Cisco unveils AI chips and partners with Nvidia

At the Open Compute Project Global Summit on October 12, 2026, Cisco Systems introduced new silicon, systems, and optical technologies tailored for open artificial intelligence infrastructure. The updates span rack-level compute fabrics as well as distributed capacity across facilities.

Cisco announced that its 102.4T Silicon One G300 is now sampling, with the first G300-powered systems currently shipping to customers. The company also expanded the portfolio with the Silicon One G302, a 51.2T switching chip with 200G SerDes designed for top-of-rack deployments. The G302 integrates 1.6T optics and direct-to-chip liquid cooling, and Cisco plans to begin customer shipments before the end of 2026.

Cisco is also collaborating with Nvidia to design an open specification for Ultra-Dense Small Form Factor Pluggable optical modules. Capable of 400Gb/s per lane, the USFP modules are expected to offer four times the bandwidth density and 1.3 times the power efficiency of 200Gb/s-per-lane optics.

In production networks, Cisco confirmed that leading hyperscalers have deployed its 51.2T systems powered by the Silicon One P200. Microsoft corporate vice president of engineering Anand Ramakrishna stated that Microsoft has entered production on P200-powered systems. Alibaba Cloud Intelligence Group vice president and head of network infrastructure Dennis Cai announced the massive-scale deployment of its in-house Quantum Cat router using the P200 to provide 51.2 Tbps of routing capacity.

For data center management, Cisco launched the Cisco 8001 management system powered by the Silicon One A100 chip, available in 10G or 1G configurations running the open-source SONiC operating system. In addition, Cisco plans to introduce Live Protect for SONiC users starting in mid-2027 to mitigate vulnerability exposure before patches are applied.