Cisco unveils Silicon One G300 to power next-generation AI networks

Launched at Cisco Live EMEA, the new 102.4 Tbps networking chip and liquid-cooled systems target AI data centre bottlenecks, as Cisco takes on Broadcom and Nvidia in the fast-growing AI infrastructure market.

Cisco Systems has introduced a new high-performance networking chip and router platform aimed at accelerating artificial intelligence infrastructure. Announced at the Cisco Live EMEA conference on February 10, the launch signals the company’s intensified push into the rapidly expanding AI data centre market.

At the core of the announcement is the Silicon One G300, a next-generation switching chip built to manage the enormous data flows generated by large-scale AI models. As organisations deploy increasingly complex AI systems, connecting thousands of processors efficiently has become a major technical hurdle. Cisco says its new chip is designed to ease these data bottlenecks and improve overall network performance.

Built for the AI-driven data centre

The Silicon One G300 delivers up to 102.4 terabits per second of bandwidth, enabling operators to build extensive AI clusters with fewer networking components. By reducing the number of switches required, Cisco aims to simplify infrastructure design while enhancing speed and reliability.

Alongside the chip, Cisco introduced upgraded N9000 and 8000 series networking systems featuring fully liquid-cooled architectures. The company stated that, when combined with advanced optical technologies, these systems can significantly reduce power consumption, improving energy efficiency by nearly 70%. Sustainability has become a growing concern as AI workloads demand higher computing power and increased electricity usage.

Cisco executives noted that while early AI infrastructure investments have largely come from hyperscale cloud providers, broader enterprise adoption is expected to drive the next wave of spending.

Intensifying competition in AI infrastructure

The launch places Cisco in direct competition with established players such as Nvidia and Broadcom. While Nvidia dominates the AI processor segment and promotes its InfiniBand networking technology, Cisco is advancing Ethernet-based solutions, positioning them as scalable and widely compatible alternatives.

Cisco also introduced AgenticOps, a software suite designed to automate and manage complex AI networking environments. The company claims performance gains of up to 28% in AI training job completion times compared with competing technologies.

With the AI infrastructure market projected to grow substantially in the coming years, Cisco’s latest innovation represents a strategic effort to strengthen its role in the evolving digital ecosystem. The Silicon One G300 is expected to become commercially available in the second half of 2026.