Cisco Enters the AI Networking Arms Race
Cisco has joined the intensifying competition to power the world’s expanding AI infrastructure with the launch of its 8223 routing system — a high-capacity, energy-efficient router designed specifically for connecting AI data centres spread across multiple locations.
Unveiled on October 8, the system features the Silicon One P200 chip, a core component capable of 51.2 terabits per second throughput. It marks Cisco’s bid to solve a mounting problem for the AI industry: how to keep scaling when traditional data centres have run out of space, power, and cooling capacity.
“AI compute is outgrowing the capacity of even the largest data centre,” said Martin Lund, Executive Vice President of Cisco’s Common Hardware Group. “That’s driving the need for reliable, secure connection of data centres hundreds of miles apart.”
D.K.
Why AI Needs “Scale-Across” Networks
Until now, the industry has relied on two main strategies to expand capacity — scaling up (making servers more powerful) and scaling out (adding more servers in the same facility). But the energy and cooling demands of AI workloads have pushed those limits.
The new frontier is “scale-across” networking — linking multiple facilities so they function as one massive AI supercomputer. This approach helps balance workloads across regions but introduces a new challenge: the network itself becomes the bottleneck.
Traditional routers weren’t built for the massive, unpredictable data surges that AI training generates. When those links get congested, GPU clusters sit idle, wasting both energy and time.
Inside Cisco’s 8223 System
Cisco’s 8223 system tackles these constraints head-on. Housed in a compact 3RU chassis, it provides 64 ports of 800-gigabit connectivity — the highest density currently available in a fixed router. The system can handle over 20 billion packets per second and scale to three exabytes per second of total interconnect bandwidth.
Its defining feature is deep buffering, made possible by the P200 chip. These large buffers act like a shock absorber for data, soaking up sudden spikes in traffic that would otherwise choke the network. The result: smoother performance during AI training bursts.
Power efficiency is another standout. The 8223 offers what Cisco describes as switch-like energy efficiency — a critical factor as data centres face mounting power constraints. The system also supports 800G coherent optics, allowing connections over distances of up to 1,000 kilometres, essential for linking distributed AI clusters.
Early Adopters and Industry Reactions
Cisco’s long-time partners are already testing the waters. Microsoft, which has used Silicon One chips in Azure’s network, said the architecture “made it easier for us to expand from our initial use cases to multiple roles in DC, WAN, and AI/ML environments,” according to Dave Maltz, Azure’s technical fellow and corporate vice president.
Alibaba Cloud plans to use the P200 to extend its eCore network architecture, replacing traditional routers with clusters of P200-powered devices. Meanwhile, Lumen is exploring how the technology can boost its network performance and customer offerings.
Future-Proofing and Security
The 8223’s programmability gives it a future-ready edge. Instead of replacing hardware to support new protocols, operators can update the chip directly — a major cost advantage as AI networking standards evolve.
Security has also been built into the system. The router features line-rate encryption using post-quantum-resilient algorithms, protecting data from potential future quantum computing threats. It integrates with Cisco’s observability tools for end-to-end network visibility and rapid troubleshooting.
Cisco vs. Nvidia vs. Broadcom
Cisco’s entry intensifies a three-way race with Broadcom and Nvidia for dominance in the emerging scale-across networking market. Broadcom’s Jericho 4 and Nvidia’s Spectrum-XGS platforms have already staked claims, but Cisco brings its decades-long enterprise presence, the established Silicon One portfolio, and deep relationships with hyperscalers.
The 8223 will first ship with open-source SONiC software support, followed by Cisco IOS XR, and will later appear in modular systems and Nexus switches — giving customers flexible deployment options and reducing vendor lock-in risks.
The Bigger Picture
Whether Cisco’s 8223 becomes the industry benchmark remains to be seen, but one thing is clear: the next phase of AI growth depends as much on networking innovation as it does on compute power. As AI models expand beyond the limits of single facilities, the companies that can move data faster, smarter, and more efficiently will define the future of AI infrastructure.
