Cisco and Nvidia expand the Secure AI Factory from the data center to the industrial edge

The Secure AI Factory now encompasses the entire infrastructure continuum, from data centers to industrial sites, hospitals, and telecom networks. This announcement reinforces three key areas: delivering inference to the edge, enhancing switches for large-scale AI factories, and integrating security at every layer of the stack, from silicon to autonomous agents.

<p>The underlying premise of this announcement is clear: AI inference occurs where data resides, and decisions often cannot wait. For instance, a hospital analyzing a patient’s vital signs in real time, a production line detecting security anomalies through image analysis, or a telecom operator offering managed AI services to industrial clients all exemplify use cases that demand local processing capabilities. These situations eliminate the need for latency-inducing back-and-forth trips to a centralized data center. The expansion of the Secure AI Factory addresses this need by providing a validated architecture that can reduce deployment times from several months to a few weeks, embedding security into the design rather than adding it later.</p>

<p>Chuck Robbins, CEO of Cisco, emphasized the importance of this venture: “Most organizations recognize the transformative potential of AI for their operations, but they are still figuring out how to deploy this technology safely and at scale. In partnership with Nvidia, we are meeting this challenge through an architecture that sets a new performance standard, simplifying the deployment, operation, and security of AI infrastructures.”</p>

<h4><strong>Edge Inference: Nvidia GPUs on Cisco UCS and a New Cisco AI Grid for Operators</strong></h4>

<p>The extension towards the edge is focused on two distinct areas. For enterprises, the support for Nvidia GPUs on Cisco UCS and Cisco Unified Edge portfolios allows critical AI workloads to run locally, sidestepping the energy costs and hardware footprint of a full data center infrastructure. This continuity of architecture offers IT teams a seamless integration between their core environments and edge sites, removing the need to piece together disparate systems.</p>

<p>For telecom operators, Cisco is launching the Cisco AI Grid in collaboration with Nvidia — a reference architecture that combines Cisco's Mobility Services Platform with Nvidia's Blackwell-series RTX PRO GPUs. This setup enables operators to leverage their existing networks to offer managed AI services at the edge, ensuring operator-level reliability and sovereignty. For a CIO working with partner operators, this approach opens the prospect of outsourcing edge inference without transferring data to centralized infrastructures.</p>

<h4><strong>Switching at 102.4 Tbit/s and Integration of Nexus Hyperfabric</strong></h4>

<p>At the level of large-scale AI factories, Cisco is enhancing its switching line with two new references. The Cisco N9100 switch, operating at 102.4 Tbit/s and based on Nvidia's Spectrum-6 Ethernet silicon, represents a high-performance option for the most demanding training and inference architectures. The N9100 800G, already available, is built on Nvidia's Spectrum-4. Together, these models complement the earlier-launched switches based on Cisco Silicon One G300 and P200, designed for scale-out and scale-across architectures.</p>

<p>With the integration of Cisco Nexus Hyperfabric into Cisco Nexus One, support has now expanded to include Cisco N9000 switches, including the N9100 series based on Spectrum-X. This integration transforms complex multivendor orchestration into a unified full-stack solution. For network teams in large organizations, this operational gain is significant: a single control point to manage heterogeneous switching infrastructures, greatly reducing the integration and maintenance burden.</p>

<h4><strong>Integrated Security from Silicon to Agents</strong></h4>

<p>The security aspect is the most foundational element of this announcement for Chief Information Security Officers (CISOs). Cisco is implementing a layered protection strategy that addresses three levels. At the hardware level, Cisco's Hybrid Mesh Firewall extends its security policy application to Nvidia BlueField DPUs integrated into GPU servers connected to Cisco Nexus One factories. Threats are intercepted at the server level before reaching the organization's data, all without compromising the performance of inference workloads.</p>

<p>At the agent level, Cisco AI Defense integrates with Nvidia NeMo Guardrails, part of the Nvidia AI Enterprise suite, to secure distributed AI environments. This coverage includes interactions between agents, which is critical in multi-tier agent architectures where edge agents delegate tasks to central agents, introducing risks that traditional perimeter security tools are not designed to handle. Cisco AI Defense will also support Nvidia’s OpenShell runtimes, part of the Nvidia Agent Toolkit, adding controls and safeguards to monitor agents’ actions and tools. Each tool call and every action taken by an agent in this runtime will undergo continuous monitoring and validation.</p>

<p>Jensen Huang, founder and CEO of Nvidia, articulated the essential philosophy: “AI factories are transforming every industry, and security must be built into every layer — from silicon to software — to protect data, applications, and infrastructure. Together, Nvidia and Cisco are constructing the secure foundations of AI infrastructure so that businesses can confidently deploy intelligence at scale.”</p>

<p>For CIOs managing AI industrialization programs, the Cisco-Nvidia architecture addresses one of the most common tensions: balancing the pressure for rapid deployment with the governance and security requirements imposed by business leaders and regulators. The promise of a validated stack covering networking, computing, perimeter security, and agent governance reduces integration risk while maintaining architectural flexibility. Two coexisting reference architectures are available: one compliant with the Nvidia Cloud Partner program and the other based on Cisco’s Silicon One, allowing technical teams to choose based on their constraints and vendor preferences.</p>
Scroll to Top