Cisco unveils industry's first 51.2T router to power distributed AI networks

The new 8223 system, powered by Silicon One P200 chip, addresses critical data center interconnect challenges as AI workloads outgrow single-facility capacity.

Cisco has launched the 8223 routing system, marking a significant milestone in AI infrastructure with the industry's first 51.2 terabits per second (Tbps) fixed router specifically engineered for connecting data centers running artificial intelligence (AI) workloads.

The announcement, made on October 8, introduces both the 8223 system and the new Silicon One P200 chip that powers it, targeting a critical bottleneck as hyperscalers and enterprises find themselves unable to expand AI infrastructure within individual data centers due to power and space constraints.

Addressing the scale-across challenge

The new routing system directly tackles what Cisco calls the "scale-across" problem – the need to distribute AI workloads across multiple data centers that may be hundreds of miles apart. As organizations exhaust options to scale-up within facilities or scale-out by adding systems, data center interconnects have become the new frontier for AI infrastructure expansion.

"AI computers are outgrowing the capacity of even the largest data center, driving the need for reliable, secure connection of data centers hundreds of miles apart," said Martin Lund, Executive Vice President of Cisco's Common Hardware Group. "With the Cisco 8223, powered by the new Cisco Silicon One P200, we're delivering the massive bandwidth, scale and security needed for distributed data center architectures."

The 8223 is already shipping to initial hyperscaler customers, positioning Cisco to capitalize on the rapidly expanding AI infrastructure market where traditional networking approaches are reaching their limits.

Technical specifications and advantages

The 8223 distinguishes itself as the only fixed routing system offering 64 ports of 800G connectivity, delivering processing capability exceeding 20 billion packets per second and scaling capacity up to 3 Exabits per second.

Housed in a compact 3-rack-unit (RU) form factor, the system achieves what Cisco claims is switch-like power efficiency while maintaining deep-buffer routing capabilities essential for handling the bursty traffic patterns characteristic of AI training workloads.

The P200 chip's deep buffering capabilities enable the system to absorb massive traffic surges without network slowdowns, while its programmability allows organizations to adopt new protocols and standards without hardware replacements – a critical feature as AI networking requirements continue to evolve.

Additional capabilities include support for 800G coherent optics enabling connections up to 1,000 kilometers, line-rate encryption using post-quantum resilient algorithms, and integration with Cisco's observability platforms for comprehensive network monitoring.

Industry validation

Major cloud providers have endorsed the technology. Microsoft's Dave Maltz, Technical Fellow and Corporate Vice President of Azure Networking, noted that as an early Silicon One adopter, the company has found the common ASIC architecture facilitates expansion across data center, wide-area network, and AI/machine learning environments.

On the other hand, Alibaba Cloud is planning to leverage the P200 as a foundational building block for its eCore architecture expansion, with Dennis Cai, Vice President and Head of Network Infrastructure, stating the chip "will enable us to extend into the Core network, replacing traditional chassis-based routers with a cluster of P200-powered devices."

Deployment flexibility

Cisco is initially offering the 8223 with open-source SONiC operating system support, with IOS XR planned for future availability. The P200 silicon will also be deployable in modular platforms and disaggregated chassis, while the Nexus portfolio will support P200-based systems running NX-OS.

This multi-platform approach reflects Cisco's strategy to provide architectural consistency across different network sizes and deployment models, addressing varied customer requirements in the AI infrastructure space.

The announcement reinforces Cisco's position in AI networking infrastructure, building on its Silicon One portfolio introduced in 2019, which has been deployed in major networks globally across hyperscaler, data center, enterprise, and service provider environments.