The new silicon will power the upcoming Cisco Nexus 9000 (N9000) and Cisco 8000 systems, which are specifically designed to meet the extreme performance, density, and efficiency demands of AI training and inference workloads.
Cisco announced a major step in its strategy to transform networks into AI innovation platforms with the launch of Silicon One G300, a 102.4 terabits per second (Tbps) switching chip designed for large-scale AI cluster deployments. The new silicon will power the upcoming Cisco Nexus 9000 (N9000) and Cisco 8000 systems, which are specifically designed to meet the extreme performance, density, and efficiency demands of AI training and inference workloads.
The G300 introduces an intelligent collective network that combines fully shared packet buffers, path-based load balancing, and proactive telemetry to improve performance across distributed AI clusters. Cisco says this improves network utilization by 33% and reduces job completion time by 28% compared to non-optimized networks, allowing organizations to maximize GPU output and improve AI datacenter economics. Security is built directly into the hardware and the chip is highly programmable, allowing for future upgrades without replacing the infrastructure.
Cisco also introduced water- and air-cooled 102.4T systems, which are nearly 70% more energy efficient compared to the previous generation. New 1.6T optics and 800G Linear Pluggable Optics (LPO) further improve bandwidth, reduce optical power consumption by up to 50%, and reduce overall switch power consumption by 30%. These advancements target hyperscalers, enterprises, neoclouds, and service providers building AI-enabled infrastructure.
On the software side, Cisco has enhanced Nexus One, a unified AI networking platform that simplifies operations across on-premises and cloud environments. New features include AgenticOps capabilities that streamline network management with AI job observability, Splunk integration, API-driven automation, and guided AI-driven troubleshooting. Cisco says this approach reduces operational complexity and supports sovereign and compliance-focused deployments.
Silicon One G300 and associated systems and optics are expected to ship this year. Cisco works with partners such as NVIDIA, AMD, Intel, NetApp, DDN, and VAST to deliver a unified AI infrastructure stack.
