Cisco Adds Rack-Scale Compute to Its AI Infrastructure Push

Cisco is taking its AI infrastructure strategy up a level, adding high-density rack-scale computing from Supermicro to its Secure AI Factory with NVIDIA. Announced recently, the move is aimed at enterprises, neocloud operators and sovereign clouds looking to deploy increasingly demanding AI workloads without piecing together complex infrastructure themselves.

The expanded portfolio will include Supermicro’s liquid- and air-cooled systems, alongside Cisco’s AI networking, security and observability technologies. The architecture is designed to support workloads ranging from large-scale inference and agentic AI to trillion-parameter model training. Cisco says the rack-scale approach can also combine liquid-cooled servers with its liquid-cooled networking, addressing the power and thermal challenges of increasingly dense AI systems. 

A key part of the announcement is NVIDIA Cloud Partner (NCP) compliance. Cisco says its architecture brings together Cisco Silicon One-based front-end networking and NVIDIA Spectrum-X-based back-end switching under the Nexus One architecture. This gives customers a validated networking and compute stack rather than requiring them to integrate and test every layer independently.

Cisco is also introducing Cisco Validated Infrastructure Services (CVIS), designed to verify that deployed infrastructure matches the intended reference architecture and performance requirements

As AI infrastructure asks more questions of the system and as compute density rises, networking, cooling, security and operational visibility increasingly determine how effectively organisations can turn expensive AI hardware into usable capacity.

Cisco plans to make the Supermicro-based solutions available through its authorised channel ecosystem late this year.