How Cisco N9100 Will Reshape AI Data Centre Networking

Cisco has introduced the N9100 series switch at GTC in Washington DC, positioning itself as the first partner to deliver a Nvidia Cloud Partner-compliant reference architecture. The switch represents the first Nvidia partner-developed data centre switch based on Nvidia Spectrum-X Ethernet switch silicon, with implications across neocloud, enterprise and telecom markets.

Cisco’s portfolio of Nexus data centre switching solutions delivers a unified operating model through Cisco Nexus Dashboard across Silicon One, Cloud-scale ASICs and switches built on Spectrum-X Ethernet switch silicon. This approach aims to provide operational consistency as organisations scale their AI infrastructure.

“We’re at the beginning of the largest data centre build-out in history,” says Jeetu Patel, President and Chief Product Officer at Cisco. “The infrastructure that will power the agentic AI applications and innovation of the future requires new architectures designed to overcome today’s constraints in power, computing and network performance. Together, Cisco and Nvidia are leading the way in defining the technologies that will power these AI-ready data centres in all their varieties, from emerging neoclouds, to global service providers, to enterprises and beyond.”

Cisco Cloud Reference Architecture targets neocloud customers

The partnership between Cisco and Nvidia extends beyond the N9100 hardware. Cisco has developed the Cisco Cloud Reference Architecture for neocloud and sovereign cloud customers, based on the design tenets of Nvidia’s Cloud Partner reference architecture. The architecture utilises Cisco’s Silicon One and Cloud-scale ASIC offerings and will include the Cisco 8223 based on the Silicon One P200 for scale-across networks, Nvidia BlueField-4 DPUs and Nvidia ConnectX-9 SuperNICs.

Gilad Shainer, SVP of Networking at Nvidia, says: “Nvidia Spectrum-X Ethernet delivers the performance of accelerated networking for Ethernet. Working with Cisco’s Cloud Reference Architectures and Nvidia Cloud Partner design principles, customers can choose to deploy Spectrum-X Ethernet using the newest Cisco N9100 series or Cisco Silicon One based switches to build open, high-performance AI networks.”

The reference architecture addresses the integration challenges that emerge when organisations scale AI infrastructure. Back-end and front-end Ethernet-based networks must work with existing infrastructure whilst remaining deployable across thousands of GPUs. Xiaohe Hu, CEO of Infrawaves, says: “The real challenge in AI infrastructure isn’t just performance, it’s maintaining operational sanity as you scale from dozens to thousands of GPUs. Cisco’s approach with NX-OS and Nexus Dashboard creates a single pane of glass across our entire AI fabric, whether we’re optimising inference latency in the front-end or maximising training throughput in the back-end. That operational simplicity translates directly to faster deployments and lower TCO.”

Cisco Secure AI Factory with Nvidia adds security integrations

Cisco has expanded the Cisco Secure AI Factory with Nvidia, which launched at GTC in March 2025. The architecture combines AI infrastructure with security and observability through Cisco AI PODs and Cisco Silicon One-powered Nexus switching.

The security integration centres on Cisco AI Defense, which now works with Nvidia NeMo Guardrails to deliver cybersecurity for AI applications. Organisations can deploy Cisco AI Defense on-premises, enabling security and AI teams to protect models and applications whilst controlling what data leaves their data centres.

Splunk Observability Cloud provides monitoring capabilities across the AI application stack, including insights into AI infrastructure health with Cisco AI PODs. Splunk Enterprise Security extends this visibility to protect AI workloads, giving teams a view of performance, quality, security and cost.

The infrastructure components available through AI PODs have expanded. Cisco Isovalent has been validated for inference workloads on AI PODs, enabling Kubernetes networking. Cisco Nexus Hyperfabric AI with a cloud-managed Cisco G200 Silicon One switch delivers 800G Ethernet and is now orderable as a deployment option in AI PODs. Cisco UCS 880A M8 rack servers with Nvidia HGX B300, and the Cisco UCS X-Series modular servers with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs are also orderable as part of AI PODs, enabling GPU support for workloads including fine-tuning and inference.

Nvidia Run:ai software is available through Cisco and its partners, enabling AI workload and GPU orchestration capabilities. The ecosystem now includes Nutanix, with Nutanix Kubernetes Platform as a supported Kubernetes platform and Nutanix Unified Storage as a supported storage option. Nutanix Enterprise AI solution serves as the interoperable software component that simplifies building and operating containerised inference services.

Cisco is collaborating with Nvidia on the Nvidia AI Factory for Government, a reference design for AI workloads deployed in regulated environments. This alignment addresses the requirements organisations face when deploying AI infrastructure in government settings.

Cisco and Nvidia develop AI-native wireless stack for 6G

Wireless networks face pressure to support connections at scale as AI moves from smartphones to augmented reality glasses, connected cars and robotics. Cisco, Nvidia and telecom partners have developed the first American AI-RAN stack for mobile networks that integrates sensing and communication, with pre-6G applications being showcased at Nvidia GTC DC.

The stack allows telecom providers to integrate AI into mobile networks, starting with 5G services and establishing groundwork for 6G. The technology combines Cisco’s user plane function and 5G core software with the Nvidia AI Aerial platform, creating a foundation that enables physical AI and integrated sensing.