Cisco is expanding Cisco Secure AI Factory with NVIDIA to rack-scale, liquid-cooled AI infrastructure, a full-stack architecture aligned to NVIDIA Cloud Partner, NVIDIA Enterprise, and Cisco reference architectures, and validated, sold, and supported by Cisco.
What stands between GPUs and first token
Power density has outrun air
An NVL72-class rack can exceed 200 kW. Conventional air-cooled racks are designed for roughly 30 to 50 kW. That gap is a facilities problem, not a preference.
Cooling as a system requirement
Beyond 200 kW, air cooling reaches its limit. Secure your high-density compute and fabric with liquid cooling to help ensure operational reliability and peak performance at any scale.
The supply squeeze
Avoid delays caused by component shortages. Get an integrated rack-scale architecture that streamlines GPU, CPU, and memory delivery to keep your AI buildouts on schedule.
The answer to integration complexity
Embrace a unified solution that helps ensure compute, networking, and software work in harmony, eliminating handoffs to accelerate your time to first token.
Cisco, NVIDIA, and Supermicro
What Cisco adds around the GPU
Reduce risk
Keep AI initiatives on track with rack-scale compute aligned to NVIDIA Cloud Partner and Enterprise reference architectures, supported by Cisco and Supermicro's manufacturing and supply chain strengths.
Speed time to value
Get AI infrastructure into production faster with Cisco Validated Infrastructure Services, which verifies deployment against the reference architecture and provides an evidence report.
Simplify operations
Spend less time troubleshooting AI infrastructure with unified management and AgenticOps. Cisco Cloud Control and Splunk provide visibility into job health across compute, NICs, optics, and networking.
Better together, by design.
Compute, built for rack scale
- NVIDIA-certified rack-scale AI system
- Supermicro Liquid-cooled systems
- Global supply chain execution
- Delivered on Cisco operating model
The fabric, security, and operations around the GPU
- Industry-leading Ethernet for AI
- One validated full stack
- Built-in security and observability
- Global channel ecosystem and financing
On-ramp
NVIDIA MGX PCIe GPU server
Inference, RAG, agents, VDI, and visual computing at lower power and cost
Premium bridge
Air-cooled HGX B300 8-GPU Server
Brings premium 8-GPU HGX performance where liquid cooling is not yet feasible
Flagship
Liquid-cooled HGX B300 8-GPU Server
Handles dense training and inference at high utilization in standard EIA racks
Flagship
Liquid-Cooled NVIDIA HGX Rubin NVL8 8-GPU Server
Works as one accelerator for high-utilization training and reasoning inference
Flagship
NVIDIA GB300 NVL72
Delivers frontier training and large-scale reasoning inference at maximum computing density
Available today
Flagship
NVIDIA GB300 NVL72
Functions as one accelerator for test-time scaling and reasoning inference at scale
Flagship
NVIDIA Vera Rubin NVL72
Frontier and trillion-parameter model training, the next flagship rack-scale tier
Modular premium
Liquid-cooled HGX B300 8-GPU Server
Dense training and inference at high utilization in standard EIA racks
Modular premium
Liquid-Cooled NVIDIA HGX Rubin NVL8 8-GPU Server
Scales extreme training and inference with next-gen density and lower cost per token
Tiered fleet
NVIDIA MGX PCIe GPU server
Delivers cost-efficient inference, visual compute, and vGPU tiers for fleets
AI factory
NVIDIA GB300 NVL72
Sovereign foundation model training and reasoning inference on infrastructure you govern
AI factory
NVIDIA Vera Rubin NVL72
Next-generation capacity for sovereign foundation models and reasoning inference
Regional clusters
Liquid-cooled HGX B300 8-GPU Server
Powers dense AI clusters for national labs, defense, healthcare, energy, and other regulated sectors
Regional clusters
Liquid-Cooled NVIDIA HGX Rubin NVL8 8-GPU Server
Scales next-gen density for national labs, defense, healthcare, energy, and other regulated sectors
Institutional Layer
NVIDIA MGX PCIe GPU servers
Enables simulation, graphics, and secure inference for institutions and the sovereign edge
Trust built in, not bolted on
Deploy with confidence, knowing every layer is validated and secured, from silicon to agents.
Enterprise AI: Get to production fast
Run AI on infrastructure your teams already know, with security and governance built into the stack.
Neoclouds and sovereign clouds: Turn GPU capacity into revenue
Dense, multi-tenant GPU infrastructure with the isolation and networking AI service providers depend on.
Cisco Secure AI Factory with NVIDIA
Get one validated architecture across compute, networking, and security—now extended to rack scale.
Rack-scale AI questions, answered
At the end of October 2026, organizations globally can order the full stack—Cisco, NVIDIA, Supermicro, and ecosystem technologies—through the Cisco authorized channel partner ecosystem.
Rack-scale systems such as NVIDIA NVL72 can exceed 200 kW per rack, where liquid cooling becomes a system-level requirement. Air-cooled options, including the 8RU HGX B300, serve facilities that cannot support liquid cooling today.
Cisco is your first point of contact for the entire stack, leading design, deployment, validation, and day-2 support, and engaging NVIDIA and ecosystem partners through an established collaborative support model.
Design your AI factory with Cisco
Features Accordion Item
Ligula ullamcorper malesuada proin libero nunc consequat interdum varius sit. Orci a scelerisque purus semper eget duis.
Rack-Scale AI Infrastructure | Cisco & NVIDIA
Ligula ullamcorper malesuada proin libero nunc consequat interdum varius sit. Orci a scelerisque purus semper eget duis.
Features Accordion Item
Ligula ullamcorper malesuada proin libero nunc consequat interdum varius sit. Orci a scelerisque purus semper eget duis.
Features Accordion Item
Ligula ullamcorper malesuada proin libero nunc consequat interdum varius sit. Orci a scelerisque purus semper eget duis.
Features Accordion Item
Ligula ullamcorper malesuada proin libero nunc consequat interdum varius sit. Orci a scelerisque purus semper eget duis.