Vertiv and NVIDIA unveil factory-integrated AI factory deployment model as Polar begins 41 MW Herøya phase
Редакция ЦОД:Главное
Vertiv and NVIDIA have introduced an AI factory deployment approach that combines virtual validation of pre-engineered architectures with factory assembly and integration of infrastructure. Polar Data Centers is applying the model to the initial 41 MW phase of its multiphase AI data centre in Herøya, Norway, where deployment is underway.
- Polar Data Centers is applying the Vertiv-NVIDIA model to an initial 41 MW phase in Herøya, Norway.
- The Herøya project is described as multiphase and scalable to 100 MW, not as a commissioned 100 MW campus.
- The approach combines virtual validation with factory assembly, integration and testing before site delivery.
- The source claims combined configuration, validation and deployment can be reduced from around 120 days to 48 days.
- Claimed schedule and token-output benefits have not been independently validated at the Polar deployment.
Vertiv and NVIDIA announced a new AI factory deployment approach at NVIDIA GTC Berlin, centred on pre-engineered architectures that can be virtually validated before physical construction and then assembled, integrated and tested in a factory before delivery to site.
Polar Data Centers is the first named operator to apply the model. The deployment is underway at the initial 41 MW phase of its multiphase AI data centre in Herøya, Norway. The Herøya campus is described as scalable to 100 MW; this is future project scalability, not operating or commissioned capacity.
The model is designed to shift a larger share of infrastructure execution away from the site. By combining standardised designs, digital validation and factory integration, it targets a reduction in the time needed to configure, validate and deploy AI data-centre infrastructure.
According to the source's stated model assumptions, infrastructure deployment time can be reduced from approximately 90 days to 45 days. Configuration and validation are claimed to fall from roughly 30 days to three days, while the combined configuration, validation and deployment stages are stated to decline from around 120 days to 48 days. The source further claims that productive AI capacity could become available roughly 10 weeks sooner and that cumulative token output could be up to 30% higher within 12 months of project kick-off, with the same installed IT capacity.
These are claimed or modelled outcomes rather than independently verified operational results from Herøya. For operators developing high-density AI capacity, the significance lies in whether factory-integrated power, cooling and distribution infrastructure can reliably reduce the dependency on sequential on-site engineering and construction. The commercial value of the approach will depend on whether the stated delivery and production effects are repeatable in live deployments.
The approach is intended to compress delivery of dense AI infrastructure by moving more assembly, integration and testing off-site and validating designs before physical construction. If its claimed schedule and production benefits are reproduced in live projects, it could offer operators an alternative to sequential, site-based delivery of power, cooling and infrastructure systems.
- Data Centre Magazine / 23 октября 2026 г. · АнглийскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.