Cisco and Nvidia Expand Secure AI Factory to Rack Scale, Exceeding 200 Kilowatts per Rack

en.Wedoany.com Reported - Cisco and Nvidia have expanded their collaborative Secure AI Factory with Nvidia to rack-scale deployments, delivering a more complete full-stack operational capability for enterprises. Built on a framework for AI deployment across the customer's entire infrastructure, the solution now integrates liquid cooling from rack to network, supporting loads exceeding 200 kilowatts per rack for use cases such as inference, agentic AI, and trillion-parameter training. The latest enhancements include Nvidia Spectrum-X switching paired with Cisco operating systems, enabling customers to deploy based on reference architectures aligned with Nvidia Cloud Partner specifications as well as architectures built on Cisco Silicon One.

As AI moves from research laboratories to factory floors and retail aisles, enterprises are seeking full-stack architectures capable of deploying autonomous technologies across high-value production environments—a core characteristic of AI factories. Cisco and Nvidia are positioning their portfolios to provide the operational foundation that drives this model.

Will Eatherton, Senior Vice President of Network Engineering at Cisco, explained that in the scale-out segment connecting GPUs, Cisco offers a network stack using Nvidia Spectrum chips running NX-OS or SONiC for front-end GPU connectivity, while also leveraging Cisco Silicon One with the same software ecosystem to enable high-performance storage networking for customers.

Eatherton spoke during an exclusive broadcast on theCUBE, SiliconANGLE Media's live studio, with host John Furrier at the "Cisco Secure AI Factory With Nvidia Expands to Rack Scale" event. He was joined by Marc Hamilton, Vice President of Solutions Architecture and Engineering at Nvidia. The two discussed how they are bridging the deployment gap from acquiring GPUs to operating production-ready AI factories. Hamilton noted that the solution extends far beyond rack scale, representing a complete AI factory-level offering. In generative AI scenarios, users may invoke agents in sandboxes or perform tool calls on the front-end network, which can impact performance, and Cisco's reference architecture provides full visibility into where to enter, inspect, debug, and optimize the entire stack.

The reference architecture integrates multiple tools and services previously developed by Cisco, including the Cisco Nexus One open unified network management plane (integrating chips, systems, software, and operational models with built-in security and observability), as well as the Cisco Cloud Control operations platform for agentic IT. Eatherton stated that Cisco uses Nexus One for day-to-day operations, management, and monitoring, while leveraging Cloud Control to introduce AgenticOps across the entire solution, providing software, sales, and support for the whole cluster. The partnership with Nvidia is critical because the two companies jointly assemble the right technology components for customers, forming a holistic solution that runs parallel to the stack.