Cisco is expanding its Secure AI Factory with NVIDIA to support rack-scale systems based on NVIDIA’s Vera Rubin platform, while connecting those systems to Cisco’s scale-out and scale-across networking. The approach combines Cisco Silicon One on the front end, NVIDIA Spectrum-X on the back end, and a common Nexus One operating model under an NVIDIA Cloud Partner reference architecture. Cisco also announced general availability of Cisco Cloud Control and expanded distributed security through Hypershield support on smart switches.
Taken together, these announcements reflect an important shift in the AI infrastructure market. The discussion is moving beyond individual GPUs and faster switches toward validated systems that can be deployed, secured, and operated as a unified environment. For enterprises, neoclouds, and sovereign cloud providers, the potential value lies less in any single component than in reducing the integration burden surrounding increasingly distributed AI workloads.
I had the opportunity to sit down with Murali Gandluru at Cisco GSX to discuss these innovations. The link to the full interview is below.
From Large Training Clusters to Distributed AI
The first wave of generative AI infrastructure investment centered heavily on large, centralized training clusters. That demand remains significant, but the next phase will be more distributed. Enterprises are moving AI into production through inference services, agentic applications, and hybrid workflows that connect models with corporate data, traditional applications, cloud services, users, and other agents.
That changes the network requirement. AI traffic is no longer confined to a specialized back-end fabric. It increasingly moves east-west within data centers and north-south between users, applications, clouds, edge locations, and data sources. Traditional and AI workloads will coexist, making separate operational silos difficult to sustain.
Murali Gandluru, senior vice president of data center networking at Cisco, described the implications clearly: “Traditional workloads [and] AI workloads are not separate silos. They’re actually part of a hybrid infrastructure.”
This helps explain Cisco’s 35% year-over-year data center networking growth in its fiscal fourth quarter. AI is generating infrastructure demand while prompting organizations to modernize the broader data center. Higher-speed connectivity matters, yet enterprises must also consider consistent policy, visibility, resilience, and operational simplicity across mixed environments.
The Business Value of a Validated Architecture
Building an AI environment from individual components burdens architecture and operations teams. GPUs, networks, storage connectivity, security, and management systems must work together predictably. Integration delays can slow time to value, while inconsistent tools increase operational risk.
Cisco’s Secure AI Factory with NVIDIA is intended to address that challenge through a tested reference design spanning the rack-scale system and the wider network. Gandluru explained that customers gain “the comfort that Cisco is going to stand behind an NVIDIA Cloud Partner reference architecture design with Cisco’s Silicon One for the frontend, NVIDIA Spectrum-X architecture for the backend, all operated by the same common operating model of Nexus One.”
For buyers, that could mean faster deployment, clearer support ownership, and less interoperability testing. A common operating model is particularly relevant when deploying AI across data centers, colocation facilities, sovereign environments, and edge locations.
However, enterprises should still validate the architecture against their own workload profiles. Important considerations include application latency, storage traffic, GPU utilization, power and cooling availability, workload portability, operational skill sets, and the degree to which a reference architecture preserves future technology choice.
Making Security Part of the Fabric
AI infrastructure also expands the security challenge. Agentic systems will generate more machine-to-machine communication, access data across multiple locations, and interact with both AI and traditional applications. Applying security through centralized appliances alone can introduce bottlenecks and force traffic onto inefficient paths.
Cisco’s strategy is to distribute enforcement throughout the fabric while retaining centralized policy. Hypershield support on smart switches is designed to make network ports enforcement points, extending policy closer to workloads—including Kubernetes environments—and reducing the need to redirect traffic through a centralized firewall.
As Gandluru put it, “We want to make your entire network fabric the enforcement points. So every port in your infrastructure becomes an enforcement point.”
The potential outcome is stronger control over lateral movement and more efficient traffic flows. This reflects a wider industry trend toward embedding security into infrastructure. Enterprises should still assess policy consistency, security integration, performance, and management complexity.
A Common Operational Layer for AgenticOps
The other major element is operations. Cisco Cloud Control, now generally available, is positioned as a common access point across Cisco infrastructure. Nexus One provides the data center architecture and operating model, while AI Canvas gives network, application, and security teams a shared workspace for troubleshooting. Cisco is also connecting this environment to AI assistants, its Deep Network Model, and Splunk-derived insights.
The near-term business case is reducing time spent switching tools, determining ownership, and handing incidents between domains. Shared context could help teams identify whether a problem sits in the network, application, Kubernetes environment, or security stack.
The longer-term opportunity is AgenticOps, using AI agents to interpret telemetry, collaborate, recommend action, and eventually execute approved changes. Enterprises should distinguish today’s assisted troubleshooting from autonomous capabilities that will mature over time. Governance, auditability, deterministic execution, and human oversight remain essential.
Why It Matters
Cisco’s latest announcements demonstrate that AI infrastructure competition is broadening. Success will not be determined solely by who delivers the fastest switch or largest GPU cluster. Vendors will increasingly compete on their ability to deliver validated architectures, unified operations, embedded security, ecosystem integration, and consistent support across distributed environments.
Cisco brings considerable assets to that contest, including silicon, systems, optics, networking software, security, observability through Splunk, and a broad partner ecosystem. Its challenge will be turning that portfolio breadth into a genuinely simpler customer experience and producing measurable proof points around deployment time, operational efficiency, resilience, and cost.
For enterprise leaders, the key takeaway is that AI readiness should not be treated as a standalone infrastructure project. AI will reshape traffic patterns and operating requirements across the existing application estate. Organizations should therefore evaluate AI infrastructure as part of a broader modernization strategy, one that connects architecture, security, observability, and operations from the rack to the cloud and edge.
For more information on Cisco’s Secure AI Data Center, please visit their website.
