Cisco Expands Secure AI Factory Infrastructure

Cisco Systems Inc. is enhancing its rack-scale infrastructure designed for secure AI operations, targeting the burgeoning demand from neocloud and sovereign cloud environments. This strategic move addresses the critical need for robust compute power to handle massive AI training and inference workloads, a key challenge in the current data center buildout phase.

The expanded offerings, developed in collaboration with Super Micro Computer Inc., integrate solutions compliant with Nvidia's Cloud Partner ecosystem. These systems leverage Cisco's own Silicon One switches for front-end connectivity and build upon their Nvidia Spectrum-X-based switches for the back end, all managed through Cisco's Nexus One networking platform. This unified approach aims to simplify the deployment and management of complex AI factories.

Jeetu Patel, President and Chief Product Officer at Cisco, emphasized the unprecedented scale of current data center expansion, stating, "Every organization is racing to scale AI — but speed only counts if it comes with control of data, managed token costs and real return on investment." This highlights the dual focus on performance and operational efficiency in Cisco's strategy.

Further bolstering enterprise adoption, Cisco has updated its Enterprise Reference Architectures. These updated blueprints facilitate rapid prototyping and deployment of networking and rack-scale solutions for AI workloads, particularly those utilizing Nvidia's AI platforms. This initiative aims to reduce risk and leverage established expertise for both small-scale projects and large data center initiatives.

Addressing the significant thermal challenges posed by high-density AI hardware, Cisco's liquid-cooled N9000 Series Switches are designed to interoperate seamlessly with Supermicro's liquid-cooled compute solutions. This integration is crucial for managing the substantial heat generated by systems like Nvidia's NVL72, which can exceed 200 kilowatts per rack, ensuring optimal performance and reliability.

The partnership with Supermicro provides enterprises with pre-validated compute and networking infrastructure, ready for immediate deployment in service-oriented environments. Cisco is also introducing Cisco Validated Services to assist customers in certifying their infrastructure against these reference architectures, ensuring adherence to best practices in design, build, and alignment.

Mark Hamilton, Nvidia's vice president of solution architecture and engineering, noted the significance of integrating AI factories within existing enterprise infrastructure. He pointed out that Cisco's networking capabilities provide access to proprietary enterprise data, a crucial differentiator for training AI models beyond publicly available datasets. This connectivity is vital for supporting the vast number of GPUs required for agentic AI and high-throughput inference.

The Supermicro compute solutions will begin rolling out as part of the Cisco Secure AI Factory with Nvidia starting in October. This collaboration underscores the growing trend of specialized infrastructure solutions tailored to the demanding requirements of modern artificial intelligence workloads.