Cisco Systems has launched its new Silicon One P200 networking chip alongside the 8223 routing system, two innovations designed to handle the enormous demands of artificial intelligence (AI) workloads.
The new systems are already being adopted by hyperscale customers including Microsoft Azure and Alibaba Cloud.
The launch will enhance data centre connections and scale. With the growth of AI training models that span multiple regions, Cisco’s P200 and 8223 systems aim to efficiently link data centres located thousands of miles apart, acting as a unified, high-performance computing network.
Martin Lund, executive vice president of Cisco’s Common Hardware Group, explained the scale of change the company is addressing: “AI compute is outgrowing the capacity of even the largest data centre, driving the need for reliable, secure connection of data centres hundreds of miles apart. 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 centre architectures.”
The P200 chip, a deep-buffer routing silicon, replaces 92 separate chips with one. It delivers more than 65% energy savings compared to traditional routers, an important development as hyperscalers face power challenges.
The chip also supports interconnect bandwidth of over 3 exabits per second, allowing AI workloads to run across multiple facilities without performance loss.
The 8223 routing system is built to complement this. It can process more than 20 billion packets per second, supporting 64 ports of 800G, the highest density of any fixed router available. Its deep buffering feature absorbs data surges generated during AI training, keeping systems stable and responsive.
The system’s design prioritises power efficiency and flexibility. It is the most power-efficient 3RU router in its category, offering switch-like energy performance while managing the intense traffic of AI-driven networks. The 8223 can adapt to changing network protocols through programmability, reducing the need for hardware upgrades and keeping operations agile.
Security is another focal point, as the 8223 integrates line-rate encryption and post-quantum resilient algorithms, backed by continuous monitoring features that provide visibility into performance and threats. This ensures that sensitive AI traffic remains protected as data moves between facilities.
Dave Maltz, corporate vice president of Azure Networking at Microsoft, said: “The increasing scale of the cloud and AI requires faster networks with more buffering to absorb bursts. We’re pleased to see the P200 providing innovation and more options in this space.”
Alibaba Cloud is also deploying the technology to enhance its eCore architecture. “We are pleased to see the launch of Cisco Silicon One P200, the industry’s first 51.2T routing ASIC that delivers high bandwidth, lower power consumption, and full P4 programmability. This breakthrough chip aligns perfectly with the evolution of Alibaba’s eCore architecture,” said Dennis Cai, vice president and head of Network Infrastructure at Alibaba Cloud.
The 8223 is initially being released for open-source SONiC deployments, with IOS XR support to follow. Cisco also confirmed that the P200 silicon will be available for modular and disaggregated platforms, allowing consistent integration across different network sizes.
With AI workloads increasing across industries, Cisco’s latest launch will boost its competitive edge with companies like Broadcom in the networking chip market. More importantly, it stresses how infrastructure innovation, not just software, will determine how effectively global AI systems evolve.
Patrick Moorhead, CEO and chief analyst at Moor Insights & Strategy, said, “As AI workloads rapidly outpace the capabilities of traditional data centres, the industry faces new challenges in bandwidth, reliability, and scale. Cisco’s 8223, powered by Silicon One P200, marks a significant step forward, delivering the industry’s first 51.2-terabit fixed Ethernet router purpose-built for secure, power-efficient scale-across networking.”
