AI demands a fundamental rethinking of how we design, build, and secure data centers. Organizations are moving past the experimentation phase and are deploying massive AI clusters that require unprecedented network bandwidth, power, and security controls. Building these giga-scale environments isn’t just about adding more GPUs. It requires a holistic, deeply integrated architecture that ensures every component—from the silicon to systems to software and the operating model—works in perfect harmony.

Cisco is delivering the foundational infrastructure needed to make this a reality. By combining our networking expertise with advanced silicon, systems, optics, software, operating models, and security innovations, we’re providing enterprises, neoclouds, and sovereign cloud providers with the tools they need to deploy AI securely at scale. Cisco stands out as the only vendor capable of delivering a true turnkey solution, seamlessly tailored to meet the demands of customers at every scale.

Through our partnership with NVIDIA, we are pushing the boundaries of what AI networks can achieve, focusing on three critical pillars: infrastructure, networking, and security.

Built for AI factories

The transition to giga-scale AI requires hardware capable of handling massive data throughput with minimal latency. Building on the expansion of the recently announced N9300 Silicon One G300 scale-out and 51.2T P200 scale-across systems, Cisco is raising the bar to meet these intense demands.

We are thrilled to introduce the new Cisco N9100 Series Switch, N9164F-NS6, powered by NVIDIA Spectrum-6 silicon. This 102.4T switch delivers a massive leap in data center capacity to support next-generation secure AI factories. We are also making the N9100 switches available with NVIDIA Spectrum-4 silicon, giving enterprises, neoclouds, and sovereign cloud providers flexible, high-performance deployment options.

We believe in the power of silicon diversity. This strategy allows you to choose the exact technology that fits your specific performance and operational needs. To keep implementation simple, we ensure full reference architecture compliance. This streamlines your deployment and guarantees smooth integration into complex data center environments.

Cisco keeps raising the performance ceiling, making advanced AI infrastructure faster and easier to deploy. We introduced the N9100 switches, powered by NVIDIA Spectrum-6 Ethernet silicon, to address the immense scale required by next-generation secure AI factories. This 100% liquid-cooled 102.4T switch represents a major leap forward in AI data center capacity.

In addition, the N9100 switch, powered by NVIDIA Spectrum-4 silicon, is also now available, further expanding deployment options for enterprises, neoclouds, and sovereign cloud providers.

To make deploying these complex systems easier, we integrated Nexus One support with cloud-managed Cisco Nexus Hyperfabric. This creates a turnkey, full-stack AI solution. AI builders no longer have to piece together disparate components and hope they function efficiently. Nexus Hyperfabric also now manages the newly available Cisco N9164E-NS4-O, powered by NVIDIA Spectrum-4 Ethernet silicon. It offers pods of plug-and-play data center fabrics, managed entirely through the cloud, dramatically reducing the time it takes to go from procurement to full operation. Nexus One includes a robust, on-premises managed Nexus dashboard that is a proven operations, automation data center, and AI networking solution.

Flexibility remains crucial when designing these environments. Organizations have different needs based on their size, security requirements, and existing infrastructure. To accommodate this, we offer robust reference architectures tailored to specific deployment types:

• Cisco Cloud Reference Architecture (CRA): For enterprises and neoclouds deploying specialized AI servers with greater than 1000 GPUs and up to 32,000 GPUs, the Cisco CRA provides a highly optimized, scalable path forward using industry-leading Cisco Silicon One and Cisco N9300 Series Switches. For deployments of fewer than 1000 GPUs, we have ready-to-consume Cisco Enterprise Reference Architecture (ERA).
• Compliant with NVIDIA Cloud Partner Reference Design: For large-scale AI infrastructures ranging from 1000 to 32,000 GPUs, the Cisco N9100 Series Switches fully comply with the NVIDIA Cloud Partner (NCP) Reference Architecture. This ensures ultimate performance for sovereign and neocloud deployments, utilizing scale-out fabrics powered by Spectrum-X Ethernet switch silicon.

By offering a choice between NCP Reference Design and Cisco CRA architectures, we ensure that every customer has a proven, validated blueprint for success, whether they’re building a massive sovereign cloud or a highly targeted enterprise AI cluster.

Securing the AI workload at the edge

As AI clusters grow in power and complexity, they become highly attractive targets for malicious actors. Traditional perimeter security models fail in these environments. Unprotected east-west traffic within the AI fabric allows lateral threats to spread rapidly. A compromised AI workload could lead to GPU resource hijacking or massive data exfiltration, resulting in a full lateral blast radius. However, running traditional security agents directly on the AI servers taxes the CPU and GPU, draining the very compute resources you’re trying to maximize.

To solve this, we’re fundamentally shifting where and how security is enforced. We extended Cisco Hybrid Mesh Firewall support directly to NVIDIA BlueField DPUs.

This innovation brings security as close to the workload as possible without compromising performance. By embedding the firewall on the DPU, we deliver integrated inline security with high-performance scalability. Administrators can define security policies once and enforce them everywhere, isolating virtual public clouds (VPCs) and blocking lateral attacks in real time. This provides choke-point-free enforcement, protecting front-end fabric servers while leaving the host CPU and GPU entirely dedicated to processing AI workloads.

To learn more about how we are protecting AI environments from lateral threats without sacrificing performance, read our detailed technical breakdown: Cisco secures AI infrastructure with NVIDIA BlueField DPUs.

Advancing AI networking for peak performance

Even the most powerful GPUs will sit idle if the network can’t feed them data fast enough. Maximizing GPU utilization and optimizing the key-value (KV) cache performance requires intelligent, high-speed connectivity across the entire fabric. This is where Cisco Nexus One fundamentally changes the game for AI networking.

Nexus One provides a unified management plane across both NX-OS and SONiC environments with an on-premises managed Nexus Dashboard and cloud-managed Nexus Hyperfabric as operational models.

Nexus Dashboard delivers unprecedented visibility into the full-stack AI environment with Cisco N9000 Series Switches. Administrators gain access to deep AI job-monitoring capabilities with Nexus Dashboard 4.2. Track exactly how data moves through the fabric, identifying bottlenecks before they impact training times. The system features intelligent, auto-adjusting load balancing and telemetry-based congestion control. Instead of relying on static routing protocols that break down under the unique, elephant-flow traffic patterns of AI workloads, the network dynamically adjusts to ensure optimal data delivery.

Furthermore, Nexus Dashboard enables real-time GPU health tracking. Network operators can see directly into the performance metrics of the compute layer, ensuring that every expensive GPU resource operates at maximum efficiency. Whether you’re building a scale-out network within a single data center or a scale-across network connecting multiple facilities with high-performance, low-latency links, Nexus Dashboard ensures the network acts as a powerful accelerator rather than a bottleneck.

Empowering the next generation of innovation

The AI revolution requires more than just raw processing power. It demands a fully integrated ecosystem where infrastructure, networking, and security are designed to operate as a single, cohesive unit.

Through our continued innovations with the Spectrum-X-powered Cisco N9100 Series Switches, the intelligent management of Nexus One, and the edge-enforced security of the Cisco Hybrid Mesh Firewall on BlueField DPUs, Cisco provides the foundation that enterprises, neoclouds, and sovereign clouds need to scale their AI ambitions. We’re building the networks that will power the next decade of discovery, ensuring they are fast, reliable, and fundamentally secure.

Explore AI networking in data center