When China-based DeepSeek gained mainstream attention with its low-cost artificial intelligence model earlier this year, some wondered if demand for data center power had been overhyped.
DeepSeek, which develops large language models similar to OpenAI’s ChatGPT, claims that its highly efficient AI technology can effectively reduce computing costs and overall power needs, casting doubt on how much energy infrastructure data centers of the future might require.
But John Medina, senior vice president at Moody’s Ratings, thinks these advancements could fuel more growth, not less.
“What that does is it lowers the cost of computing and potentially increases the number of new companies and new applications that can be created,” said Medina. “It actually could lead to more usage than less usage.”
John Medina
Permission granted by Moody’s Ratings
Mitchell Osborne, director of MEP at Adolfson & Peterson Construction, a Minneapolis-based general contractor, echoed that sentiment.
“While more efficient AI models will reduce the amount of power consumed, we are still in AI and quantum computing’s infancy,” said Osborne. “I do not foresee data centers or power infrastructure construction slowing down.”
Power construction is surging alongside data center projects as contractors rush to keep up with soaring demand, according to industry sources. Tech giants such as Amazon, Microsoft and Meta are rapidly expanding their data center footprints. Meanwhile, a joint venture among OpenAI, Softbank and Oracle will invest $100 billion in artificial intelligence infrastructure, with the potential to scale up to $500 billion, according to a White House press conference.
That shift is fueling a surge in power construction, said Ryan Wobbrock, vice president at Moody’s Ratings. He said utilities and power producers are rapidly increasing expenditures to meet the energy needs of data centers.
“We’re certainly keeping track of that, and we’re seeing the fruits of it really come through in companies’ announcements,” said Wobbrock. “Every time a company has a public statement, it seems to increase its capital budget, and a lot of that is tied to the data center growth.”
New starts and renovations
The surge in new power construction is also driving local grid work as well, said Osborne.
“I absolutely see data center and power project trends continuing,” said Osborne. “In the local market, substations are being specifically built for data centers and stability upgrades are being subsidized by these companies for existing grids.”
Older power plants, once slated for retirement, are being enhanced rather than decommissioned, said Michael Byrne, vice president at Linesight, a Dublin-based construction consultancy firm. That’s largely because new power projects still can’t keep up with the pace of data center expansion in key regions.
From 2014 to 2016, data center energy use reached about 60 terawatt-hours. By 2018, the figure jumped to about 76 terawatt-hours. In 2023, data center energy use hit 176 terawatt-hours, about 4.4% of total U.S. electricity consumption. That will jump to between 325 and 580 terawatt-hours by 2028, according to the Energy Department.
The U.S. power grid was not designed to handle this level of demand growth, said Theodore Paradise, chief policy and grid strategy officer at CTC Global, an Irvine, California-based advanced conductor manufacturer for overhead transmission powerlines.
“The transmission system is the critical enabler of all this. It is the existential piece to whether you can add additional supply,” said Paradise. “Interconnection costs [are] getting bigger. Those might have been $20 million or $40 million for substation work. Now, we’re seeing interconnection costs that can be hundreds of millions of dollars or even over a billion dollars in some cases.”
Yet, that isn’t slowing the pace of data center construction. Some developers are even proceeding without fully secured power contracts, said Byrne. Osborne said that depends on what specific phase of construction the data center or campus is in.
Mitchell Osborne
Permission granted by Adolfson & Peterson Construction
“If it is a single building or an expansion of an older facility, we see a gap in power availability due to permitting and general substation design,” said Osborne. “Hyperscalers and colocation campuses are usually pretty good at pre-planning during the real estate procurement phase and coordinating with local utility providers on the amount of power they need and when to start these projects.”
Tech giants, on the other hand, are putting matters in their own hands. These deep-pocketed companies are increasingly building their own power construction projects, said Medina.
“The new trend is to build a new data center, very large, and to build a power plant right next to it. Build them together,” said Medina. “The hyperscaler will pay for both. They build a gas plant, some solar and some batteries and a big data center, all together. It’s kind of like an island.”
Headwinds on power construction
Skilled labor, a persistent headwind for the construction industry, continues to be a drag on construction activity, said Osborne.
“Due to the ever-increasing amount of data center construction starts, the number of experienced teams that have completed successful projects will become competition in itself,” said Osborne. “Not only are contractors competing for work with developers and clients, but we are now competing for the people to build the number of projects being started. This creates an issue for developers and contractors alike.”
Some sustainability goals could also slow the pace of construction activity, said Byron Sarhangian, partner at Snell & Wilmer, a Phoenix-based law firm. He warned that strict environmental requirements could burden U.S. progress in AI development, especially in relation to other nations.
“When a company says it needs a gigawatt of energy and it’s got to be green, all of a sudden it’s like, ‘Listen, not sure you can get a gigawatt of energy in the first place,’” said Sarhangian. “Frankly, we’re in a race with a number of countries, including China, as it relates to quantum computing and AI. It’s almost like being in the space race and saying, ‘We’re going to go all green.’ Not the time. We need absolutely every piece of energy infrastructure that we can have.”
Looking ahead
The urgency to address these challenges has reached the federal level. The Trump administration recently issued an executive order declaring an energy emergency, aiming to fast-track new power generation and transmission construction.
“That is no small thing, that is a significant executive order,” said Paradise. “What the administration is doing in issuing that is saying, ‘What is happening now is not meeting our energy needs, this is not working.’”
Interest in nuclear energy as a potential solution is growing too, particularly among hyperscalers with aggressive goals. Microsoft, for instance, is experimenting with nuclear energy sources for data centers, but large-scale development remains years away.
“This is still some time before we see a rollout of small modular reactors on data center campuses,” said Byrne. “Nuclear energy will predominantly be used for large-scale campuses, 1 gigawatt or more. And these projects are now only beginning.”
State and federal regulators may also start weighing in on construction expansion, said Osborne.
“As power consumption increases, we may see changes in state and federal regulations to prohibit or encourage data center expansion and usage as well as AI usage,” said Osborne. “This could impact how quickly data centers and infrastructure construction can be started.”
