Now loading...
AMSTERDAM, Feb 10 (Reuters) – Cisco Systems unveiled a new high-speed networking chip and related hardware on Monday, aimed at supporting the growing demands of artificial intelligence data centers as the technology expands beyond major cloud providers.
The companys Silicon One G300 chip offers 102.4 terabits per second of switching capacity, designed to handle large-scale AI clusters for tasks like model training and real-time processing. It powers updated versions of Ciscos N9000 and 8000 series routers and switches, which are built to deliver hyperscale performance while improving energy use and operational simplicity for a range of users including cloud services, enterprises and telecom providers.
Key to these advancements is a focus on efficiency, with the new systems available in fully liquid-cooled configurations that, combined with advanced optical modules, can cut energy consumption by almost 70 percent compared to older setups. The G300 chip incorporates features like shared packet buffering and intelligent load balancing to boost network use by 33 percent and shorten AI job times by 28 percent, according to Cisco.
Alongside the hardware, Cisco enhanced its Nexus One management platform to provide a single control layer for AI networks, whether on-site or in the cloud. This update includes tools for faster deployment, better security and integration with analytics software like Splunk, helping operators monitor from network to graphics processing units without moving data externally.
Jeetu Patel, Ciscos president and chief product officer, described the innovations as foundational for AI infrastructure, supporting diverse customers shifting to intelligence-driven operations. Martin Lund, executive vice president of Ciscos hardware group, emphasized the networks role in enabling reliable data flow, making it integral to compute performance.
The G300 chip stands out for its programmability, allowing post-deployment upgrades to adapt to new needs and embedding security at the hardware level. Cisco, which launched the Silicon One family in 2019, positions it as a versatile architecture spanning multiple network environments.
New optical technologies include 1.6 terabit per second modules and linear pluggable optics that halve power use in AI connections, potentially reducing overall switch energy by 30 percent. Cisco also expanded its lineup with systems based on the earlier P200 chip, adding modular cards and support for long-haul optics to serve various routing roles.
Cisco plans to ship the G300-powered products and optics later this year. The company collaborates with partners such as AMD, Intel, Nvidia and NetApp to integrate networking with compute and storage for optimized AI setups.
Industry analysts praised the developments. Matt Eastwood of IDC noted they address key challenges in scaling AI beyond hyperscalers, focusing on cost and sustainability. Alan Weckel of 650 Group predicted the technology could drive the data center switch market past 100 billion dollars annually as speeds hit 1.6 terabits.
Partner endorsements highlighted interoperability and performance. Yousuf Khan from AMD stressed open standards for flexible AI builds, while Anil Nanduri of Intel pointed to efficient inference clusters using the new switches with their accelerators.
