Cisco on August 25, 2026 announced an expansion of its Secure AI Factory with NVIDIA, adding rack-scale AI computing through a partnership with Supermicro. The move is aimed at enterprises, neoclouds, and sovereign clouds, which puts it in front of operators planning power, cooling, networking, and procurement, not only AI platform teams.
What changed
The official announcement is titled “Cisco Expands Secure AI Factory with NVIDIA for the Rack-Scale Era.” Cisco said the expansion adds rack-scale AI computing solutions to its portfolio through a partnership with Supermicro.
Cisco positioned the architecture for three explicit deployment targets: enterprises, neoclouds, and sovereign clouds. Cisco also said the solutions are NVIDIA Cloud Partner compliant; that is Cisco’s stated compliance positioning, not an independent analysis of security or regulatory certification.
On workload scope, Cisco said the architecture supports trillion-parameter training and high-throughput inference, and named the platforms behind it: NVIDIA Vera Rubin NVL72 and NVIDIA HGX Rubin NVL8. Cisco described the span as reaching from trillion-parameter training models to edge inferencing.
Cisco disclosed a specific network stack for the design: Silicon One-based switches on the front end, NVIDIA Spectrum-X-based switches on the back end, and Nexus One as the unifier for the networking architecture. Cisco also named a software control layer, saying Cisco Cloud Control is used with NVIDIA AI Enterprise software and AgenticOps.
One line in the announcement carries Cisco’s core differentiation argument: the company says it is the only NVIDIA technology partner using its own networking switches and network operating system inside an NCP-compliant solution. That is Cisco’s own positioning claim, not a verified survey of what competing partners ship.
For infrastructure teams, the practical detail is that the offering includes both liquid- and air-cooled Supermicro systems. Cisco said customers can deploy rack-to-fabric liquid cooling with Cisco and Supermicro systems, and secondary reporting said Cisco N9000 Series switches are liquid-cooled and interoperate with Supermicro liquid-cooled compute.
Cisco set the date itself: it said it will begin offering Supermicro compute solutions as part of the Secure AI Factory with NVIDIA in October 2026.
Why B2B teams should care
Cisco is aiming this below the hyperscalers as well as at them: its own announcement names enterprises, neoclouds, and sovereign clouds, while SiliconANGLE listed hyperscale alongside the other two. What Cisco is selling into those segments is a validated full-stack architecture, pitched at buyers for whom control over networking, operations, and deployment model matters as much as raw accelerator access.
That changes the evaluation lens for B2B infrastructure teams. Instead of asking only which GPU platform is available, buyers need to assess whether the full stack — networking, cooling, management, and architecture validation — can be operated inside their own compliance and governance boundaries.
SiliconANGLE reported that NVIDIA Vera Rubin NVL72 can exceed 200 kilowatts per rack. That figure is tied to that NVIDIA system reference, not to Cisco’s whole portfolio, but it matters because rack-scale AI decisions now intersect with power delivery, heat removal, and data center floor design.
Who is affected
Neocloud operators and sovereign cloud builders are the named targets, and Cisco is pitching them the same two things: rack-scale compute paired with a networking design meant for large training and inference environments, and NCP compliance as the procurement credential for controlled environments.
Enterprise AI platform teams evaluating the architecture will need to translate it into internal platform standards, particularly where Cisco is extending Enterprise Reference Architectures to the latest generation of NVIDIA AI infrastructure.
Data center operations teams are affected because rack-scale AI deployment here is not just a server procurement exercise. Cooling method, rack power density, switch interoperability, and liquid-cooling integration all become prerequisites when buyers evaluate Supermicro compute with Cisco networking.
What teams should check now
- Verify whether current facilities can support liquid cooling and very high rack power densities.
- Check whether the network design aligns with Cisco’s disclosed architecture.
- Determine whether Cisco’s stated NVIDIA Cloud Partner-compliant reference architectures matter to your procurement model, partner strategy, or sovereign cloud review process.
- Assess the role of the new Cisco Validated Infrastructure Services (CVIS), which Cisco says certifies infrastructure is built as designed and aligned with reference architectures, and which Cisco describes as aligned with the NVIDIA Infrastructure Services offer.
- Treat Cisco’s statement that it is investing in a dedicated large-scale AI Lab as a forward-looking validation and tooling signal, not evidence of current customer adoption.
- Map how Cisco Cloud Control, NVIDIA AI Enterprise software, and AgenticOps would fit into the existing operations stack, including whether the promised correlation of job health with compute, NIC, optics, and network metrics replaces or duplicates current monitoring.
- Press on delivery timing: Cisco frames the Supermicro partnership as supply-chain risk mitigation and names GPU and memory access as the constraint it is meant to address.
What remains unclear
- Not yet confirmed: pricing for the expanded Cisco Secure AI Factory with NVIDIA portfolio.
- Not yet confirmed: shipment volumes, customer counts, or revenue impact tied to the expansion.
- Not yet confirmed: the detailed launch scope beyond the described rack-scale portfolio and architecture elements.
- Not yet confirmed: broader availability timing for the full portfolio beyond Cisco’s stated October 2026 start for Supermicro compute solutions.
- Not yet confirmed: deeper configuration and SKU detail for the Cisco, Supermicro, and NVIDIA combinations described in the announcement.
- Not yet confirmed: any deployment of the new Supermicro-based systems. The one customer quoted in the announcement, Sharon AI, speaks to NCP validation in general terms, and its quote is vendor-supplied.
What to watch next
The next concrete milestone is October 2026, when Cisco says it will begin offering Supermicro compute solutions as part of the Secure AI Factory with NVIDIA.
Buyers should also watch for more complete SKU and configuration detail, the first disclosed deployments of the Supermicro-based systems, and more specific information from Cisco on validated architectures, software tooling, and operational security controls.
