Cisco has expanded its Secure AI Factory with NVIDIA initiative with a very clear message for 2026: enterprise AI no longer needs to be deployed only in large data centers but also at edge locations where inference must run close to the data and almost in real-time. The company unveiled this expansion at NVIDIA GTC 2026, targeting businesses, service providers, neoclouds, and sovereign clouds seeking to move from isolated pilots to production deployments without stitching multiple components separately. Cisco claims this architecture can reduce deployment times from months to weeks, though their announcement also notes that many of the announced functions are still subject to future availability.
This announcement fairly well reflects the direction in which the market is heading. Inference is no longer seen solely as a centralized load on large clusters, but as something that must also run in hospitals, warehouses, industrial plants, or even connected vehicles—where data is generated and delayed decisions lose value. Cisco proposes that this shift requires redesigning infrastructure to support AI “from core to edge,” with integrated networks, security, and operations from the outset.
More edge, more network, and more choice
One of the key elements of the announcement is support for NVIDIA RTX PRO 4500 Blackwell Server Edition within the Cisco UCS and Cisco Unified Edge families, aiming to bring critical AI workloads to edge locations without relying on data-center-scale hardware at each site. Cisco also introduced Cisco AI Grid with NVIDIA, a reference design combining its mobility service platform with RTX PRO Blackwell GPUs to enable managed AI at the edge focused on reliability and sovereignty.
In terms of networking, Cisco aims to strengthen its position with both proprietary silicon and NVIDIA technology. The manufacturer announced a new Cisco N9100 of 102.4 Tbps based on NVIDIA Spectrum-6, complementing the already available N9100 of 800G based on Spectrum-4. Additionally, Nexus Hyperfabric, now integrated into Cisco Nexus One, will support the N9000 series, including the N9100 family. This aims to simplify traditional complex integrations involving multiple vendors and provides two validated pathways for building large “AI factories”: one aligned with the NVIDIA Cloud Partner program and another based on Cisco Silicon One, following similar design principles.
This is a significant detail. Cisco is not betting on a single closed architecture but on a dual approach that preserves its narrative of flexibility: customers preferring a design close to the NVIDIA ecosystem can pursue that, and those favoring a more Cisco-centric infrastructure based on Silicon One and its own stack will have a defined path. In a market where networking is increasingly becoming a bottleneck for large-scale AI, this choice capability is part of the commercial argument. This interpretation is a reasoned inference from Cisco’s official announcement.
Integrated security in DPUs and agents
The second major aspect of the announcement centers on security. Cisco has decided to extend its Secure AI Factory to include Hybrid Mesh Firewall protection for NVIDIA BlueField DPUs integrated into GPU servers connected via Cisco Nexus One fabrics. The straightforward thesis is that, given AI workloads are high-value assets, protection cannot be limited to perimeter or hypervisor security but must also be applied closer to the server and internal traffic moving data and models. Cisco assures that this extension enables threat blocking at the server level without performance penalties—an assertion that forms part of its positioning but remains to be validated in real deployments.
The other layer of security involves agents. Cisco states that AI Defense integrates with NVIDIA NeMo Guardrails, part of NVIDIA AI Enterprise, providing specific controls for distributed AI agents—especially in edge scenarios where multiple agents interact with each other and central systems. The company presents this as a way to reduce risks in agent-to-agent interactions and to apply protections not only to models but also to operational behaviors of the system.
AI Defense will support NVIDIA OpenShell, NVIDIA’s new open platform for agent development. Described as part of NVIDIA’s Agent Toolkit, OpenShell aims to define how agents access data, use tools, and operate within policy boundaries. In this context, Cisco plans to contribute guardrails and continuous validation of agent actions, aligning with the growing corporate concern over autonomous systems capable of executing tasks beyond simple chat functions.
From lab to operations
The most significant aspect of this movement may not be a specific hardware or software component but the strategic direction it signals. Cisco positions itself as a company capable of connecting networks, security, and AI operations within a unified narrative—especially as many companies begin to realize that the real challenge isn’t just testing a model but operating it reliably, at scale, and securely. Cisco emphasizes that it is taking AI “from testing to production,” with security integrated from silicon up through software. Meanwhile, NVIDIA leverages the partnership to reinforce its message that “AI factories” represent a new industrial layer of computing.
That said, it’s important to interpret the announcement with the caveat Cisco adds at the end of its statement: many of the products and functions described are still under development and will be available when ready. This nuance doesn’t diminish the strategic importance of the movement but underscores that enterprise AI continues to evolve through roadmaps, gradual integrations, and promises that still need to prove themselves in real-world deployments.
Frequently Asked Questions
What exactly has Cisco expanded in its Secure AI Factory with NVIDIA?
Cisco has extended its architecture to encompass not only large data centers but also edge environments, introducing new reference designs, support for NVIDIA RTX PRO Blackwell GPUs, new networking options, and additional security layers for infrastructure and agents.
What role do NVIDIA BlueField DPUs play in this announcement?
Cisco will extend Hybrid Mesh Firewall policies to BlueField DPUs embedded in GPU servers connected via Cisco Nexus One fabrics, aiming to protect AI workloads closer to the server.
What is OpenShell, and why is it part of this alliance?
OpenShell is an open agent development platform introduced by NVIDIA as part of its Agent Toolkit. Cisco AI Defense will add controls and guardrails to govern these agents and their tools.
Is everything announced already available for purchase and deployment?
Not necessarily. Cisco explicitly states that many of the described products and functions are in different development phases and will be released when available.
via: newsroom.cisco