Vertiv (VRT) says a new approach developed with Nvidia could cut the time required to design and deploy power and cooling infrastructure for an AI data center from about 120 days to 48 days.
The company supplied the electrical and cooling equipment that keeps high-density data centers running. Nvidia’s newest AI system requires large amounts of both, making those systems part of the buildout that has to be completed before expensive GPUs can start processing customer workloads.
Vertiv estimates that bringing the infrastructure online roughly 10 weeks earlier could produce up to 30% more cumulative AI token output during the first 12 months, assuming the same amount of installed computing capacity.
The figure comes from a modeled scenario. Vertiv’s model compares 317 productive days under the new approach with 245 days under a conventional process. Actual results will vary by project.
Vertiv (VRT) shares rose about 4% to around $212.51 in midday trading on Oct. 21 after closing at $203.56 the previous day.
The announcement comes a week before Vertiv reports third-quarter results on Oct. 28.
Vertiv’s model cuts 120 days of work to 48
Vertiv’s estimate breaks the infrastructure process into two stages.
Design, configuration, and virtual testing: about 30 days under a conventional process, versus three days using pre-engineered designs.
Power and cooling deployment: about 90 days conventionally, versus 45 days using factory-built infrastructure.
Together, those two stages fall from about 120 days to 48 days in Vertiv’s model.
Also read: Broadcom fought AT&T over VMware: Now a telecom giant is all in
The first step happens before equipment reaches a construction site.
Instead of designing the power and cooling systems for each AI facility from scratch, engineers can start with reference designs based on Nvidia’s Vera Rubin DSX AI Factory architecture. The proposed setup can then be tested in a digital model before construction begins.
The system also uses technology from the French industrial software company Dassault Systèmes to create a virtual model of the facility, allowing engineers to test how computing, power, and cooling systems work together before construction begins.
After the design is approved, Vertiv will shift more assembly and testing into a factory. Larger sections of the power and cooling infrastructure can arrive at the data center site preassembled rather than being assembled piece by piece during construction.
This is important for AI operators. GPUs generate revenue only after the facility has sufficient power, cooling, and networking to run customer workloads. Shortening the infrastructure schedule gives operators more days to use the computing equipment they have already paid for.
Vertiv CEO Giordano Albertazzi has described the goal as reducing the time from a project’s requirements to productive AI capacity.
The company has been building its relationship with Nvidia around the same constraint. Vertiv has developed power and cooling designs around Nvidia’s newer systems, including infrastructure capable of handling increasingly power-dense server racks.
Vertiv’s new approach could help cut time required to design and deploy power and cooling infrastructure.primeimages / Getty Images
A 41-megawatt Norway project is applying the approach
Data-center operator Polar Data Centers is using Vertiv’s factory-built deployment approach for the first 41-megawatt phase of an AI data center in Herøya, Norway.
The site is designed to scale to 100 megawatts. Vertiv will provide prefabricated power and thermal modules along with overhead infrastructure for power distribution, liquid cooling, and networking.
The project gives Vertiv a real site where the deployment approach can be used, but neither Vertiv nor Polar has reported that the Norway facility has already achieved the 10-week time saving or 30% token-output increase in Vertiv’s model.
How much time can actually be saved also depends on problems outside the equipment supplier’s control.
A data center can have preassembled cooling and electrical equipment ready to install and still wait for enough electricity to reach the property.
That problem hit Australian data center operator NextDC on Oct. 21.
NextDC shares plunged 41.7% after delays to electricity-network upgrades at its M3 Melbourne facility forced about A$140 million in contracted customer revenue to be pushed into a later period. The company said the next increase in grid supply at the site is now expected in 2028.
Microsoft ran into an even longer delay in Britain.
The company pulled out of an agreement to anchor an Nscale data center in Loughton, England, after the site’s expected power connection was pushed into the 2030s, Reuters reported. The planned facility was supposed to house 23,040 Nvidia GB300 GPUs and provide 50 megawatts of AI computing capacity.
Those cases involve grid connections that Vertiv’s factory-built deployment model cannot remove on its own. They also show why data-center companies are spending more money on ways to secure electricity sooner.
Vertiv is moving closer to the data center’s power source
Vertiv agreed in September to acquire UtilityInnovation Group for about $1.45 billion in upfront cash, adding technology designed to help data centers source and manage power before it reaches server racks.
UtilityInnovation Group builds microgrid controls, coordinates on-site generation, and designs so-called behind-the-meter power systems. Those systems can enable a data center to use local generation or energy storage when the utility grid cannot quickly provide enough electricity.
More Artificial Intelligence:
AI’s biggest business opportunity may not look like AI
AI racks are outgrowing old power: 2 chipmakers have a strategy
BofA splits software into winners and losers as AI agents arrive
The acquisition pushes Vertiv further upstream from the cooling systems and power-distribution equipment it already sells inside data centers.
Its existing equipment helps move electricity from the point where power enters a facility down to individual server racks. UtilityInnovation Group would add tools for deciding where that electricity comes from and how on-site generation can fill gaps when the grid is constrained.
Vertiv has also been expanding production capacity for equipment used in high-density AI facilities and developing systems around newer Nvidia rack designs. The company describes the broader strategy as connecting infrastructure from the power source down to the chip.
Vertiv’s financial growth already reflects the AI infrastructure buildout. Second-quarter revenue rose 24% from a year earlier to $3.27 billion, and management raised its full-year outlook.
The company’s next update comes quickly. Vertiv is scheduled to report third-quarter earnings before the market opens on Oct. 28, followed by an 11 a.m. ET conference call.
Polar’s 41-megawatt Norway project will provide an early look at how Vertiv’s factory-built approach works outside a company model. The remaining question is how much of the proposed 10-week savings survives once construction, permitting, and access to electricity are factored into the schedule.
Related: Goldman Sachs sends a stern warning to stock market investors