Vertiv, Nvidia target greater AI token output

Vertiv and Nvidia have introduced an AI factory approach that combines rapid configuration and virtual validation of pre-engineered reference architectures with factory-built deployment.

The approach is designed to move AI factories from project kickoff to productive capacity faster and, under defined assumptions, generate up to 30% more cumulative token output within the first 12 months with the same installed capacity.

Polar Data Centers is applying the factory-built deployment approach to the first 41 MW phase of its multiphase AI data centre at Herøya, Norway.

For AI factory operators, securing compute is only part of the challenge. Traditional infrastructure development can require power, cooling and facility systems to be engineered and integrated through separate, sequential processes, creating complexity and potential delays.

The Vertiv and Nvidia approach starts instead with pre-engineered AI factory reference architectures that can be rapidly configured to customer and site requirements, virtually validated before deployment, and then translated into factory-built infrastructure.

Teams can start with pre-engineered AI factory reference designs and rapidly configure them for specific compute, power, cooling and facility requirements in a shared virtual environment. 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.

Rather than developing an integrated infrastructure design from scratch, teams can configure and virtually validate proven reference designs against project requirements, interfaces and constraints. In the modelled scenario, this reduces the time from requirements to a configured and validated AI factory solution from approximately 30 days to three days, a 10x improvement.

Vertiv is also contributing SimReady OpenUSD assets of its power and cooling equipment to SimReady Central Library on Hugging Face. The library helps engineering teams discover and reuse validated equipment assets in AI factory digital twins, reducing asset preparation and helping teams evaluate infrastructure options during design.

Once the AI factory configuration is validated, Vertiv prefabricated solutions shift more assembly, integration and testing into the factory before equipment reaches the site.

These solutions include Vertiv OneCore, a hybrid-built data centre solution that integrates critical power, thermal management and supporting infrastructure, and Vertiv SmartRun, a prefabricated overhead infrastructure system for high-density power distribution, liquid cooling, networking and containment.

By moving more infrastructure work off-site, teams can reduce field complexity and on-site integration requirements. In the modelled scenario, the factory-built approach reduces the infrastructure deployment period from approximately 90 days to 45 days, a reduction of up to 50%.

Combined with the faster configuration and validation enabled by pre-engineered reference designs, the modelled approach reduces these stages from approximately 120 days to 48 days. That can make productive AI capacity available about 10 weeks sooner, generating up to 30% more cumulative token output within 12 months of project kickoff with the same installed capacity.

At its Herøya campus in Norway, Polar is using Vertiv’s factory-built deployment model for the first 41 MW phase of its multiphase AI data centre, scalable to 100 MW.

By shifting more infrastructure assembly, integration and testing into the factory before equipment reaches the site, the deployment model is designed to reduce on-site complexity and help make high-density, liquid-cooled capacity available to customers sooner. Vertiv is supplying prefabricated power and thermal modules, Vertiv SmartRun overhead infrastructure and Vertiv SmartMod, supported by Vertiv™ 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,” says Giordano Albertazzi, CEO of Vertiv. “Success depends on how quickly operators can move from requirements to productive AI capacity.

“By starting with pre-engineered reference designs that can be rapidly configured and validated, then extending that approach through factory-built deployment, we help customers reduce complexity, deploy infrastructure faster and bring AI capacity online sooner. Polar Data Centers is demonstrating how that approach can be applied at scale in a real-world AI factory deployment.”

Vladimir Troy, vice-president of AI infrastructure at Nvidia, comments: “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.

“With the Nvidia Omniverse DSX Blueprint, companies such as Vertiv and Dassault Systèmes can simulate these systems together and carry an integrated design into physical deployment, helping operators resolve conflicts earlier and bring productive AI capacity online sooner.”

Stéphane Sireau, vice-president: high-tech industry at Dassault Systèmes, adds: “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.Integration issues get resolved before construction, not on site, helping bring token-producing capacity online sooner.”

Andy Hayes, CEO of Polar Data Centers, says: “Our customers measure us on how quickly capacity is ready for them. At Herøya, Vertiv’s factory-built infrastructure helps us manage the complexity of deploying power, cooling and white-space systems for high-density computing. It gives us a more disciplined, repeatable deployment model while preserving the flexibility to meet the requirements of our customers.”

The post Vertiv, Nvidia target greater AI token output appeared first on IT-Online.