Cisco (NASDAQ: CSCO) has expanded its Secure AI Factory with NVIDIA into rack-scale territory, teaming with Supermicro to add high-density, liquid- and air-cooled compute to a full-stack, validated architecture aimed at enterprises, neoclouds and sovereign clouds, the company announced on August 25, 2026.
Rack-scale compute meets Cisco networking
Under the deal, Cisco will resell Supermicro's liquid- and air-cooled server systems, validated and sold as part of the broader Cisco AI infrastructure portfolio. The combination unlocks rack-to-fabric liquid cooling for platforms including NVIDIA's Vera Rubin NVL72 and HGX Rubin NVL8, letting operators run trillion-parameter training and high-throughput inference alongside conventional workloads. Cisco's Silicon One switches sit on the front-end while NVIDIA Spectrum-X-based Nexus 9100 switches handle the back-end, unified by Cisco Nexus One.
Neocloud- and sovereign-cloud-ready
The expanded Secure AI Factory is delivered as an NVIDIA Cloud Partner (NCP) compliant reference architecture — Cisco is the only NVIDIA technology partner to use its own switches and NOS inside an NCP-compliant stack. New Cisco Validated Infrastructure Services (CVIS), aligned with NVIDIA Infrastructure Services (NVIS), certify that deployed clusters match reference designs. NVIDIA AI Enterprise software runs through Cisco Cloud Control's AgenticOps, correlating job health with compute, NIC, optics and network telemetry for end-to-end observability.
Availability and market context
Cisco will begin offering Supermicro compute solutions inside the Secure AI Factory with NVIDIA in October 2026. The move lands amid the biggest datacenter buildout in history, and just weeks after NVIDIA's Cloverleaf partnership on greenfield AI-factory sites and the AWS-NVIDIA Vera CPU deployment. Cisco cites Sharon AI, WWT, ePlus and NTT DATA as launch partners, and points to NVIDIA's $105B OpenAI commitment as evidence that rack-scale demand is only accelerating.
Reporting based on coverage from Cisco Newsroom, SiliconANGLE and HPCwire.
