Cisco Systems scales programmable AI network capacity with new silicon and repluggable optics innovations

Cisco Systems recently unveiled a suite of open silicon and optical networking systems at the annual Open Compute Project Global Summit, designed to meet the increasing demands of AI workloads. These innovations aim to tackle the bottleneck in AI data centers, where data center operators need to move significantly more data from chip-to-chip without increasing hardware costs or power consumption.

The company built a "scale-across" architecture to link distributed AI data centers into a massive, unified compute fabric, focusing on coherent optics and deep-buffered systems. Cisco's new Silicon One chip family, including the flagship 102.4-terabit Silicon One G300 chip, and the 51.2-terabit top-of-rack networking switch, Silicon One G302, are key components of this vision.

These chips support faster data transmission between GPUs and enable liquid cooling for up to 1.6-terabit optics. Cisco also collaborated with Nvidia Corp. on a repluggable optics module, the USFP, which significantly increases bandwidth density and power efficiency. A new out-of-band network management system, Cisco 8001, powered by Silicon One A100 chips, has been introduced to streamline network recovery operations during outages.

Cisco's SONiC network operating system has also been updated with new features, including a Live Protect runtime security feature. Kevin Wollenweber, Cisco's senior vice president and general manager of data center and internet infrastructure, emphasized the need for a new approach to scaling data center networks in the era of AI, highlighting Cisco's role in delivering the open infrastructure required for these evolving needs.