Cisco (CSCO) has launched a large-scale product line of core network chips and accessories, targeting the next-generation AI network infrastructure market. On October 10th, at the "Cisco Live EMEA" event, Cisco unveiled the "Silicon One G300," an Ethernet-dedicated switching chip boasting a processing speed of 102.4 terabits per second, positioning it as a core solution for building large-scale AI clusters.
The newly released G300 chip not only focuses on bandwidth enhancement but also prioritizes maximizing network efficiency for AI workloads. Cisco emphasizes that the chip will be used in its mainstay switching systems, the Nexus 9000 and Cisco 8000 series, and will support both air-cooled and liquid-cooled models, simultaneously reducing power consumption and operational complexity.
As AI infrastructure expands beyond hyperscale cloud to encompass all enterprises, the need for network configurations to support GPU-centric large-scale data processing is growing rapidly. Kevin Walter Weber, Senior Vice President of Cisco's Data Center Business, explained, "Recent industry trends are rapidly shifting from AI focused on model training to agent applications and inference," with more and more enterprise data centers and on-premises clouds building their own AI clusters.
The G300 chip utilizes "Intelligent Convergence Networking" technology to more precisely control the bursty traffic unique to AI workloads. Cisco also released simulation data for this technology: by integrating packet buffer sharing, path-based load balancing, and telemetry capabilities, it can improve network utilization by 33% and reduce task completion time by 28%. Sami Bujerben, Vice President of Dell Oro Group, commented, "This is not a simple upgrade, but an interesting attempt to directly address AI network bottlenecks and east-west traffic limitations."
The G300 is also noteworthy for its programmable design, which allows for the addition of functionality based on standard changes after deployment. Furthermore, Cisco plans to launch the Nexus 9000 and Cisco 8000 switching platforms based on this chip, offering liquid-cooled options in addition to air cooling. The liquid-cooled model offers 70% better energy efficiency compared to traditional solutions and can handle the bandwidth that previously required six systems in a single system, thus attracting the attention of enterprises deploying at scale.
Cisco has also upgraded its switch driver software, enhancing flexibility by improving support for Linux-based NX-OS and the open-source network operating system Sonic. Cisco also released a family of high-density optical modules for AI clusters, including a next-generation OSFP type 1.6Tbps optical module and a new 800Gbps optical module with 50% lower power consumption.
At the same time, Cisco also released the "Nexus One" management platform, which integrates the entire AI network environment into a single operating system. This platform is characterized by its ability to manage on-premises and cloud networks through a single view, providing API-based automation and visibility related to AI workloads. A particularly noteworthy feature is its native integration with Splunk, acquired in 2024, enabling network telemetry analytics without requiring mobile data.
The AI networking market is currently dominated by companies such as NVIDIA (NVDA), Celestica, Juniper Networks, and Arista Networks. However, Cisco is seeking rapid expansion by focusing on existing enterprise customers initiating AI infrastructure transformations, leveraging its vertically integrated product portfolio, from chipsets to systems, software, optical components, and operational platforms. Vice President Bujerben commented, "Cisco's integrated stack is highly competitive for companies that prioritize stability over new partners."
Cisco is closely monitoring the accelerating adoption of agent-based AI applications. As a large number of autonomous agents emerge, traditional centralized firewalls or authentication systems alone will be insufficient to address security risks. Therefore, Cisco anticipates a transformation across its entire architecture, reflecting this reality by integrating new security management strategies at the network edge.
Most of the products announced this time are still under development and will be launched to the market sequentially upon completion. Cisco stated that, given that network infrastructure itself is emerging as a core competitive advantage in AI environments, the company plans to strengthen its market control through a two-pronged approach of technological differentiation and operational optimization.
