Cisco Expands AI Networking Lineup With G302 Chip and USFP Optics

Cisco is expanding its AI networking portfolio with higher-capacity switching silicon, a planned optical form factor and security tools designed to keep open networks operating during maintenance.

The announcements came at the Open Compute Project Global Summit in San Jose, where Cisco presented networking as a core part of AI infrastructure. Cisco showcased its 102.4-Tbps Silicon One G300 and introduced the 51.2-Tbps G302, which uses 200G SerDes and is designed for top-of-rack systems. G300-powered systems are shipping to customers, while G302-powered systems are planned to begin shipping before the end of 2026.

Cisco and NVIDIA also developed USFP, a planned optical form factor designed for 400 Gb/s per lane. The design targets up to 12.8 Tbps of aggregate capacity through a future 32-lane configuration. Initial transceivers are planned to use 16 full-duplex lanes in an OSFP-sized footprint, with a path to 32 lanes.

USFP is designed to provide 4x the bandwidth density and approximately 1.3x the power efficiency of pluggable optics using 200G-per-lane signaling. TE Connectivity and Amphenol are working with Cisco and NVIDIA on a multisource hardware ecosystem, while an open industry specification is planned through a community-led project. USFP remains a planned design and is not currently shipping.

“AI builders need an interconnect architecture that can scale performance without locking them into a more complex operating model,” Bill Gartner, Cisco’s senior vice president and general manager of optical systems and optics, said.

Cisco is also extending its open-networking strategy to security. SONiC, or Software for Open Networking in the Cloud, is an open-source network operating system used in hyperscaler, neocloud, service-provider and enterprise environments. Customers can build their own SONiC distribution or use a Cisco-hardened version with validated releases, lifecycle management and technical support.

Live Protect support for Cisco-validated SONiC on Nexus One is planned for the second quarter of 2027. The technology provides vulnerability shielding on a live system while patches are applied during planned maintenance, without downtime or a switch reboot. Availability for SONiC customers is planned to begin in mid-2027.

“As AI training and inference continues to scale, data movement is the key to efficient AI compute; the network becomes part of the compute itself,” Martin Lund, Cisco’s executive vice president of the Common Hardware Group, said.

Alibaba Cloud has deployed in-house Quantum Cat routers powered by Silicon One P200 at scale. The systems provide 51.2 Tbps of routing capacity and use SONiC.

The announcements extend Cisco’s earlier OCP coverage of its G300 platform and its role in connecting accelerators across large AI clusters. The 2026 OCP Global Summit is scheduled for Oct. 12–15 in San Jose, California.