Cisco Showcases 102.4-Tbps Silicon for AI Networks at OCP Summit

Cisco is showcasing networking silicon, optical links and liquid-cooled systems for AI infrastructure at the 2026 Open Compute Project Global Summit in San Jose, highlighting technologies aimed at moving data through large AI clusters.

The event runs Oct. 12–15. Cisco held its executive session, “Breaking the Walls: Open Scale-Across for the AI Era,” on Oct. 12 from 12:50 to 1:10 p.m. ET (16:50 to 17:10 UTC).

Cisco’s Silicon One G300 delivers 102.4 Tbps of switching capacity for large AI clusters. G300-powered N9000 and 8000 systems are expected to support 102.4 Tbps switching and ship in 2026.

Cisco’s Silicon One P200 provides 51.2 Tbps of switching capacity. The portfolio is intended to support networking requirements across AI infrastructure, including high-bandwidth connections within clusters and data-center interconnection.

Cisco also announced 21.6T OSFP optics for links operating at 1.6T, 800G, 400G and 200G. Its 800G linear pluggable optics can reduce optical-module power consumption by 50% compared with retimed modules and reduce total switch power by 30%, Cisco said.

Cisco said its 100% liquid-cooled systems could improve energy efficiency by nearly 70% and provide the bandwidth of six prior-generation systems in one system.

“We are spearheading performance, manageability, and security in AI networking by innovating across the full stack - from silicon to systems and software,” Jeetu Patel, Cisco’s president and chief product officer, said.

The products target both scale-out networking within AI clusters and scale-across data-center interconnection. AI clusters use networks to move data among GPUs and other accelerators, making bandwidth, latency, power consumption and recovery from failures important constraints.

“It’s not just about faster GPUs – the network must deliver scalable bandwidth and reliable, congestion-free data movement,” Martin Lund, Cisco’s executive vice president of the Common Hardware Group, said.