Cisco Takes On Nvidia and Broadcom With New AI Chip

SAN FRANCISCO: Cisco Systems has entered the AI chip wars with the Silicon One G300. This switching processor offers 102.4 terabits per second of bandwidth. It is built to handle data movement in massive AI clusters. Cisco unveiled the chip this week at Cisco Live EMEA in Amsterdam. The launch takes direct aim at Nvidia and Broadcom. These rivals are fighting for a share of the $600 billion AI infrastructure market.

The G300 is manufactured on TSMC’s 3-nanometer process. It introduces a feature called Intelligent Collective Networking. This technology combines shared packet buffering and path-based load balancing.

It also uses real-time telemetry to prevent traffic jams. These jams often stall AI training runs. The push for optimized infrastructure is driving innovation at the silicon level. Startups are also developing new chip technologies for better efficiency.

Cisco claims the chip completes AI computing jobs 28% faster. It achieves this by automatically rerouting data around congested links. This happens within microseconds. The chip also improves overall network utilization by 33%.

The G300 powers two new switching platforms: the Nexus N9000 and Cisco 8000. Both are available in fully liquid-cooled configurations. Cisco claims this improves energy efficiency by nearly 70%.

The chip supports up to 512 ports and 1.6 terabit Ethernet connections. This gives hyperscalers room to grow without replacing hardware. As AI clusters scale, data demand is causing a global memory chip shortage and driving up DRAM prices. Cisco plans to put the G300 on sale in the second half of 2026.

This move signals a strategic pivot toward the AI networking layer. Nvidia is embedding its own networking silicon into rack-scale systems. Broadcom is expanding its Tomahawk series simultaneously. A credible third competitor in switching silicon could reshape the market. It would impact both pricing and procurement strategies for cloud providers.