Vertiv, NVIDIA target 30% more AI token output with pre-engineered factory model

The announcement is aimed at a bottleneck that appears after compute is secured: the sequential engineering and on-site integration of power, cooling and facility systems, which the two companies describe as creating complexity and potential delays. Their answer is to start from pre-engineered AI factory reference architectures that are configured and virtually validated in a shared digital environment before any equipment reaches the site.

The modeled numbers split into two stages. Configuration and validation fall from roughly 30 days to three days, a 10x improvement, while factory-built deployment reduces the on-site infrastructure period from about 90 days to 45 days, a reduction of up to 50%. Combined, the two stages drop from approximately 120 days to 48 days, which the companies say can make productive AI capacity available about 10 weeks sooner.

That time saving is the mechanism behind the token-output claim. Vertiv CEO Giordano Albertazzi framed the stakes around tokens, saying every day spent configuring or integrating infrastructure is output that never comes back. The approach draws on NVIDIA's Vera Rubin DSX AI Factory reference design and Omniverse DSX Blueprint, Vertiv SimReady power and cooling assets, and Dassault Systèmes' 3DEXPERIENCE agentic platform, with Vertiv also contributing SimReady OpenUSD assets to a library on Hugging Face.