Summary
- Polar is using Vertiv's factory-built infrastructure approach for the first 41MW phase of its Herøya data centre, with the wider campus potentially reaching 100MW.
- Vertiv and NVIDIA have combined pre-engineered reference designs, virtual validation, and prefabricated electrical and thermal equipment to reduce site integration work.
- A modelled comparison suggests 120 days of design and infrastructure deployment could fall to 48, potentially increasing first-year cumulative token output by up to 30%; actual results remain unverified.
Polar Data Centers is applying factory-built electrical and cooling infrastructure supplied by Vertiv to the first 41MW phase of its AI data centre at Herøya, Norway, as the equipment manufacturer extends its collaboration with NVIDIA on integrated computing facilities.
The Norwegian development is intended to expand to as much as 100MW across multiple phases. Vertiv’s equipment scope includes prefabricated power and thermal modules, overhead infrastructure, and modular systems designed to support high-density computing.
Vertiv confirmed the arrangement on 21 October at NVIDIA GTC Berlin, introducing an approach that combines pre-engineered reference designs, virtual configuration and validation, and factory assembly of infrastructure before delivery to a data centre site.
Polar’s involvement provides a named European application for the approach. However, the announcement does not establish that the Herøya first phase has been commissioned, that the entire 41MW is available to customers, or that the planned 100MW campus has secured every milestone required for operation.
The first phase represents the scale of the infrastructure development to which the factory-built model is being applied. The company has not provided a detailed reconciliation between the reported megawatt figure, the intended IT load, and the total facility electricity connection.
Vertiv is supplying prefabricated power and thermal modules alongside its SmartRun overhead infrastructure and SmartMod modular systems, supported by Vertiv Unify software.
SmartRun is designed to accommodate electrical distribution, liquid cooling infrastructure, networking, and containment within an integrated overhead arrangement. Combining those functions can reduce the number of separate installation activities required after equipment reaches the destination.
The broader factory integration approach includes Vertiv OneCore, a modular infrastructure system combining electrical and thermal equipment. However, Vertiv has not identified OneCore as part of the specific equipment package being supplied to Herøya, so its general capabilities should not be confused with a confirmed installation at Polar’s site.
Moving assembly into a manufacturing environment changes the sequence of infrastructure delivery. Equipment can be integrated and checked before shipment while civil engineering and external utility work continue at the destination.
Electrical modules can undergo selected inspections and functional tests, while cooling assemblies can be checked for the integrity of their connections and interfaces. Such work offers an opportunity to identify integration problems before equipment arrives on site.
The completed installation nevertheless requires final connections, site testing, and commissioning. Factory validation cannot establish that external electrical infrastructure, coolant distribution, and heat rejection equipment will perform correctly under the conditions found at Herøya.
Vertiv and NVIDIA are also developing a shared approach to infrastructure design using NVIDIA’s Vera Rubin DSX AI Factory reference architecture and Omniverse DSX Blueprint.
These provide a framework for modelling computing systems and the power, cooling, networking, and other services needed to support them. The intention is to coordinate equipment requirements before manufacturers begin producing the final infrastructure.
Vertiv is integrating its electrical and thermal equipment models with Dassault Systèmes’ 3DEXPERIENCE platform, allowing design teams to examine the interaction between components inside a virtual representation of the proposed facility.
The company is also contributing SimReady OpenUSD models of its equipment to the SimReady Central Library, enabling engineering teams to reuse digital representations when evaluating infrastructure configurations.
Virtual validation can identify clashes between equipment arrangements, service routes, and design requirements before installation. Its usefulness depends on the accuracy of the underlying models and their correspondence with the products eventually manufactured.
Electrical ratings, cooling temperatures, available space, and installation constraints still require confirmation for the actual project. A successful simulation does not replace physical testing of the completed infrastructure.
Vertiv has published a modelled comparison of the time needed to design and deploy an integrated AI computing installation. Under its assumptions, configuring and validating an initial design falls from approximately 30 days to three days.
The same comparison estimates that factory-built infrastructure could reduce on-site electrical and cooling deployment from 90 days to 45 days. Together, those two stages fall from 120 days to 48 days.
The reduction amounts to 72 days across the activities included in the model, allowing computing equipment to become productive earlier if the surrounding construction and infrastructure programme is ready.
Vertiv calculates the commercial effect over the first 12 months from design commencement. Under the conventional timetable, it assumes 245 productive days, compared with 317 days for the accelerated approach, at the same installed computing capacity.
That difference underpins the company’s estimate of up to 30% more cumulative AI token output during the first year. The claimed increase results from a longer period available for productive computing, rather than improved processor performance or a demonstrated reduction in electricity consumption per token.
All of these comparisons are modelled. Vertiv states that results vary by project, and it has not published measured evidence showing that the Herøya installation has achieved the predicted design, construction, or computing output improvements.
The comparison also excludes several development activities that can determine when a data centre begins operating. Land preparation, planning permission, grid connection works, external power infrastructure, and construction of supporting facilities may involve separate programmes.
Even when prefabricated modules are ready for delivery, unavailable electrical capacity or incomplete site infrastructure can prevent installation and commissioning. The model therefore describes potential acceleration of specified engineering stages rather than the entire period from site acquisition to commercial operation.
The approach also introduces logistical requirements associated with transporting integrated infrastructure. Modules must be designed for lifting, handling, and installation while protecting the electrical and mechanical equipment assembled during manufacturing.
After delivery, project teams must verify that connections and equipment have not been damaged in transit and that the systems meet the final installation requirements. Maintenance access, component replacement, electrical isolation, and cooling system serviceability remain relevant throughout the operating life of the facility.
At Herøya, these systems will support infrastructure intended for liquid-cooled computing. High-density equipment transfers substantial amounts of heat into coolant circuits, which must be connected to facility equipment capable of controlling temperatures and rejecting or recovering the resulting energy.
The electrical installation must similarly accommodate the intended computing load and the electricity required by cooling, distribution, monitoring, and other supporting systems. Vertiv has not disclosed a complete engineering specification or measured efficiency target for the finished Herøya phase.
The project is separate from Polar’s Tørdal campus, where Crusoe has confirmed 12MW of operating AI capacity and plans a further 40MW of liquid-cooled infrastructure. Herøya and Tørdal involve different sites, customers, and development programmes, and their reported capacities must not be treated as interchangeable.
Vertiv’s Herøya arrangement is also distinct from NVIDIA’s wider DSX POD reference design and its separate announcements about dry cooling, workload power allocation, and grid-responsive computing. The Norwegian development concerns Polar’s use of a specified manufacturer’s prefabricated electrical and thermal systems.
The next measurable milestones will be delivery and installation of the infrastructure modules, completion of external electrical and cooling connections, and confirmation of the capacity commissioned for customer use.
Until operating results become available, Herøya provides a defined deployment programme for Vertiv’s factory integration model, while the proposed time savings and additional first-year computing output remain supplier estimates rather than demonstrated project performance.
