Vertiv and NVIDIA target earlier AI output with factory-built data centers

Vertiv and NVIDIA introduced an approach to building AI data centers on October 21 that links virtual design checks with factory-built infrastructure. Vertiv projects up to 30% more cumulative AI token output in the first year by getting the same installed capacity working sooner. Polar Data Centers is applying the deployment approach at its Herøya site in Norway.

Resolve the design before equipment reaches the site

The approach, introduced at NVIDIA GTC Berlin, starts with a coordinated infrastructure design rather than separate engineering processes for power, cooling and facility systems. Teams adapt pre-engineered reference designs to customer and site requirements, check them virtually, and carry that configuration into physical deployment.

The software environment combines NVIDIA’s Vera Rubin DSX AI Factory reference design and Omniverse DSX Blueprint with digital representations of Vertiv power and cooling equipment. Dassault Systèmes’ 3DEXPERIENCE platform integrates those elements. Engineers can evaluate requirements, interfaces and constraints in a shared virtual environment before moving to construction.

Vertiv is also contributing SimReady OpenUSD representations of its power and cooling equipment to the SimReady Central Library on Hugging Face. The stated purpose is reuse: engineering teams can find validated equipment assets for their digital facility models instead of preparing each asset themselves, then compare infrastructure options during design.

Once a configuration is validated, more assembly, integration and testing move into the factory. Vertiv says that reduces work and complexity at the customer’s site. Two of its infrastructure offerings illustrate how the design becomes physical equipment:

• OneCore combines critical power, thermal management and supporting infrastructure in a hybrid-built data-center solution.
• SmartRun provides prefabricated overhead infrastructure for high-density power distribution, liquid cooling, networking and containment.

The output gain is a calendar gain

Vertiv’s modeled comparison puts configuration, validation and infrastructure deployment at 48 days combined, against a conventional baseline of approximately 120 days. That brings productive capacity forward by about 10 weeks. These are conditional projections, not measured customer results; Vertiv says actual outcomes will vary by project.

The first-year clock starts at design kickoff, not when the facility opens. Vertiv’s underlying assumptions give the faster build 317 productive days, versus 245 for the conventional schedule, at equal installed capacity. The additional operating days explain the projected increase in cumulative output; the comparison does not depend on making that capacity compute faster.

Polar takes the factory-built approach to Norway

Polar is applying the factory-built deployment model to the first 41 MW phase of its multiphase Herøya campus, which is scalable to 100 MW. Its use of the approach concerns physical infrastructure deployment: shifting assembly, integration and testing off-site to help make high-density, liquid-cooled capacity available sooner.

Vertiv is supplying prefabricated power and thermal modules, SmartRun overhead infrastructure and SmartMod, supported by its Unify software. Polar CEO Andy Hayes described the appeal as a more disciplined, repeatable deployment model that still leaves room to meet customer requirements. He said customers judge Polar on how quickly capacity is ready for them.