Cisco Systems has launched its new Silicon One G300 switch chip, positioning it as a direct competitor to high-bandwidth offerings from Nvidia and Broadcom in the AI data center networking race. The announcement, made on February 10, 2026, led to a modest rise in Cisco's stock (NASDAQ: CSCO) as investors reacted to the strategic move.
The G300 is a 102.4 terabits per second (Tbps) switching silicon engineered specifically for large-scale GPU deployments in AI data centers. It utilizes TSMC's advanced 3nm manufacturing process and features 512 SerDes (Serializer/Deserializer) that process data at 200 Gigabits per second. This design allows the chip to support deployments of up to 128,000 GPUs using only 750 switches, a significant improvement over the 2,500 switches previously required for similar output.
Cisco claims the G300 offers substantial performance and efficiency gains. Its Intelligent Collective Networking feature is said to increase network utilization by 33% and reduce AI job completion time by up to 28% compared to non-optimized methods. The chip also integrates advanced traffic management that can automatically reroute data during spikes, preventing packet drops that can stall AI training jobs.
On the efficiency front, the G300, when paired with Cisco's N9000 and 8000 series switches and 800 Gbps Linear Pluggable Optics (LPO), is reported to cut power consumption by approximately 30%, with the pluggable modules themselves using 50% less energy than previous generations. The chip also features the latest liquid-cooling systems.
Company executives highlighted the strategic importance of the release. Jeetu Patel, Cisco's President and Chief Product Officer, stated the company is "spearheading performance, manageability, and security in AI networking by innovating across the full stack – from silicon to systems and software." Martin Lund, Executive Vice President of Cisco’s Common Hardware Group, added that the G300 provides "high-performance, programmable, and deterministic networking" to help clients fully utilize their compute resources.
The launch is part of Cisco's broader push to strengthen its position among hyperscale data center operators. The company reported $1.3 billion in AI infrastructure commitments from hyperscaler clients in its recent fiscal Q1 2026. The G300 also aligns with industry efforts, like the Ultra Ethernet Consortium (UEC), to make Ethernet a high-performance, open alternative to proprietary solutions like InfiniBand for AI clusters.
Alongside the hardware, Cisco announced the release of AgenticOps for data center networking through its AI Canvas platform. This software aims to simplify troubleshooting through guided, human-in-the-loop conversations.
Sales of the Silicon One G300 chip are slated to begin in the second half of 2026.
