Cisco has launched the Silicon One P200 chip and 8223 routing system, directly challenging Broadcom's dominance in connecting AI data centres. These new offerings aim to facilitate faster data transfers across geographically dispersed data centres, essential for training large AI models. Microsoft and Alibaba are currently evaluating Cisco's new platform.
The P200 chip boasts a 51.2 terabits per second routing performance, consolidating the functionality of 92 separate chips into one and reducing power consumption by 65%. The 8223 routing system is designed to maximise rack efficiency and support 400/800 Gbps interfaces over long distances. Its deep-buffer architecture prevents packet loss, crucial for maintaining GPU utilisation in distributed AI workloads.
Cisco's entry into the market aims to provide enterprises with more negotiating power and accelerate innovation in AI infrastructure. The company is also embedding AI-ready security innovations into its Hybrid Mesh Firewall and Universal Zero Trust Network Access offerings. The P200 chip is now available in the 8223 system, with deployment in modular platforms and disaggregated chassis to follow.
