Cisco launches Silicon One G300 chip to speed AI data-center networks

SAN FRANCISCO — Cisco Systems unveiled its Silicon One G300 switch chip and a new router aimed at moving data faster through massive AI data centers, positioning the company against Broadcom and Nvidia in the fast-growing AI infrastructure market, according to Reuters.

Cisco said the G300 uses TSMCs 3-nanometer process, will go on sale in the second half of the year, and is designed to keep AI systems talking to each other across hundreds of thousands of links while handling traffic spikes; the company says it can cut job completion time by about 28% by rerouting data around network problems, Reuters reported.

What Cisco says it adds

In its announcement at Cisco Live EMEA, the company said the 102.4 Tbps Silicon One G300 will power new Cisco N9000 and Cisco 8000 systems and support liquid-cooled designs and higher-density optics to improve energy efficiency. Cisco also said its Nexus One management plane is being updated to simplify AI fabric deployment across on-prem and cloud environments, according to Ciscos news release.

How it stacks up

The Register reported that the G300 packs 512 200-Gbps SerDes lanes and can support AI clusters up to roughly 128,000 GPUs using about 750 switches, versus 2,500 previously. The chip can aggregate lanes for port speeds up to 1.6 Tbps. Cisco says its collective networking engine improves link utilization by about 33% and reduces time to completion by up to 28%, though those are vendor claims.

Why it matters

Networking is a bottleneck for large-scale AI training and inference, and Reuters noted the competition with Broadcoms Tomahawk and Nvidias networking silicon as companies chase a multiyear AI infrastructure spending boom.

How we report: We select the days most important stories, confirm facts across multiple reputable sources, and avoid anonymous sourcing. Our goal is clear, balanced coverage you can trust”because transparency and verification matter for informed readers.