AI models and the data centers filled with chips to power them require an enormous amount of power. And the industry is scrambling to keep pace.
After a decade of largely flat electricity needs, somewhere around 50 gigawatts of new power capacity — or enough to run roughly 40 million homes — will be required in the US to sustain the AI boom, according to Goldman Sachs.
But should some of the marquee deals Big Tech giants are signing not materialize, utility companies and their customers could be stuck footing the bill.
“Many of the [grid connection requests] appear to be from a developer that is proposing data centers in multiple utility service territories looking for, ‘Where can I connect the fastest? Where can I get a deal?'” said Brendan Pierpont, director of electricity modeling at the research firm Energy Innovation.
But the speed-at-all-costs approach has its risks. “What are the long-term business models? How much compute will those services actually require? [There’s] just huge amounts of uncertainty about that whole space,” Pierpont said.
The process for turning power demand into power generation takes years.
When a utility receives a power load request, such as from a tech company looking for, say, 2 gigawatts for a new data center, the utility spends millions buying the equipment, materials, and hiring the personnel to make it happen.
Should demand ultimately fall short of estimates, utilities can be stuck with stranded assets generating no revenue. Their options then are to find a way to pass that cost — which averages around $102 per kilowatt, or $102 million for a 1-gigawatt load — on to ratepayers or write down the loss themselves.
AI has pushed a bevy of tech companies into deals with utilities across the country, but some cracks have started to show in these best-laid plans as the AI boom rapidly evolves.
Microsoft (MSFT), one of the largest data center developers in the country, decided in March to walk away from proposed data center projects in the US and Europe with a combined 2-gigawatt load, according to Bloomberg.
While it is unclear whether utilities had begun to spend money building out connections for those projects, it is evidence of the potential threat that TD Cowen analysts attributed to an oversupply of the computers that power AI technology.
This past week, Monitoring Analytics, the independent market monitor for PJM Interconnection, filed a brief with the Federal Energy Regulatory Commission arguing that the federal regulator should reject a recently signed transmission agreement between Pennsylvania utility PECO Energy and Amazon’s (AMZN) Data Services division.
PJM is the largest electricity transmission operator in the country, serving more than 65 million people across 13 East Coast and mid-Atlantic states.
Monitoring Analytics’ brief argues that unless the operator can prove the massive load request from Amazon won’t impact reliability and cost for PJM ratepayers, the transmission deal should not be allowed to go through.
While the long-term demand picture from AI data centers might be murky, what’s clear is the pressure these projects are putting on the US electrical grid today.
The increased energy load of data center development is already showing up in Americans’ electric bills.
The average utility payment for electricity and gas rose by 3.6% year over year in the third quarter as the heightened demand from the AI build-up pushed prices upward.
“The impact runs through the spending on enhancements to the transmission and distribution grid required for data center buildouts, which is incorporated into the tariffs of all the ratepayers (residential, commercial and industrial) on the system,” Bank of America senior economist David Tinsley wrote in a recent note.
“How will the growth in electricity demand impact consumer bills from here? In BofA Global Research’s view, there is likely further upside ahead,” Tinsley added.
“[BofA research analysts] point to the fact that electricity supply is still struggling to catch up with the rapid increases in demand because of the capital intensity and regulatory requirements around building more generation and transmission capacity.”
The explosion in AI development is similar to the natural gas boom at the turn of the century, experts told Yahoo Finance.
Throughout the 20th century, most electricity in the US came from vertically integrated, government-regulated utilities that controlled everything from generation to distribution and billing.
In the 1990s, many states began to restructure their electric industries, allowing non-regulated companies to both generate and buy power, and then sell it to both utilities and end consumers directly.
At the same time, technology in natural gas extraction was rapidly advancing, the shale boom was just starting, and turbines became easier and faster to build. Companies rushed in, eager to make money building gas plants and then selling the energy on the market. Between 1999 and 2003, 175 gigawatts of capacity was built, according to Power Research Group, but the demand never caught up. Companies like Calpine Corporation and Energy Future Holdings, which had poured money into building natural gas plants and pipelines, went under.
“It was hot and [turbines] were scarce, and companies paid a lot of money to get those orders, and then the load didn’t materialize, and a few of the companies went bankrupt,” Rob Gramlich, the president of energy consulting firm Grid Strategies, said. “This industry has trouble when there are changes in forecast.”
For utilities today, however, this increased demand is being greeted as good news.
On Duke Energy’s (DUK) second quarter earnings call, president and CEO Harry Sideris highlighted an investment announcement of $10 billion from Amazon Web Services to build a data center campus in North Carolina.
“I’m proud to say that our team played an integral role in making this happen,” Sideris told analysts. “Our team continues to build on their track record of success, moving at pace with our customers to deliver what they need when they need it.”
Southern Company (SO) president and CEO Christopher Womack said much of the same on Southern’s last earnings call, highlighting that the company’s load pipeline from data centers and large manufacturers “remains well above 50 gigawatts of potential incremental load by the mid-2030s with project commitments totaling 10 gigawatts” and “ongoing advanced discussions for even more interest from large load customers in all of our electric service territories.”
In other words, requests for power from large industrial customers like AI data center developers are only picking up.
The speed at which utilities are working to source the equipment they need for new connections is evident at industrial giants like GE Vernova (GEV), Powell Industries (POWL), and Eaton Corporation (ETN), the manufacturers of the equipment utilities need to order.
Demand for data centers in GE Vernova’s electrification division brought in nearly $500 million in orders in the first half of 2025 compared to $600 million in all of 2024, CEO Scott Strazik said on a recent earnings call. Bank of America (BAC) is projecting the company will report 7 gigawatts of turbine orders in the third quarter, compared to 5.1 gigawatts in the second quarter, according to a recent note.
GE Vernova stock has gained over 80% this year, one of the best performers in the S&P 500.
Because demand is hot and supply is constrained, Grid Strategies’ Gramlich said, suppliers can up their prices, creating another potential problem for utilities trying to recoup the costs of these buildouts down the road.
“Those huge premiums that utilities are paying for that equipment could end up being something they regret if the load doesn’t materialize,” Gramlich said. “Since all new electrical equipment is really scarce and expensive, those marginal additions to connect up new loads are adding a lot of cost. Somebody is going to have to pay for that.”
Some utilities leaders have begun to consider what’s realistic and what’s not, Pierpont said.
Calvin Butler, the president and CEO of Chicago-based utility Exelon Corporation (EXC), which serves much of Illinois, took a measured tone on a recent earnings call when an analyst asked about data center demand.
“You’ll see more of those announcements coming in [on data center deals],” Butler said on the call. “But doing it the right way, with reliability in mind,” or without compromising power stability or cost for other ratepayers, “is key.”
Part of the solution to meeting this near-term power demand is on-site, or “behind-the-meter,” generation, largely with renewable energy sources.
Apple (AAPL), for example, built a solar array to partially power a data center in North Carolina. Amazon Web Services recently contracted AEP Ohio to install a series of on-site fuel cells that will provide direct power to two proposed data centers in central Ohio.
But as AEP Ohio noted in its announcement of the AWS deal, the fuel cells are intended to “provide the energy that allows these data centers to begin operations quickly, while the electric grid grows to support their needs.” In other words, these are stopgap measures until grid infrastructure is built.
Meta (META) has contracted Entergy Corporation (ETR) to build three natural gas turbines that could provide power to its planned $10 billion data center in Richland Parish, La. — one of the largest of such projects in the country — but it has also applied for connection to the larger grid.
“We work closely with utilities and grid operators to plan for future growth,” Amazon said in a statement to Yahoo Finance. “Where we require specific infrastructure to meet our needs (such as new substations), we work to make sure that we’re covering those costs and that they aren’t being passed on to other ratepayers.”
Meta and Apple did not respond to requests for comment.
Key for the utilities is trying to find ways to pass the risk on to the companies asking for the power, Energy Innovations’ Pierpont told Yahoo Finance. If a utility begins to prep for a major load request and then the client drops out, the thinking goes, the financial consequence should fall on the developer, not the utility.
That dynamic is on display at national electric utility American Electric Power (AEP). The utility has already signed on 24 gigawatts of incremental load to its system by 2030, “all backed by signed customer agreements, protecting us from changes in usage-driven volatility,” AEP president and CEO Bill Fehrman said on a recent earnings call. AEP’s customers include Amazon Web Services and Google (GOOG).
The COL4 AI-ready data center is located on a seven-acre campus at the convergence point of long-haul fiber and regional carrier fiber networks on July 24, 2025 in Columbus, Ohio. (Photo by Eli Hiller/For The Washington Post via Getty Images) ·The Washington Post via Getty Images
AEP’s backlog of demand, however, has reached 190 gigawatts. Not only is that “five times our current system,” Fehrman said, but it’s equal to nearly 15% of the entire US power grid, according to data from the Edison Electric Institute.
“Helping accelerate economic growth while also making sure we are paying for the infrastructure and electricity required to serve our operations is critical for Google,” Google said in a statement to Yahoo Finance. “We have and will continue to work closely with utility partners, generation owners and developers, and grid operators to plan and develop energy solutions that benefit the grid and all who use it.”
In October 2024, the AEP division covering Ohio, where Columbus has quickly become a data center hub, sent a request to the state’s utility regulator with a series of conditions.
First, require new data center customers to pay for at least 85% of the energy they sign up to use, even if they never end up using it; second, make them pay an exit fee for terminating a previously planned contract; and third, compel them to prove that they are financially strong enough to meet those requirements.
In July 2025, the Public Utilities Commission of Ohio, the state’s grid regulator, approved the proposal.
The plan, called a tariff in the utilities industry, worked. Only two months later, in September, AEP Ohio cut its data center load demand forecast by more than half.
“It helps [utilities] see who is willing to put money behind [these projects],” Pierpont told Yahoo Finance.
Assistance is also coming from the federal government. After pausing or canceling a litany of grants handed out by the Department of Energy at the tail end of the Biden administration, the Trump administration’s DOE approved Thursday a $1.6 billion loan to AEP that will fund efforts by the utility to rebuild more than 5,000 miles of power transmission cables throughout the mid-Atlantic region as AEP stares down a slew of connection requests from new Big Tech clients like Amazon, Google, and Microsoft (MSFT).
Microsoft declined to comment.
AEP, Fehrman said in a statement on the grant, is “experiencing growth in energy demand that has not been seen in a generation.”
Jake Conley is a breaking news reporter covering US equities for Yahoo Finance. Follow him on X at @byjakeconley or email him at [email protected].
Click here for in-depth analysis of the latest stock market news and events moving stock prices
Read the latest financial and business news from Yahoo Finance