Cisco & NVIDIA unveil Secure AI Factory to boost safety for enterprises

Cisco has introduced an AI infrastructure architecture in collaboration with NVIDIA, with a focus on embedding security across all layers of enterprise AI deployments.

The joint initiative, dubbed the Cisco Secure AI Factory with NVIDIA, is aimed at enabling customers to build and secure data centres for developing and managing artificial intelligence workloads. According to Cisco and NVIDIA, the offering is designed to address the requirements for both high-performance computing and comprehensive protection, responding to the growing importance of AI-specific threat vectors and regulatory frameworks.

Chuck Robbins, Chair and Chief Executive Officer at Cisco, stated: "AI can unlock groundbreaking opportunities for the enterprise. To achieve this, the integration of networking and security is essential. Cisco and NVIDIA's trusted, innovative solutions empower our customers to harness AI's full potential simply and securely."

Jensen Huang, Founder and Chief Executive Officer, NVIDIA, said: "AI factories are transforming every industry, and security must be built into every layer to protect data, applications and infrastructure. Together, NVIDIA and Cisco are creating the blueprint for secure AI—giving enterprises the foundation they need to confidently scale AI while safeguarding their most valuable assets."

The companies have outlined that the new architecture sets out to simplify the process for enterprises to deploy, manage and secure AI infrastructure at any scale. Cisco is integrating solutions such as Cisco Hypershield to help protect AI workloads, as well as Cisco AI Defense, which is intended to support the secure development, deployment and utilisation of AI models and applications. These elements are incorporated into the AI factory on top of the NVIDIA Spectrum-X Ethernet networking platform.

The Cisco Secure AI Factory with NVIDIA is positioned to provide enterprises with flexibility when designing infrastructure for tailored AI requirements, combining networking, security and AI technologies. The companies aim to support modular and scalable deployment of AI factories that encompass both computational demands and emerging security considerations associated with artificial intelligence.

Recent analysis by Cisco, through its State of AI Security report, underscores the changing landscape of AI-related threats and laws. The report evaluates dozens of AI-specific threat vectors and more than 700 pieces of relevant legislation. Cisco emphasises that companies which address their AI infrastructure and security concerns simultaneously are better placed to scale swiftly and extract faster business value amid evolving digital risk scenarios.

The architecture detailed by Cisco and NVIDIA incorporates technologies from a group of ecosystem partners. Compute resources will use Cisco UCS AI servers alongside NVIDIA HGX and MGX systems. Networking is to be based around Cisco Nexus Hyperfabric AI and Nexus hardware supported by Silicon One and Spectrum-X Ethernet technology from NVIDIA. Storage solutions will come from certified partners including Pure Storage, Hitachi Vantara, NetApp, and VAST Data. Enterprises can further leverage the NVIDIA AI Enterprise software suite for development and deployment of AI workloads.

On security, the Cisco Secure AI Factory architecture includes the Hybrid Mesh Firewall for unified security management across network switches, firewalls, and workload agents. This approach is intended to provide consistent policy enforcement, deep packet inspection and the capacity to detect and contain security threats within the infrastructure. Cisco Hypershield is expected to expand its security coverage by integrating with NVIDIA BlueField-3 DPUs, supporting zero-trust principles at the level of individual AI nodes.

For AI workloads, Cisco Hypershield offers lateral movement prevention and proactive vulnerability mitigation without the requirement for patching, using a single management interface. Planned enhancements will increase workload protection capabilities with NVIDIA BlueField-3's DOCA AppShield, providing real-time threat detection for workloads in virtual machines and containers.

Cisco AI Defense addresses both safety and security risks within AI applications by enabling automated vulnerability testing and runtime security across model lifecycles. The tool integrates with CI/CD workflows and supports compliance with recognised standards such as NIST, MITRE ATLAS, and the OWASP LLM Top 10. Future integration with NVIDIA AI Enterprise is planned to further streamline security workflows for AI development teams.

Patrick Moorhead, Founder, Chief Executive Officer and Chief Analyst at Moor Insights & Strategy, commented: "In today's fast-moving market, businesses need more than just technology—they need end-to-end solutions that address their most pressing challenges. I see Cisco and NVIDIA combining their strengths to deliver integrated solutions that I believe will drive innovation, simplify deployment, and streamline operations. AI isn't easy but the combination of the two could be an 'easy button' for AI infrastructure. By making AI infrastructure easier to adopt and manage, they could empower enterprises to accelerate digital transformation and achieve their strategic goals with more confidence."

Cisco and NVIDIA plan to offer multiple deployment routes, including a ready-to-deploy model with vertically integrated AI solutions and a build-your-own option enabling modular customisation. These configurations are designed to support enterprises at different points in their AI adoption journeys.

Cisco has produced new reference architectures featuring deployment options such as Cisco Nexus Hyperfabric AI and Cisco Nexus 9000 Series Switches, validated in conjunction with the NVIDIA Enterprise Reference Architecture for HGX H200 and Spectrum-X platforms. Individual technology components outlined are already available, while complete solutions based on the joint architecture are projected to be available for purchase before the end of 2025.