Vertiv and NVIDIA have introduced an AI factory design and deployment approach combining pre-engineered reference designs, virtual validation and factory-built infrastructure. Polar Data Centers is applying the approach to a 41 MW development in Norway.
At NVIDIA GTC Berlin, Vertiv and NVIDIA introduced an approach to AI factory design and deployment that combines rapid configuration and virtual validation of pre-engineered reference architectures with factory-built infrastructure. The companies say the approach can reduce the time between project planning and productive computing capacity.
Polar Data Centres is applying the factory-built deployment approach to the first 41 MW phase of its multiphase AI data centre at Herøya, Norway. The development is designed to scale to 100 MW.
For AI factory operators, securing computing hardware is only part of the challenge. Power, cooling and facility systems are often engineered and integrated through separate processes. The proposed approach starts with pre-engineered AI factory reference architectures that can be configured to site requirements, virtually validated and translated into factory-built infrastructure.
Configure and validate
The approach combines the NVIDIA Vera Rubin DSX AI Factory reference design and NVIDIA Omniverse DSX Blueprint with Vertiv SimReady power and cooling assets, integrated with Dassault Systèmes’ 3DEXPERIENCE agentic platform. Vertiv is also contributing SimReady OpenUSD assets of its power and cooling equipment to the SimReady Central Library on Hugging Face.
In the companies’ modelled scenario, configuration and validation time falls from approximately 30 days to three days, a tenfold improvement. These figures describe a modelled comparison and will vary by project.
Factory-built infrastructure
Vertiv’s prefabricated solutions shift more assembly, integration and testing into the factory. These include Vertiv OneCore, integrating critical power, thermal management and supporting infrastructure, and Vertiv SmartRun, a prefabricated overhead system for high-density power distribution, liquid cooling, networking and containment.
In the modelled scenario, the infrastructure deployment period falls from approximately 90 days to 45 days. Combined with the shorter design stage, the two stages fall from approximately 120 to 48 days. Vertiv estimates that this could produce up to 30% more cumulative AI token output in the first 12 months, at equal installed capacity, by bringing computing capacity online earlier.
Polar deployment
At Herøya, Vertiv is supplying prefabricated power and thermal modules, SmartRun overhead infrastructure and SmartMod, supported by Unify software.
“An AI factory earns its return in tokens, and every day spent configuring infrastructure or integrating it on site is output that never comes back,” said Giordano Albertazzi, CEO of Vertiv. “Success depends on how quickly operators can move from requirements to productive AI capacity.”
“AI factories must be designed as one system, and the NVIDIA Vera Rubin DSX AI Factory reference design helps teams align compute, power and cooling requirements from the outset,” said Vladimir Troy, vice president of AI infrastructure at NVIDIA.
“With Vertiv and NVIDIA, we’re accelerating AI factory design and deployment by engineering the factory as one system – compute, networking, power and cooling co-designed and co-simulated in a virtual twin,” said Stéphane Sireau, vice president, high-tech industry, at Dassault Systèmes.
“Our customers measure us on how quickly capacity is ready for them,” said Andy Hayes, CEO of Polar Data Centres. “At Herøya, Vertiv’s factory-built infrastructure helps us manage the complexity of deploying power, cooling and white-space systems for high-density computing.”