Cisco Unveils New AI Networking Chip and Router.

Cisco Systems has launched a powerful new networking chip and accompanying router systems, directly challenging Broadcom and Nvidia in the race to capture part of the projected $600 billion AI infrastructure market.

Announced at Cisco Live EMEA in Amsterdam on 10 February 2026, the Silicon One G300 delivers 102.4 terabits per second of Ethernet switching capacity in a single device. This switch silicon targets massive AI cluster networks, enabling efficient communication among hundreds of thousands of links that connect chips training and running AI models.

Cisco engineers built the G300 on Taiwan Semiconductor Manufacturing Company's 3-nanometre process. The chip incorporates innovative "shock absorber" features that monitor congestion, share a collective network map across devices, and reroute data within microseconds during traffic spikes. Martin Lund, executive vice president of Cisco's common hardware group, explained these capabilities in an interview: the mechanisms prevent AI networks from bogging down, allowing some AI computing jobs to complete up to 28% faster.

The G300 powers new Cisco N9000 and Cisco 8000 systems, which support high-density optics, innovative liquid cooling, and designs that maximise GPU utilisation. These advancements aim to scale AI data centres to gigawatt levels for training, inference, and emerging agentic workloads, where AI agents operate autonomously. Cisco positions the technology as the networking foundation for the "agentic era" of AI, emphasising high performance, security, reliability, and lower total cost of ownership.

Consider the pressure on data centre operators today: sudden bursts in AI demand can throttle performance, much like traffic jams halt commuters during rush hour. Cisco's approach addresses this by enabling networks to adapt dynamically, potentially saving enterprises significant time and energy as they scale operations.

The company also enhanced Nexus One software to simplify AI network management, whether on-premises or in the cloud. This reduces complexity that often delays adoption in enterprises.

Systems powered by the G300 target hyperscalers, neoclouds, sovereign clouds, service providers, and enterprises. The Silicon One G300 SDK is now available, with first customer designs underway and system availability slated for the second half of 2026.

Author:Oje. Ese