Cisco’s Secure AI Factory with NVIDIA expands to include integrated Supermicro rack-scale systems to meet larger AI cluster demands.
Cisco this week announced that its Secure AI Factory with NVIDIA is expanding to rack-scale infrastructure built around Supermicro liquid-cooled and air-cooled systems. Anchored by NVIDIA Cloud Partner Reference Architecture compliance, the design integrates Cisco Validated Infrastructure Services (CVIS), Cisco AI networking, and unified operations. Cisco plans to pair its reference architecture, networking gear, validation services, and management software with Supermicro’s rack-scale systems, launching in October through its enterprise sales and channel partner ecosystem. Cisco support and lifecycle services will be bundled alongside the hardware, along with the company’s Ethernet silicon, optics, security, and observability portfolio.
The reference design divides the network across two switch families. The frontend fabric runs on Cisco N9300 Series switches powered by Cisco Silicon One, while the backend fabric utilizes Cisco N9100 Series switches built around NVIDIA Spectrum-X silicon. Cisco positions this architecture for clusters spanning roughly 1,000 to more than 100,000 GPUs, mapping it directly onto NVIDIA’s NVL72 builds for Vera Rubin and Grace Blackwell, HGX NVL8 systems for Rubin and B300, and MGX PCIe GPU platforms.
Cisco segments its validated designs by scale. The Enterprise Reference Architecture targets AI server clusters under roughly 1,000 GPUs using Cisco Silicon One N9300 Series switching, while the Cloud Reference Architecture extends from approximately 1,000 to over 100,000 GPUs, pairing an N9300 frontend with an N9100 Series backend carrying NVIDIA Spectrum-X silicon. These Supermicro-focused deployments are positioned beyond Cisco’s existing Unified Edge and AI POD offerings, targeting model training and trillion-parameter workloads rather than edge inferencing or optimization.
CVIS serves as the operational backbone for the build. A CVIS Toolkit handles repeatable provisioning and automated validation, while specialist-led delivery ensures handover of a fully validated cluster complete with performance and compliance reporting. Dedicated CVIS performance clusters further support software testing and tool development. For physical maintenance, such as replacing an SSD or GPU within a Supermicro server, Supermicro will handle the direct hardware service.
The planned Supermicro lineup covers a significant portion of NVIDIA’s accelerated computing portfolio, including Supermicro MGX systems, air-cooled and liquid-cooled NVIDIA HGX B300 NVL8 builds, NVIDIA HGX Rubin NVL8, Vera Rubin NVL72, GB300 NVL72 rack-scale systems, and Cisco’s UCS X-Series modular servers. At the rack scale, networking touches every fabric in the cluster. A management fabric connects via an N9364E-SG2-O switch with Cisco Silicon One, while frontend and backend fabrics utilize an N9164E-NS4-O with NVIDIA Spectrum-X silicon. Cisco states this same fabric configuration can scale beyond 100,000 GPUs using operational models already familiar to network teams.
Unified management converges on Cisco Cloud Control, scheduled for release in calendar Q4 2026. The console provides unified login, inventory, topology tracking, server management, infrastructure power and cooling oversight, and network management, integrating Intersight and Nexus One into a single Day 0–2 operating model. Operators gain a comprehensive view of the entire AI cluster, with Cloud Control rendering the cluster fabric, frontend/backend/storage/management switches, servers, and storage. An integrated AI Canvas delivers continuous visibility and streamlined troubleshooting.
Cisco is enveloping Supermicro’s rack-scale AI infrastructure with its proprietary networking, validated architectures, sales channels, support, lifecycle services, and management layer. This collaboration eliminates the need for enterprises standardizing on Cisco networking to separately integrate Supermicro hardware, third-party switches, and disparate management tools. Regarding the strategic rationale behind the partnership, Patrick Kennedy, Editor-in-Chief, noted: “Supermicro has a high-volume AI cluster business, including all integration, testing, and deployment facilities. They also have a much faster development cycle for new servers than Cisco UCS, which is of paramount importance for AI infrastructure where generations advance much more rapidly. I asked Cisco about whether the outside-the-rack components like chillers, in-row CDUs, and so forth, and it sounds like those will be available under a reseller agreement, but they require additional engineering versus just delivering a rack.”