AI is redefining power. As a result of its rapid adoption and ever-increasing energy demands, we're entering the gigawatt era where the need isn't just for more compute, but different compute. Driving high densities and irregular power spikes, AI is now reshaping everything from facility design to power procurement. The AI evolution has fast transformed into a full-blown AI revolution.

Power, cooling, and compute must now be treated as one integrated unit – designed and managed together to support scalability and unprecedented power demands. The way facilities are designed, manufactured, delivered, and operated must adapt to support the proliferation of AI – and unlock its full potential.

Already supporting some of the globe’s largest customers, Vertiv is well-positioned to help deliver this transformation. The company is innovating to power the next generation of infrastructure technologies, enabling its solutions to remain agile and scalable for the demands of today, tomorrow and beyond.

During Vertiv’s Power Innovation Day, representatives from the company and several key partners laid out their vision for the future of data center power. Together, they explored how Vertiv manages the complex power ecosystem of AI-ready data centers and how real-world testing is critical to scaling effectively in this era of rapid and radical change.

The gigawatt era

AI workloads are exceptionally high-density, often with unpredictable, spiking energy demands above and beyond what legacy data centers and aging grid infrastructure were initially built to support. The need to deliver dramatically more power within the same footprint is driving a complete rethink of infrastructure, starting at the server.

“Over the last 18 years, there have been discussions around growing the rack to the next five, and then the next 10kW,” says Giovanni Zanei, vice president of large power conversion at Vertiv. “But now we’re talking about doubling, or even getting ten times more in the next few years. So, racks are going to be consuming a lot more energy, which will have a significant impact on the rest of the power infrastructure.”

The entire structure, from utility connections to backup generators and UPS systems, must be equipped to handle rapid, erratic load spikes, driven by high-density AI applications and machine learning. These fluctuations require a more robust and flexible infrastructure.

“This is not necessarily a challenge, but something that the power infrastructure needs to cope with and needs to be validated with. It must be understood,” says Zanei. Alongside this validation, sourcing reliable, carbon-efficient, and scalable backup power is becoming just as critical. As the industry continues to grow, so does the onus on sustainability, and operators must be able to meet energy-hungry AI demands while staying within environmental limits.

To better manage these challenges and bolster a deeper understanding of this newfound energy landscape, predictability remains a central goal. Vertiv’s industry-first AI simulator – a finalist for the Best AI Innovation at the Data Cloud Awards – provides clarity on how these loads impact upstream systems. Arturo Di Filippi, offering director of large power at Vertiv, explains:

“This is not a model, it's a real tool. It's a piece of hardware that we use to simulate analog profiles and to verify the effect of these AI loads on the equipment upstream. For instance, what is the impact on the grid, on the generator, or even the impact on batteries?”

Vertiv’s solution architects run detailed simulations to examine how shifts in density, power profiles, and cooling mechanisms affect infrastructure. This work is essential in helping customers scale effectively in a fast-changing landscape.

Crucially, Vertiv solutions are modular and scalable by design, meaning whether you're building a new greenfield data center or upgrading existing infrastructure, facilities are afforded the flexibility to adapt to changing applications and capacity demands alongside business needs. This is particularly advantageous for brownfield sites, as it mitigates the need to start from scratch, as Filippi explains:

"You get more power in a site, but you also need the equipment in order to process that power. Our infrastructure solutions provide additional processing capacity to the load, without requiring a complete restructuring of the data center.”

A power first mindset

The race to adopt and deploy AI workloads is fundamentally altering the philosophy of data center design. What used to be a real estate-first approach has evolved into a power-first mindset. Getting power to the site, and then delivering it efficiently to the rack, is now the dominant concern.

“As a data center operator in the past, the first question would be, what land can we build on?” says Phillip Marangella, chief marketing and product officer at EdgeConneX. “But today, it’s about where we can get power at scale, and where we can get carbon-efficient power.”

Greenfield builds are gaining favor due to flexibility in on-site generation and fewer regulatory hurdles, especially in the US. “It’s much quicker to get permits for gas turbines on site in America than in Europe,” notes Filippi. “That’s why the AI boom is largely happening in the US at the moment.”

This boom has caused some existing markets to become oversaturated, expanding the global data center footprint into new regions. “We’re seeing Malaysia and Indonesia pop up, for instance, and in the US, it's extending south and to the north,” says Marangella.

But location is only one piece of the puzzle. Tax incentives, land prices, local regulations, and labor availability also play significant roles in determining viable sites. Capital costs are also higher now due to dense compute and specialized infrastructure – further emphasizing the need for innovation.

Vertiv is already exploring alternative energy sources like hydrogen and solar to support future power demands. “We developed a proof of concept in Ohio, with 400kW of fuel cells connected through UPS and batteries,” says Filippi. “We’re using real hydrogen, and we’re learning how challenging the hydrogen supply chain still is today.”

Vertiv is also championing a grid-friendly mindset, encouraging data centers to only use power when needed and even feed power back into the grid when possible. In Sweden, the company partnered with Conapto to install UPS grid support that actively manages the absorption and injection of power based on grid balance.

“Being good to the grid will help enable more renewables and more stability,” says Filippi. “We have to work collaboratively across the industry to perfect that mindset shift.”

Alternative backup solutions like battery storage and gas turbines are also gaining traction, though diesel generators still dominate due to reliability. “We have proved that it is possible to connect fuel cells to UPS as an alternative,” Filippi adds, “but run time expectations are still a challenge.”

Delivering more power to the rack

“There are two ways to deliver more power to the rack,” says Zanei. “One is to deliver more current. The other is to increase voltage. Physics tells us that increasing voltage is more efficient.”

However, implementing this in data centers requires rethinking every part of the power infrastructure – from the grid to the gray space, to the IT room. The number of power conversion and transformation steps must be reduced, and inefficiencies must be eliminated.

That’s why Vertiv works closely with partners and engineering teams to make sure newly built data centers are future-ready. “Every new chip that comes out represents a step function in density,” adds Zanei. “Flexibility to adapt is key.”

Cooling these ultra-high-density racks is another critical piece of the puzzle. “Air was easier and safer. Water was taboo. But now it’s integral,” says Marangella. “These racks are now multi-million dollar assets, and you can’t risk damage. So operationally, everything must be rethought.”

Stefan Nilsson, chief commercial officer at Conapto, adds, “AI drives energy consumption, but also increases return heat. We can reuse that heat more efficiently with liquid cooling. It’s about squeezing more kilowatts per square meter, increasing return temperature, and saving money. That’s thanks to liquid cooling.”

Speed and standardization for the new age of data centers

As data center components become increasingly interdependent, the lack of clear standards, especially around liquid cooling, is a growing issue. Operators must support varying customer requirements without a uniform baseline to guide the way.

“Delta T, flow rates, and filtration vary so significantly,” says Marangella. “That’s why partnerships with Vertiv, Nvidia, and others are so critical to finding a common design language.”

Hybrid cooling systems are becoming more common, as facilities frequently need to support both Cloud and AI workloads. This increases the need for robust safety, operational, and training procedures. “If there’s a leak, is it a drip or a gush? Who’s responsible? Where’s the demarcation point?” notes Marangella. “We need clarity and training.”

High power capacity is no longer enough. It must be delivered fast. “Speed is essential – from production of UPS systems to delivery of PDUs,” says Zanei. “Customers want the same experience globally.”

Vertiv uses modular design and manufacturing to deliver standardization and flexibility at scale. Pre-validated systems can be assembled off-site and delivered rapidly, streamlining deployment while allowing customized configurations.

“This modular approach is a major shift,” says Nilsson. “You can grow with your customers, adding capacity at the same time as their needs scale.”

What the next generation of data centers look like

“I don't think there will be just one model for a next-generation AI-ready data center,” says Zanei. “We’re in a moment of rapid change. But I do expect certain design standards to emerge – especially around power distribution and minimizing transformation steps.”

While the architecture of the future remains fluid, what’s certain is that innovation will continue to redefine every part of the data center. Products, systems, and infrastructure are evolving now to meet the needs of tomorrow.

“The really interesting era will begin when we start using AI language models in enterprise applications, so that enterprise companies can utilize all of that calculating power that has been developed over time in business models and in strategy models, and actually make an impact with it,” says Nilsson. “That’s the next phase – the inference era.”

The rise of AI has triggered a chain reaction across the industry, catalyzing both challenges and opportunities. To meet these demands, a shift in mindset is essential – balancing speed to market with long-term flexibility and sustainability.

Vertiv is committed to powering this next era, with solutions that are helping shape the future of AI-ready data centers by delivering what matters most – information, insight, and infrastructure that scales.

For more information please visit here, or, book a tour or a testing at our Global Power Customer Experience Centers.