AI’s growth is colliding with electricity limits. Data center operators are finding that powering ever-larger “cluster farms” in single locations is becoming unfeasible. In a move that may help alleviate this barrier, Cisco just introduced the Silicon One P200 chip and 8223 router, which are designed to enable AI workloads to be distributed across multiple data centers.
Cisco says the 8223, powered by P200, delivers 51.2 Tbps throughput while cutting power consumption by about 65%compared to its previous generation systems. The system supports 64 ports of 800G coherent optics and is capable of connecting data centers separated by distances up to 1,000 km.
Cisco says the 8223, powered by P200, delivers 51.2 Tbps throughput while cutting power consumption by about 65%compared to its previous generation systems. The system supports 64 ports of 800G coherent optics and is capable of connecting data centers separated by distances up to 1,000 km.
Martin Lund, Executive Vice President of Cisco’s Common Hardware Group, said, “AI compute is outgrowing the capacity of even the largest data center, driving the need for reliable, secure connection of data centers hundreds of miles apart.” Lund added that AI firms are placing data centers “wherever you can get power.”
Cisco also emphasizes “deep buffering” as a key feature. Rakesh Chopra, SVP & Fellow, Silicon One, commented that “deep buffers slow down AI workloads … if you dig into the details, it’s actually not true” when paired with proper congestion management. He said the system is designed so that “everything has shrunk way down, we can drive our fan power down; we can save on power conversion … we’re chasing out every single watt in that system … because that is the fundamental limitation in the industry.”
In parallel, broader reports quantify how sharply data center power demand is rising. Goldman Sachs Research forecasts that global data center power demand will increase 165% by 2030 (relative to 2023). The International Energy Agency reports that data center electricity consumption is expected to double, reaching 945 TWh by 2030, roughly the current consumption of Japan. McKinsey has observed that average power density per rack in AI-ready data centers has more than doubled in two years, from about 8 kW per rack to around 17 kW, and projects densities up toward 30 kW by 2027.
Eric Schmidt, former CEO of Google, warned in 2025: “We need energy in all forms… and we need it quickly.” He estimated that by 2030, U.S. data centers could require tens of gigawatts of additional power.
Gartner, one of the leading analyst firms, predicts that 40% of AI data centers will be operationally constrained by power availability by 2027.
Cisco’s announcement frames its new equipment as a tool for data center operators to address those constraints. The “scale-across” term appears repeatedly in Cisco’s descriptions: Cisco argues that AI workloads must be distributed across multiple geographic sites because increasing scale within a single site (“scale-up”) or across racks in one facility (“scale-out”) alone cannot meet rising demands for power, space, and cooling.
Alibaba, as an early customer, stated: “We are pleased to see the launch of Cisco Silicon One P200 … a routing ASIC that delivers high bandwidth, lower power consumption, and full P4 programmability. … The introduction of this advanced routing chip marks a pivotal step forward, empowering Alibaba to accelerate innovation and drive infrastructure expansion in the AI era.” said Dennis Cai, Vice President, Head of Network Infrastructure, Alibaba Cloud.
Microsoft’s Corporate VP of Azure Networking, Dave Maltz, said: “The increasing scale of the cloud and AI requires faster networks with more buffering to absorb bursts of data.”
These data points and statements establish a chain: rising power demand → grid limitations → need for distributed AI clusters → need for higher-capacity, efficient interconnects. Cisco’s P200/8223 system is clearly positioned to serve that need.
The significance lies in matching infrastructure design to the scale and geometry of the power problem. If power is the limiting resource, then distributing compute to where power is abundant, with hardware that minimizes overhead and maximizes throughput, becomes central.
The questions for data center operators, cloud providers, and AI labs are straightforward. Can they deploy such distributed clusters with the operational discipline needed: balancing latency, synchronization, and traffic bursts? Will utilities, regulators, and regional governments invest in the transmission, generation, and cooling infrastructure required? Can security, cost, and reliability be maintained across geographically dispersed sites?
Cisco’s P200 potentially addresses the network‐side of these questions. Its ability to link distant data centers with high bandwidth, lower energy cost, and programmable features is a material advance in AI infrastructure design. This is especially timely, as power limits are forcing the data-centre industry to rethink how and where it builds.
