Key Points
- Cisco partners with Supermicro to add rack-scale AI computing to Secure AI Factory
- Solutions will be NVIDIA Cloud Partner compliant for neocloud deployments
- Availability begins October 2026 with liquid and air-cooled compute options
Cisco has expanded its Secure AI Factory with NVIDIA by partnering with Supermicro to add rack-scale artificial intelligence computing systems to its infrastructure portfolio, targeting enterprises, neoclouds and sovereign cloud operators building large-scale AI workloads.
The expanded architecture, announced on Tuesday (25 August), combines Supermicro’s liquid-cooled and air-cooled server systems with Cisco’s networking infrastructure to deliver what the company calls a validated full-stack approach for AI deployments.
The solution will be compliant with NVIDIA Cloud Partner (NCP) requirements, meaning it meets NVIDIA’s specifications for cloud service providers running AI workloads.
“We are at the beginning of one of the largest datacenter buildouts in history,” said Jeetu Patel, president and chief product officer, Cisco. “Every organisation is racing to scale AI, but speed only counts if it comes with control of data, managed token costs, and real ROI.”
The partnership addresses a growing challenge for organisations deploying AI at scale. Training large language models and running high-throughput inference, the process of generating outputs from trained models, requires dense clusters of graphics processing units (GPUs) that generate substantial heat.
Liquid cooling systems circulate coolant directly to server components, enabling higher computing density than traditional air-cooled facilities can support.
Infrastructure components
The expanded offering supports NVIDIA’s latest AI platforms including Vera Rubin NVL72 and HGX Rubin NVL8, systems designed for training models with trillions of parameters.
Parameters are the internal variables that AI models adjust during training, with larger numbers generally enabling more sophisticated capabilities.
Cisco said its architecture uses its own Silicon One-based switches for front-end networking and NVIDIA Spectrum-X based switches for back-end traffic between GPUs.
These are unified through Cisco Nexus One, the company’s network management platform. Cisco claims to be the only NVIDIA technology partner using its own networking switches and operating system in an NCP-compliant solution.
By the numbers
- October 2026
- Availability date for Supermicro solutions
- Trillion parameters
- Training model scale supported
Justin Boitano, vice president of Enterprise AI, NVIDIA, said the expansion would help enterprises get AI infrastructure into production faster. “AI factories are revenue-generating infrastructure, where compute produces intelligence, and intelligence drives revenue,” he said.
The company is also introducing Cisco Validated Infrastructure Services (CVIS), aligned with NVIDIA’s Infrastructure Services certification. The programme certifies that deployed infrastructure matches reference architecture designs. Cisco said it is investing in a dedicated large-scale AI lab to develop testing tools for this certification process.
Cisco enterprise and sovereign cloud focus
The neocloud segment, comprising cloud providers that specialise in AI workloads rather than general-purpose computing, has emerged as a significant market as enterprises seek alternatives to hyperscale providers for AI training. Sovereign cloud deployments, where data must remain within national boundaries for regulatory compliance, represent another growing segment.
For operations management, the solution integrates NVIDIA AI Enterprise software with Cisco’s Cloud Control platform. This allows correlation between AI job performance and underlying compute and network metrics, providing observability across the full stack.
James Manning, co-founder and chief executive officer, Sharon AI, a customer cited by Cisco, said the NCP validation provided confidence that infrastructure would be optimised from deployment. “We no longer have to choose between performance, reliability or ease of management,” he said.
Neil Anderson, vice president and chief technology officer for cloud, infrastructure and AI solutions, WWT, said the partnership would enable more comprehensive full-stack solutions. “Our customers span almost every vertical industry, and our priority is always to offer them solutions and services that right-fit their organisation,” he said.
Ken Farber, president of strategy, software and alliances at ePlus, emphasised the focus on helping customers manage agentic workloads, AI systems that can take autonomous actions.
“This partnership enables us to offer a robust full-stack infrastructure spanning high-performance AI compute, networking, power, liquid cooling and rack-scale solutions,” he said.
Cisco said it will begin offering Supermicro compute solutions as part of the Secure AI Factory with NVIDIA in October 2026. The company did not disclose pricing or regional availability details.
Your Questions, Answered
What is the Cisco Secure AI Factory with NVIDIA?
It is Cisco's full-stack AI infrastructure architecture that combines networking, compute and management systems. The platform is designed for enterprises, neoclouds and sovereign cloud operators deploying large-scale AI workloads with integrated security.
What does NVIDIA Cloud Partner compliance mean?
NCP compliance indicates that infrastructure meets NVIDIA's technical specifications for cloud service providers running AI workloads. This certification ensures the architecture is validated for modern AI use cases and interoperability with NVIDIA platforms.
When will the Supermicro solutions be available?
Cisco will begin offering Supermicro compute solutions as part of the Secure AI Factory with NVIDIA in October 2026. The company has not disclosed regional pricing or availability details.
What AI platforms does the expanded architecture support?
The solution supports NVIDIA's latest AI infrastructure including Vera Rubin NVL72 and HGX Rubin NVL8 platforms. These systems are designed for training models with trillions of parameters and high-throughput inference workloads.
