Cisco Expands AI Infrastructure with New Silicon and Optical Standards

Cisco is targeting the growing need for "scale-across" networking, which coordinates AI workloads across geographically dispersed facilities. By integrating its Silicon One P200 architecture with coherent optics, the company aims to turn disparate sites into a unified compute fabric. This approach has already seen practical application, with Microsoft and Alibaba confirming the production deployment of routers powered by the P200, which supports 51.2 Tbps of routing capacity.

To address the 1.6T era, Cisco is expanding its Silicon One portfolio with the G302, a 51.2T switching silicon featuring 200G SerDes and direct-to-chip liquid cooling. This component is engineered for top-of-rack deployments to maximize throughput. Complementing these hardware updates, Cisco and NVIDIA are collaborating on the Ultra-Dense Small Form Factor Pluggable (USFP) module, an open industry specification intended to quadruple bandwidth density compared to current standards while enhancing power efficiency.

Beyond hardware, the company is bolstering its commitment to the open-source SONiC operating system. New security integrations, such as Live Protect, are scheduled for mid-2027 to help customers mitigate vulnerabilities without interrupting active network operations. Additionally, the new Silicon One A100-powered Cisco 8001 management system aims to simplify out-of-band network oversight, replacing fragmented legacy tools with a unified, programmable interface for large-scale GPU clusters.