Cisco has strengthened its push into artificial intelligence infrastructure with the launch of Silicon One G300, a next-generation switching chip designed to support the massive scale and performance demands of AI training and inference workloads.
The new 102.4 terabits-per-second (Tbps) silicon marks a significant upgrade in Cisco’s strategy to transform the network into a foundation for AI innovation. The G300 will power upcoming Cisco Nexus 9000 (N9000) and Cisco 8000 platforms, which are purpose-built for large-scale AI clusters requiring extreme bandwidth, density and efficiency.
At the core of the new platform is Cisco’s Intelligent Collective Networking technology, which combines a fully shared packet buffer, path-based load balancing and proactive telemetry. According to Cisco, this architecture delivers up to 33 percent higher network utilisation and enables AI jobs to complete up to 28 percent faster compared with traditional, non-optimised networks—helping organisations extract more value from expensive GPU resources.
Security and flexibility are also embedded at the hardware level. The G300 integrates built-in security features and high programmability, allowing operators to adapt to future requirements without replacing existing infrastructure.
Alongside the new silicon, Cisco introduced liquid-cooled and air-cooled 102.4T systems that deliver nearly 70 percent improvement in energy efficiency over previous generations. The company also unveiled new 1.6T optics and 800G Linear Pluggable Optics (LPO), which increase bandwidth while cutting optical power consumption by up to 50 percent, reducing overall switch power usage by about 30 percent.
These advancements are aimed at hyperscalers, enterprises, neocloud providers and service operators building AI-ready data centres at scale.
On the software front, Cisco enhanced Nexus One, its unified AI networking platform, to simplify operations across on-premises and cloud environments. New capabilities include AI workload observability, integration with Splunk, API-driven automation and AgenticOps features that apply guided, AI-driven troubleshooting to reduce operational complexity. Cisco said the platform also supports sovereign and compliance-sensitive deployments.
Cisco confirmed that the Silicon One G300, the associated switching systems and the new optics are scheduled to ship later this year. The company is also collaborating with ecosystem partners including NVIDIA, AMD, Intel, NetApp, DDN and VAST to deliver fully integrated AI infrastructure stacks for customers worldwide.
