Cisco, Nvidia Boost Enterprise AI to Rack-Scale

Cisco Systems Inc. and Nvidia are significantly advancing the deployment of enterprise-grade artificial intelligence by extending their integrated solutions to the rack-scale level. This strategic evolution aims to equip businesses with comprehensive, full-stack operational capabilities for sophisticated AI workloads, including advanced reasoning, agentic AI, and the training of trillion-parameter models. The enhanced offering addresses the critical need for robust infrastructure that can support AI's transition from research environments to high-value production settings.

At the core of this upgrade is the integration of liquid cooling technology capable of managing power demands exceeding 200 kilowatts per rack. This capability is crucial for the intensive computational requirements of modern AI development. The partnership also introduces new networking components, notably Nvidia Spectrum-X switch silicon, which is being integrated with Cisco's network operating systems, including NX-OS and SONiC. This synergy allows for high-performance GPU interconnectivity, a fundamental requirement for efficient AI training and inference.

“We are essentially bringing a unified networking stack that connects these powerful GPUs, leveraging Nvidia’s cutting-edge silicon alongside our established operating systems,” explained Will Eatherton, senior vice president and head of networking engineering at Cisco. “Furthermore, we are utilizing Cisco Silicon One, running the same software, to provide flexible options for customers, particularly those focused on high-performance networking and storage integration.” This approach ensures a cohesive and optimized data flow essential for AI factory operations.

The collaboration addresses a key challenge in the AI lifecycle: bridging the gap between acquiring advanced hardware, such as Nvidia's GPUs, and establishing fully operational, production-ready AI factories. Marc Hamilton, vice president of solutions architecture and engineering at Nvidia, highlighted the broader implications: “This is more than just a rack-scale solution; it’s about creating an entire AI factory. In the realm of generative AI, debugging and optimizing across complex, distributed systems can be a significant hurdle. The reference architecture that Cisco is adopting provides the visibility and control needed to manage these intricate environments effectively.”

To facilitate this, Cisco is incorporating its established management tools, including Cisco Nexus One, a unified networking management plane emphasizing embedded security and observability. Additionally, the Cisco Cloud Control operations platform is being leveraged to manage agentic IT operations across the entire solution. This integrated approach ensures that customers receive not only the hardware and networking but also the software, sales, and support necessary for seamless AI deployment and management.

The market for AI infrastructure is experiencing rapid expansion, with enterprises increasingly investing in dedicated solutions to harness the power of artificial intelligence. The global AI market size was valued at approximately USD 150 billion in 2023 and is projected to grow at a compound annual growth rate (CAGR) exceeding 37% from 2024 to 2030, according to various industry reports. Solutions like the Cisco and Nvidia partnership are critical in enabling this growth by providing scalable, efficient, and secure platforms for AI development and deployment.