Cisco has unveiled a new routing system and chip that aim to connect multiple data centers to process the demands of AI traffic.

The Cisco 8223 is described as the industry’s first 51.2-Tb/s fixed form Ethernet router, featuring 64 ports of 800GE, and capacity for over 20 billion packets per second and a scale-up potential of 3 Exabits per second.

Under the 8223’s hood is Cisco’s latest release from the Silicon One family, the P200 chip, which is composed of deep-buffer, fixed-leaf systems. The 51.2 Tb/s routing silicon is available in fixed or centralized forms, designed for deployment in fixed systems, modular platforms, and disaggregated chassis.

P200-based systems such as the 8223 can deliver a total of over 500 terabits per second, which enables full and flexible connectivity between all ports. Cisco said that this setup can replace six 25.6 Tb/s fixed systems or a four-slot modular system, with less power consumption.

Early adopters of Cisco's new wares include Alibaba, who plan to leverage the P200 to build a single-chip platform

"Beyond supporting our Cisco Silicon One Q200 deployment scenarios, this new routing chip will enable us to extend into the Core network, replacing traditional chassis-based routers with a cluster of P200-powered devices," said Dennis Cai, VP and head of network infrastructure at Alibaba Cloud. "This transition will significantly enhance the stability, reliability, and scalability of our DCI network while keeping the simplicity.”

"In addition, we are developing and exploring innovative disaggregated architectures using Cisco G200 for our high-performance data center network ... empowering Alibaba to accelerate innovation and drive infrastructure expansion in the AI era," Cai added.

Other companies exploring the tech include Microsoft and Lumen.

ASIC and AI

The release of its routing ASIC sets out Cisco’s stall for the demands of AI traffic on the network, with recent research by Dell’Oro Group reporting that AI led to a 51% increase in global data center capex in 2024, hitting $455 billion in investment.

Real estate consulting firm JLL’s Data Center Project Development and Services, meanwhile, expects around 40% of total data center workloads to be AI-related by 2030.

This demand has seen Broadcom release Jericho4 in August, also capable of 51.2 Tb/s, designed to interconnect AI hardware across multiple data centers.

Similarly, Nvidia’s Spectrum-XGS Ethernet is designed to connect multiple data centers to form a unified 'AI super-factory.’ That system leverages the Spectrum-X networking fabric, which is powered by Spectrum SN5600 switches and their 51.2 Tb/s capacity specs.

Silicon One, security & OS options

Cisco also underlined its security heritage with its new release, pointing to how security breaches can undermine trust in AI, and the necessity of deep buffers in absorbing dynamic traffic shifts during outages.

The firm also pointed to security credentials and built-in hardware analyzers found in the Silicon One portfolio, which supports media access control security (MACsec), Internet Protocol Security (IPsec), post-quantum cryptography (PQC), and cloud security (CloudSec) to secure data across network links and strengthen protection.

Cisco reported the 8223 will first support SONiC (Software for Open Networking in the Cloud) deployments, an open-source network OS developed by Microsoft and the Open Compute Project community. Compatibility with Cisco’s own proprietary network OS, IOS XR, will come later.

The P200 silicon is also available outside of the 8223 system, deployable in modular platforms and disaggregated chassis, while the Cisco Nexus switch portfolio will soon support P200-based systems running on Cisco’s data center network operating system, NX-OS.