Back
Essentially everyone agrees: Americans shouldn’t pay higher electric bills to feed AI data centers’ insatiable demand for power. But what will it actually take to prevent cost spikes?
Lots of states have decided the answer is a “large load tariff” — an unsexy term that basically translates to special utility rates and requirements designed for huge energy users, like data centers.
As of late 2025, more than 65 such tariffs have been proposed or approved in over 30 states, according to data tracked by the Smart Electric Power Alliance and the North Carolina Clean Energy Technology Center.
These efforts are largely trying to solve the same problem: The explosive growth of AI data centers is outpacing utilities’ ability to build power plants and upgrade the grid. If data centers don’t show up and stick around to buy all the power that’s justifying those investments, other customers could be trapped paying them off for decades to come.
This puts enormous pressure on regulators to “hold the line on ensuring that these large-load customers carry the costs that they bring,” said Jay Griffin, executive chair at the Regulatory Assistance Project. The nonprofit last month launched a report series to help regulators and policymakers navigate these complexities.
The trend of states adopting data center–focused large load tariffs began to take off in 2024, led by early movers like Ohio and Indiana. More such tariffs were approved in Kansas, Michigan, and Virginia last year, and now Illinois and Wisconsin are debating their own proposals. With roughly a year and a half of data on how different states have tackled the problem, “there’s enough time and transparency into decision-making that commissioners are able to make appropriate decisions,” Griffin said.
Progress is decidedly mixed, said Louisa Eberle, a senior associate at the Regulatory Assistance Project who co-wrote its first data center report. “Some are just getting started. We haven’t reached full ‘best practices’ anywhere — but we have found better practices.”
Those start with contracts requiring these giant new customers to pay a minimum amount of money for power for a set period — usually 10 to 15 years — whether or not they end up being built or staying open that long. This offers some insurance against data centers pulling out and leaving customers at large holding the bag, although some advocates fear those terms aren’t lengthy enough to cover the cost of power plants and grid investments, which must be paid off over decades.
Some tariffs also lay out what kind of power such massive customers must use — namely, clean energy. These can match up nicely with both state climate targets and the clean energy goals of the tech giants, like Amazon, Google, Meta, and Microsoft, that are driving the AI boom — although plenty of utilities and data center developers are going big into fossil gas–fired power as well.
And on the cutting edge of large load tariff policy, some utility regulators are asking data centers to “bring their own” generation or grid capacity, Eberle said. The idea here is to make developers play a more active role in sourcing and contracting for new energy resources for their computing facilities. That might not be utilities’ favorite option, since it cuts into the profits they earn from investing in power plants and power lines. But it’s an opening for data center developers willing to pay a premium to get onto the grid faster.
These negotiations aren’t easy, Griffin said. Tech companies are asking utilities to invest billions of dollars to serve power demand equal to that of entire cities springing up on their grids over just a few years. The sheer scale and speed of the boom have overwhelmed regulatory processes built for slow and low growth.
And the future is highly uncertain. Tech companies keep upping their AI spending plans, even amid mounting signs that the sector is a bubble about to pop. The Trump administration’s call in recent months for data centers to build their own power plants as a means to protect utility customers from rate increases conflicts with hard limits on how quickly new generation can be built and connected to the grid.
As the former chair of the Hawaii Public Utilities Commission, Griffin knows that regulators are constantly balancing the risk of letting utilities build too much power with the risk of preventing them from building enough. The former threatens to burden customers with unnecessary costs, while the latter can constrain economic growth and even endanger grid reliability.
Right now, public opposition to data centers is squarely focused on the financial and environmental dangers of overbuilding. Laws passed in Minnesota, Oregon, and Texas last year, and bills being debated in states including Florida, Georgia, Illinois, Virginia, Washington, and Wisconsin, propose everything from stripping tax breaks for data centers to imposing full-on construction moratoriums.
However, data centers that cover their costs and finance more-sustainable resources could help in “reducing cost for everyone,” Griffin said, both by increasing utility revenues to cover shared expenses and by pushing “innovation for emerging technologies,” such as virtual power plants and on-demand clean energy resources like geothermal power. Tech giants “have the demand for power and the need for speed to drive those in a way we’re probably not going to see for another generation,” he said.
While no two large load tariffs are exactly alike, many share common characteristics, as think tank RMI highlighted in a November review.
About a third of the 65 large load tariffs on deck as of late 2025 require big customers to make minimum payments over a set period of years, whether or not they remain operational over that time. More than half include some form of collateral requirements or other credit risk protections. And roughly half require large customers to pay fees if they exit their contracts early.
These requirements can help cull the speculative data center proposals now crowding utility interconnection queues, whether from companies with projects that are highly unlikely to win financing or from major developers “shopping” single projects across multiple utility territories. American Electric Power’s Ohio utility, for example, saw its large load pipeline drop from 30 gigawatts to 13 gigawatts after it instituted a large load tariff last year. In that sense, “not only do strong tariffs help protect customers, they also help the utility in forecasting what’s coming,” Eberle said.
But the tariffs might not be sufficient to pay off the cost of power plants and grid investments that last for decades, said Ben Hertz-Shargel, global head of grid edge at research firm Wood Mackenzie. Last year, he ran an analysis that found none of the large load tariffs on the books at that time were sufficient to fully recover the cost of new gas-fired power plants that would need to be built to serve big energy users.
The scale of fossil fuel build-out being contemplated to serve the high side of the AI bubble would be ruinous on both cost and climate terms. The Sierra Club is tracking a startling 248 gigawatts of gas-fired power plants being planned across the U.S. as of the first quarter of 2026, nearly five times the amount planned in 2021. Data center expansion is the primary driver of that increase, including for build-outs planned in Georgia, Louisiana, and North Carolina— states that have yet to impose large load tariffs.
“There are some utilities that are starting to creep up and charge for what it takes to build a new power plant today,” Hertz-Shargel said. “But it’s still uneven.”
Nor can tariffs guarantee that data centers will pay for transmission built to accommodate their impact on regional grid networks, he said, since those costs are allocated via complex structures that make it hard for utilities to force expenses on individual customers. The Illinois attorney general’s office has raised that issue in challenging utility Commonwealth Edison’s proposed transmission service agreements for data centers.
Even tariffs specifically designed to force individual data centers to cover the costs of utility investments expose customers to financial risk, said Jeremy Fisher, principal adviser on climate and energy with the Sierra Club’s Environmental Law Program.
He cited Wisconsin, where utility We Energies has proposed two tariffs meant to isolate the cost of building power plants and transmission grids to the gigawatt-scale data centers being planned in its territory. Those tariffs allow data centers to pay for new solar, wind, and battery storage. But they also offer an option for the facilities to contract for power and capacity from two gas-fired plants that the utility is planning to build. Under that latter option, everyday customers would remain responsible for paying for 25% of the cost of building these plants, as well as for the fuel they burn.
Meanwhile, two of the planned data centers in We Energies’ territory will consume as much power as the utility’s entire residential customer base, Fisher noted. “I don’t know how you quantify the concentration risk of two customers doubling the size of your load,” he said. “We’ve never seen anything like this.”
Hertz-Shargel added that the risk of a handful of customers driving most new demand is compounded by the nature of AI growth. The sector is fueled by hundreds of billions of dollars of debt financing and circular deals that could unravel if one or more major players fail to deliver.
“If the utility is going to have half of its assets caused by and paid for by a small number of customers, you need to be very concerned about that level of business risk,” he said.
That’s why Hertz-Shargel and other clean energy advocates are pushing a solution adopted by only a handful of utilities and regulators so far: requiring data centers to contract for their own clean energy and capacity.
The concept goes by many names — one of the catchiest is BYONCE, for “bring your own new clean energy.” But Hertz-Shargel uses the term “clean transition tariff,” a phrase coined by Google and Nevada utility NV Energy for a tariff approved by state regulators last year. That agreement allows the search giant to directly tap a geothermal plant being built by startup Fervo Energy.
Last week, Google announced a plan with Minnesota utility Xcel Energy that expands on this premise. Like Google’s agreement with NV Energy, it is a one-off deal rather than a tariff that applies to other large-load customers. But under it, Google will pay for the construction of 1,400 megawatts of wind, 200 megawatts of solar, and 300 megawatts of energy storage, and cover the grid infrastructure costs to bring it all online. It will also invest $50 million in the utility’s proposed Capacity*Connect distributed battery program.
“All parties should love it,” Hertz-Shargel said. “Data center companies get to choose the generation technology that supplies them, generation developers can play in new markets, and utilities get to sleeve the agreements between them.”
Utilities that profit from building power plants may not be as enthused, he conceded. But they already have enormous investments to make in distribution and transmission. “Adding on power plants to serve data centers would add additional revenue, but at enormous political cost,” he said.
Utilities can also squeeze more clean capacity out of the existing grid, Eberle noted. That could look like improving energy efficiency, paying customers to use less power when demand is high, and leveraging rooftop solar systems and home batteries to ease strain on the grid. These strategies “can be scaled up quickly and cheaply,” and they “will be useful even if the load doesn’t emerge,” she said.
Data centers could also agree to strategically reduce their own power use when the system is strained or to install batteries that can relieve near-term grid pressures.
How can large load tariffs tap into this kind of clean and flexible capacity? Fisher highlighted last year’s settlement agreement between Kansas utility Evergy, which has some significant data center projects in its territory, and groups including the Sierra Club, the Natural Resources Defense Council, Google, and the Data Center Coalition.
The tariff allows data centers to earn credit for flexibility they contract directly, Fisher said. But it also gives them the option to contract for renewables, energy storage, or efficiency programs in Evergy’s integrated resource plan, the regulator-mandated process to determine the mix of new power plants and programs the utility can invest in.
That’s an important wrinkle on the “bring your own” concept, Eberle said. It allows data centers to “engage with the utility and say, ‘We really like this resource that you identified but didn’t select — we’d like to pay for it.’”
Another option under the tariff would allow Evergy to seek out and directly charge a developer for the capacity needed to allow a data center to come online, she noted.
Griffin highlighted that these kinds of collaborative agreements take more time and require concessions from utilities and data center companies alike. But “you’ll be more successful if you give commissioners and stakeholders time and space to do the vetting — and that should support the more sound business models,” he said.
As for data center developers trying to push their costs onto consumers, Griffin said, “the more you force commissions to stick their neck out — well, you don’t get that pass many times.”
A clarification was made on March 4, 2026: This story has been updated to clarify that the Smart Electric Power Alliance’s data on large load tariffs was compiled in partnership with the North Carolina Clean Energy Technology Center.
.svg)
