Cisco launches 102.4Tbps networking to scale AI data centers

Cisco has launched its Silicon One G300, a 102.4 terabits-per-second switching chip designed to enhance AI data center networking performance, efficiency, and GPU utilization. Unveiled at Cisco Live EMEA in Amsterdam, the G300 powers new Cisco N9000 and 8000 data center switches aimed at enterprises, service providers, and sovereign cloud builders expanding AI infrastructure beyond Big Tech. The platform addresses the growing network bottlenecks caused by large, distributed AI models by improving data movement between GPUs, which directly impacts AI job completion times and compute capacity. Cisco claims the G300 chip reduces job completion time by 28% and increases network utilization by 33% compared to non-optimized networking paths. At the core of Cisco’s solution is Intelligent Collective Networking, which integrates a shared packet buffer, path-based load balancing, and real-time telemetry to prevent packet loss and congestion in large AI workloads. The G300 is programmable for post-deployment updates and includes built-in security features