AI networking and digital infrastructure company Cisco unveiled its Silicon One G300 networking chip and a new portfolio of switching systems and optics to support large-scale AI data centres. This comes as the company moves deeper into infrastructure for training, inference and real-time agentic AI workloads. The launch was announced at Cisco Live EMEA in Amsterdam, The Netherlands.
The Silicon One G300 is a 102.4 terabit-per-second chip that will power the company’s new Cisco N9000 and Cisco 8000 Ethernet systems. These systems are designed for hyperscalers, neocloud operators, sovereign cloud providers, service providers, and enterprises building AI clusters.
According to the company, the G300 silicon can support gigawatt-scale AI clusters and improve job completion time by 28% by increasing GPU utilisation across training and inference workloads.
“We are spearheading performance, manageability and security in AI networking by innovating across the full stack, from silicon to systems and software,” Jeetu Patel, President and Chief Product Officer at Cisco, said in a statement.
Cisco said the new systems are designed to support dense optics and advanced cooling to improve performance and reliability in large data centres. This will ultimately improve energy efficiency, lower operating costs, and simplify operations as AI infrastructure expands beyond large cloud providers.
Cisco said the liquid-cooled designs, combined with new optical modules, can improve energy efficiency by nearly 70% while delivering the same bandwidth that previously required multiple systems.
Cisco is also introducing new 1.6-terabit optics and 800-gigabit linear pluggable optics for AI networks. The company said the 800G linear pluggable optics reduce optical module power consumption by 50% compared with retimed optical modules and cut overall switch power by up to 30% when deployed on new platforms.
This new optics and Cisco said the Silicon One G300, the G300-powered systems and the new optics will begin shipping later this year.
The launch places Cisco more directly in competition with networking silicon suppliers, including Broadcom and NVIDIA, both of which are expanding their Ethernet and AI networking portfolios to serve large AI clusters.
Broadcom also recently launched the Tomahawk Ultra Ethernet switch, expanding its role in the AI infrastructure landscape. Regarding the evolving AI data centre ecosystem, Broadcom earlier told AIM that its focus is on providing merchant silicon based on open standards to accelerate the entire AI ecosystem.
It said that open standards and multi-sourcing are always preferred to walled gardens and single-sourcing, given the broader impact of open Ethernet architecture versus proprietary systems like NVLink.
Alongside the hardware, Cisco said it has upgraded its Nexus One platform to provide a unified management plane for AI networks, bringing together silicon, systems, optics, and software under a single operational framework.
Martin Lund, Executive VP of Cisco’s Common Hardware Group, said the network is becoming part of the AI compute stack itself. “It’s not just about faster GPUs; the network must deliver scalable bandwidth and reliable, congestion-free data movement,” he said.
ALSO READ: OpenAI Begins Testing Ads on Free and Go Plans in the US
