Vertiv Packages AI Rack Power and Cooling for Existing Facilities
Vertiv introduced its SmartIT infrastructure platform on October 26, 2026, offering a pre-engineered way to install high-density AI computing in settings ranging from existing enterprise facilities to new data centers. The announcement puts a practical problem in focus: buying accelerators is only one part of putting them to work.
The power and cooling around a rack have to be ready for the equipment that arrives. A facility designed for conventional servers may have spare floor space while lacking the electrical capacity, heat removal or monitoring needed for a dense AI installation. Those constraints can turn an apparently simple expansion into a construction and engineering project.
SmartIT combines power, cooling and monitoring in a design based on NVIDIA MGX and Open Compute Project ORV3 architectures. Vertiv says the platform is intended for deployments in existing enterprise facilities, new builds, commercial spaces, colocation sites and hyperscale environments. Its SmartRun system extends integration across a row, while rack power distribution units and the Unify monitoring and control system form other parts of the offering.
The rack becomes a system
For a buyer, the attraction of a pre-engineered design is the possibility of reducing the work needed to bring separate components together. A rack, its power distribution and its cooling must work as a system. If the interfaces have already been designed and tested together, there may be fewer decisions to resolve during installation.
That does not make every building ready for high-density computing. A standard rack still depends on power delivered to the site, suitable space, heat rejection and a network connection. An enterprise considering an installation needs to check its own electrical and mechanical capacity before assuming that a packaged platform will fit. Local permits and utility connections can set the schedule, regardless of how quickly the equipment itself can be assembled.
The monitoring layer matters as well. Vertiv says Unify brings together analytics and controls across power and thermal infrastructure, including leak detection and coordinated response. For operators moving into liquid-cooled environments, knowing where a fault occurs and what should happen next is a basic operational question, not an optional dashboard feature.
A smaller starting point
Vertiv showed a demonstration unit at NVIDIA GTC Berlin designed to support NVIDIA DGX Rubin NVL8 deployments. The company presents SmartIT as a way to start at an appropriate scale and expand using a common architecture. That could appeal to research institutions, enterprises and colocation providers that want AI capacity without committing at the outset to a campus-sized project.
The commercial test is more demanding than the announcement. Buyers will want to know the delivered cost for their configuration, the cooling and power available at the site, installation time, service arrangements and what happens when a component needs replacement. A validated design can reduce integration work; it cannot establish those answers for every building.
There is also a question of flexibility. Common architectures can make expansion more predictable, but customers should ask which components can be changed later, which generations of hardware are supported and how much of the operating stack is tied to a particular vendor. A platform that is easy to buy today should also be manageable when computing hardware changes.
The wider signal is that AI infrastructure is being sold in increasingly complete packages. This reflects the difficulty of deploying dense computing outside facilities built for it from the beginning. The opportunity for infrastructure suppliers is substantial, though the hard limit remains physical: electricity has to arrive, and heat has to leave.