• Experts say developers must learn from renewable and utility markets to keep pace with demand
• Better coordination between compute and power communities is also a necessity

Data center energy infrastructure experts say to speed up energization of AI data center projects, developers need to deploy utility-grade modeling.

Following the announcement of their newly formed partnership, Devin Dilley, EPC Power co-president and founder, and Hari Sivadas, Generac Power Systems senior vice president and general manager, told NPM the booming data center development sector must quickly learn from solar, storage and utility markets to keep up with the complexity and tight timelines of gigawatt-scale projects.

Dilley said the sector is still “new to engineering full power systems” at this scale, whereas solar and storage are more mature.

“The ecosystem of companies that build these data centers are having to learn how to absorb a decade’s worth of experience from these adjacent sectors so that they can build power systems,” he explained.

That means shifting from trial-and-error power design to utility-grade modeling and simulation when planning AI data center power systems, especially for off grid or complex sites.

“You design it up front, model it well, and then when you build it, it will work exactly how you engineered it to work,” Dilley said. “That [mindset] needs to be embraced more by this sector.”

Sivadas agreed, further adding that better coordination between compute and power communities is also needed.

Too often, he said, each side assumes the other will solve the emerging grid and power challenges – either through algorithm improvements or infrastructure upgrades.

“[We need] tighter coordination between the folks building data centers, and that’s why you see some of the NVIDIA initiatives of bringing more of that industrial ecosystem together to say, ‘This is an us problem, not just you versus somebody else…it’s a collective problem that needs to get solved,’” Sivadas said.

NVIDIA led initiatives aim to bring data center developers, infrastructure providers and technology vendors into a more integrated AI data center ecosystem.

‘Everything in the toolkit’

Improved coordination can help speed up project timelines, which affects a project’s bottom line – projects that sit idle waiting on power lose compute revenue.

Sivadas said the industry is used to viewing projects mainly through power equipment and EPC costs but argued that power is actually a smaller share of the overall cost.

“The time to power, by orders of magnitude, outweighs anything else in terms of their ability to monetize the AI, algorithms and software faster,” he said.

Thus, off-grid and behind the meter solutions were born to deploy necessary power faster than traditional utility upgrades.

That’s the gap Generac and EPC Power are targeting through their partnership, combining reciprocating engines, batteries and inverters in systems that can come online faster than new grid capacity.

EPC’s inverter-based resources have a valuable role to play to “placate the utilities and grid operators, to make sure that data centers are good citizens,” Sivadas said, by helping those facilities ride through grid glitches, smooth out volatile AI loads and support the grid when it’s under stress.

With a national grid infrastructure system not ready or capable of supporting the rapidly growing power needs of the current data center industry, “off-grid data centers will be, by necessity, constructed to a substantial degree,” Dilley said.

“Those off grid data centers, and whether they’re connected in the future or not, doesn’t change the near-term reality that that power is required in fully islanded systems. … There are a whole host of energy sources that can be used, but everything in the toolkit needs to be used in order to accomplish this,” he said.