Artificial intelligence (AI) data centers are a power play — literally.
The technical backbone to power AI’s computers and machine learning algorithms need significantly more electricity than existing data centers that have fueled the internet age.
A server rack in an existing, conventional data center consumes 5 to 10 kilowatts of electricity compared to AI data center racks, which require 40 to 100 kW, according to Hanwha Data Centers.
A single AI query or prompt uses 10 to 15 times more electricity than a traditional web search.
A data center uses 10 to 50 times more energy per square foot than a commercial office building.
The need for power is in tandem with the need for speed to market with new data centers. Hyperscalers and developers are projected to spend $7.6 trillion on AI data centers through 2031, according to Goldman Sachs.
That “force of capital” is the AI bus driving the search for consistent electricity flows, said Brandon Michalski, chief economist at MOCA Systems Inc., a Boston-based infrastructure technology firm.
“It’s all about available power,” Michalski said.
Data center developers and users are looking at various energy options for planned and proposed AI projects.
Some energy sources are more doable than others, at least for the first phases of the AI wave.
Natural gas is a top power source for new electricity-hungry data centers, according to Parag Nathaney, a Washington-based energy and electricity markets analyst.
Nathaney said right now data center builders’ preference is to be on the grid with close connections to natural gas pipelines. That helps get projects to the AI marketplace quicker.
“Large gas-powered generators are what most data centers are using,” he said.
GE Vernova, Siemens, Mitsubishi, Generac and Caterpillar have seen upswings in orders for generators, turbines and energy equipment orders related to data centers.
Some data center builders are facing supply chain backlogs.
Britt Burt, senior vice president for power industry research at Texas-based Industrial Info Resources, said some data center builders and users are looking at using repurposed aerospace and engines as well as hydrogen fuel cells to power AI’s computers and machines.
Natural gas demand from data centers could increase by close to 20% by 2030, according to an analysis by Christopher Louney, a commodity strategist for RBC Capital Markets.
“U.S. natural gas offers strategic advantages due to insularity from geopolitical volatility, abundant domestic supply, and fully utilized LNG (liquefied natural gas) capacity,” Louney said.
Some data centers are looking to locate near natural gas pipelines with some new connections being built.
Natural gas supplied 40% of energy needed for data centers in the U.S. in 2024, according to a report by the International Energy Agency (IEA).
Renewable energy accounted for 24% followed by nuclear and coal with 20% and 15% respectively.
The agency said natural gas and renewables have the supplies and growth trajectories to offer additional energy for the expected data center surge.
Louney agrees.
"On-site gas generation is being increasingly viewed as a permanent fixture of the digital infrastructure landscape through decade's end at least, and potentially beyond,” he said in his analysis.
The U.S. Energy Information Administration projects data centers growth to help increase U.S. solar generation by 21% in 2026 and 18% in 2027.
U.S. President Donald Trump is a strong advocate for the AI wave and the data centers to power them.
The Trump administration wants to allow data center developers and users to build their own off-meter power plants for their AI projects.
Trump has also eased regulations and encouraged more natural gas and other domestic energy production such as oil and coal. Some coal-fired power plants are being given new life by the data center wave, to the chagrin of those concerns about emissions.
In August, the U.S. Energy Information Administration projected the highest domestic natural gas production since 2016.
The U.S. agency said developers plan to bring 44.9 billion cubic feet per day of new pipeline capacity in the U.S. in 2026 and 2027. Sixty-six percent of that new capacity is in Texas, which is seeing significant data center growth, and Louisiana, which is seeing data center interest.
In Texas, Chevron and Microsoft announced in June a 20-year deal for the energy conglomerate to provide off-meter natural gas for a 2.67 gigawatt power plant and AI data center near Pecos, Texas.
The project is off-grid from the local utility.
Kim Lundgren, CEO of sustainability consulting firm Kim Lundgren Associates (KLA), said the global energy surge raises concerns about emissions linked to climate change.
“This is all coming to the local level,” she said.
While natural gas does not have the emissions of coal-fired power plants, methane leaks from transport and pipelines are a significant driver of climate change, according to the Climate & Clean Air Coalition.
Some developers and users have talked about nuclear energy, via leveraging nuclear plants or installing small micro-reactors, to power AI data centers.
Constellation Energy is pumping $1.6 billion into reopening a reactor at the Three Mile Island Nuclear Generating Station in Pennsylvania as part of a 20-year deal to provide power for Microsoft’s new AI data centers.
NextEra Energy and Alphabet/Google are partnering to reopen the 615-megawatt Duane Arnold Energy Center nuclear plant in Iowa. The facility closed in 2020 after weather damage and is slated to reopen in 2029.
TerraPower, a Bill Gates-backed energy company, is building a nuclear reactor in Wyoming, adjacent to a coal plant, to help offer energy to large users, including data centers.
But nuclear power has its own set of worries over safety, Lundgren said.
Nuclear energy also has high startup costs and long approval processes, according to Arif Gasilov, a partner with the Gasilov Group, an Arizona-based energy and infrastructure consulting firm.
“It just requires a lot of upfront investment,” he said.
Gasilov said he does not expect nuclear energy options to be ready for AI data centers until 2030-2035.
That could result in new data centers relying on natural gas going forward.
“Some might stay on gas,” Gasilov said.
New data centers also face potential supply chain backlogs and delays for energy and technology equipment.
Demand is high for generators, turbines and other equipment.
The U.S. power grid and local energy infrastructure also needs to be ready for a potential doubling or tripling of electricity consumption driven by AI data centers.
“The nation is vastly underbuilt in transmission lines,” Michalski said.
Much of the country’s electricity infrastructure is also older. Seventy percent of transformers and transmission lines are 25 years or older and 60% of circuit breakers are 30 years or older, according to the American Society of Civil Engineers (ASCE) and Georgia-based utility MEAG Power.
On Sept. 24, the Trump administration announced it was dispensing $1.9 billion to utilities across the country for transmission lines and power grid improvements.
Utilities, including ones in Washington state, North Carolina, Tennessee, Florida and Ohio, will put $3.35 billion towards the efforts to rebuild more than 1,500 miles of transmission lines and upgrade another 21,000 miles of lines, according to the U.S. Department of Energy.
Energy use and the potential for higher utility bills for ratepayers are some of the concerns cited by AI data center opponents in their pushes for development moratoriums, as well as codifying cost assurances and protections for other ratepayers.
“Data centers must pay their own way to lower our electricity costs,” said North Carolina Gov. Josh Stein, a Democrat, in a statement, calling for state regulators to codify utilities’ rate pledges.
Will natural gas remain the top power source for AI data centers?
How will the U.S. power grid handle the potential doubling or tripling of electricity consumption?
What are the concerns about emissions linked to climate change from AI data centers?
