Cisco is making a bold move in AI data center connectivity with its latest Silicon One P200 processor and the Cisco 8223 router, a platform designed to power distributed and large-scale AI clusters. The new system marks a major upgrade in the company’s networking silicon portfolio, targeting hyperscale and enterprise environments that require both performance and efficiency at unprecedented levels.

A leap in AI data center connectivity

At the heart of the new system is Cisco’s Silicon One P200, a 51.2-terabit programmable chipset engineered for routing and switching in demanding AI and cloud environments. The processor features deep-buffer architecture and advanced congestion management, enabling networks to sustain the massive east-west traffic generated by AI training and inference workloads.

The Cisco 8223 routing platform, built around the P200, supports both OSFP and QSFP-DD optical standards—key enablers for connecting geographically distributed AI clusters. Its design reflects the current industry trend toward disaggregated, high-capacity fabrics, capable of extending connectivity across multiple data center locations while maintaining low latency and energy efficiency.

Power, scale, and sustainability

The P200’s architecture delivers the performance that previously required multiple 25.6 Tbps systems, consolidating them into a single 3RU chassis that consumes up to 65% less power. The system provides 64 ports of 800G coherent optics and processes more than 20 billion packets per second—a throughput suitable for AI workloads that rely on massive inter-GPU communication.

Internally, the router supports a 512-radix design and scales up to 13 petabits using a two-tier topology or 3 exabits in a three-tier configuration. This scalability makes it particularly suited for environments where AI training clusters must be distributed across regions while maintaining high data consistency.

Deep buffering plays a crucial role in this generation of networking silicon. While traditional networking designs often view large buffers as a source of latency or jitter, Cisco’s approach focuses on intelligent congestion control. The enhanced buffer management ensures predictable performance even during traffic surges—vital for AI workloads that generate short but intense bursts of data.

Architecture for modern AI traffic patterns

AI workloads are deterministic but data-intensive, often generating traffic that floods the network during synchronization between compute nodes. The 8223’s design directly addresses this challenge through programmable traffic management and flow optimization, allowing network operators to anticipate and balance traffic more effectively.

With high-radix connectivity, the router also reduces the number of hops between compute devices, cutting down latency and simplifying rack layouts. The result is a flatter, more energy-efficient network architecture optimized for low-latency, high-bandwidth communication between GPUs and AI accelerators.

Open networking software support

In a notable shift, Cisco’s 8223 does not initially rely on the company’s proprietary IOS XR operating system. Instead, it launches with support for open-source platforms, including SONiC (Software for Open Networking in the Cloud) and FBOSS, developed by Meta.

These network operating systems are designed around the principle of hardware abstraction, allowing software to run across multiple switch vendors and ASIC types. SONiC, in particular, provides a flexible and modular environment with full support for protocols like BGP, RDMA, QoS, and Ethernet/IP, offering network operators the freedom to integrate Cisco’s hardware into open, multi-vendor ecosystems.

IOS XR support is planned for a later release, signaling Cisco’s continued transition toward interoperable, hybrid infrastructures that combine its proprietary technology with open-source innovations.

Reimagining the AI network backbone

The introduction of the Silicon One P200 and Cisco 8223 router underscores a broader transformation in data center networking. As AI workloads grow more distributed and power-hungry, hyperscalers and service providers are seeking systems that balance throughput, flexibility, and energy efficiency.

Cisco’s latest hardware aims to provide exactly that—bridging traditional networking with AI-native design principles, and setting the stage for the next generation of high-performance, open, and programmable data center fabrics.