Cisco drives AI networking innovation with NVIDIA

In February of this year, Cisco announced an expanded partnership with NVIDIA to combine two of the most widely used networking portfolios: Cisco Silicon One and NVIDIA Spectrum-X Ethernet. This collaboration aimed to provide enterprises with unparalleled flexibility and options for their AI data center investments.

I’m delighted to announce three significant milestones in our partnership:

• The Cisco Cloud Reference Architecture, featuring switches based on Cisco Silicon One with the add-on license for deeper integration with NVIDIA Spectrum-X Ethernet technology.
• A Cisco NVIDIA Cloud Partner(NCP)-compliant Reference Architecture, built on the NVIDIA Spectrum-X Ethernet silicon–powered Cisco N9100 Series switches, with early orderability in November 2025.
• The orderability in November 2025 of Cisco Nexus Hyperfabric AI.

These milestones highlight the strength of our partnership and demonstrate our commitment to innovation and excellence in delivering cutting-edge solutions to our customers.

Cisco Cloud Reference Architecture

The new Cisco Cloud Reference Architecture is based on NCP design tenets for Cisco Silicon One based switches with an add-on license for deeper integration with NVIDIA Spectrum-X Ethernet technology.

This add-on license enables the NVIDIA Spectrum-X adaptive routing capability to work seamlessly with Cisco Nexus 9300 Series switches and NVIDIA Spectrum-X Ethernet SuperNICs. The combination of low latency and congestion-aware per-packet load balancing on Cisco 9300 switches, along with out-of-order packet handling and end-to-end congestion management on NVIDIA SuperNICs, significantly enhances network performance. These improvements are essential for AI networks, optimizing critical metrics such as job completion time.

Cisco NVIDIA Cloud Partner Compliant Reference Architecture

With the Cisco N9100 Series, we now introduce to the market a NVIDIA Cloud Partner(NCP)-compliant reference architecture. This development is particularly significant for neocloud and sovereign cloud customers building data centers with capacities ranging from thousands to potentially hundreds of thousands of GPUs, as it allows them to diversify their supply chains effectively.

The Cisco N9100 Series delivers an impressive 51.2 Tbps of bandwidth, ensuring exceptional scalability, performance, and security. With the addition of NVIDIA Spectrum-X Ethernet silicon alongside Cisco Silicon One and Cisco Cloud Scale ASICs, Cisco offers a wider range of switch silicon options. This allows customers to standardize on a unified AI networking platform that supports both Cisco Silicon One and NVIDIA Spectrum-X Ethernet silicon, streamlining management through the Cisco Nexus Dashboard. Additionally, users have the flexibility to choose between different Network Operating Systems, including Cisco NX-OS or the open-source SONiC.

The Cisco N9100 Series empowers organizations to efficiently deploy both AI backend and frontend networks, effectively meeting the demanding requirements of modern AI workloads

Cisco Nexus Hyperfabric AI

Our customers have shared that they feel pressure to deploy larger AI clusters, and they are seeking assistance from infrastructure partners like Cisco. We developed Cisco Nexus Hyperfabric AI solution specifically to help IT teams shorten their time to value realization and simplify the delivery and operation of complex AI infrastructure.

Since its initial announcement in June 2024, many customers and partners have expressed interest in its capabilities, among them: Wake Forest University, Sharon AI and Gruve.

“We are thrilled to participate in the Cisco Nexus Hyperfabric AI pilot, made possible through our collaborative partnerships with Cisco, VAST Data, and MCNC. This cutting-edge technology allows us to explore new ways for Wake Forest University faculty and students to access essential tools that will significantly accelerate their research, advance AI initiatives, and lead to groundbreaking discoveries.”
— Adam Carlson, Assistant Director of Research Computing

”Sharon AI is excited to test Cisco Nexus Hyperfabric AI because it offers a turnkey, cloud-managed, full-stack AI infrastructure solution that simplifies deployment and accelerates AI innovation, perfectly aligning with our mission to power sovereign AI solutions and AI factories.”
— Dan Mons, Chief Technology Officer at Sharon AI

“As a Cisco partner, Gruve views Cisco Nexus HyperFabric AI as a major step forward in simplifying AI infrastructure—integrating compute nodes, fabric interconnects, and VAST storage into a cohesive, scalable framework. Our partnership with Cisco enables us to accelerate customer deployments, enhance throughput, and ensure secure, high-performance AI operations.”
— Rajeev Khanolkar, Chief Strategy Officer of Gruve, Inc

As customers build their own AI clusters on premises, they deal with an increasing amount of complexity as they need to manage multiple systems with specialized technology stacks for which they don’t necessarily have the right resources with the right skills.

To simplify and accelerate their AI journey, Cisco is transforming the operational experience for deploying and managing AI clusters. Many customers invest in large GPU clusters but often spend months getting them fully operational, which can lead to a poor return on investment.

Cisco Nexus Hyperfabric AI aims to reduce deployment time from several months to just a few weeks or even less. By planning the fabric layout in advance (figure 1), creating detailed cabling plans (figure 2), and offering online guided assistance for installers (figure 3), we can significantly save time during the deployment process.

Figure1: Design an NVIDIA Enterprise Reference Architecture in one-click from a template (no guesswork)

Figure 2: Complex interconnect cabling layout is automatically generated and auto-documented by the Cisco Nexus Hyperfabric controller (no more spreadsheets)

Figure 3: Day-1 Deployment is automated using Cisco Nexus Hyperfabric’s on-site interface, which guides installers through all the steps of deployment.

This increased efficiency directly translates to a shorter time to first token and the simplified processes for deploying complex AI clusters encourage adoption among customers who may not have the necessary expertise or resources. 

These three milestones mark an exciting new chapter in the Cisco-NVIDIA partnership and provide a foundation for customers to meet the demanding network requirements of AI reasoning and other inference workloads.