The attraction of a denser pluggable module is that the optical interface remains removable. A failed transceiver can be exchanged without replacing the switch ASIC. It also retains a separate module-selection and service workflow, while the electrical connections between switch and faceplate remain part of the equipment design.
Raising lane rate places more weight on the complete electrical channel: package escape, PCB loss, connector transitions and module interface. A module with a smaller frontage does not remove those constraints. It changes how much bandwidth the enclosure can accommodate once that channel is engineered successfully.
Cooling becomes part of the platform
Cisco's reference to liquid cooling alongside G302 is significant because a switch enclosure contains more than one concentrated heat source. The ASIC, optical modules and supporting power circuitry all contribute to the thermal design, but they occupy different locations and have different service needs.
Liquid cooling of the central chip does not establish the cooling arrangement for modules at the faceplate. Those modules must remain accessible for replacement, with their electrical connectors and heat-removal paths incorporated into the enclosure. Increasing bandwidth per opening can reduce the frontage needed for a given capacity, but the proposed USFP interface still requires a specification that equipment and module suppliers can implement consistently.
