AI is driving up electricity demand—will you pay the price?

The explosive growth of AI is driving a huge surge of demand for power and shows no signs of slowing down.

Electricity consumption from U.S. data centers, which is crucial for powering generative artificial intelligence (AI) models such as ChatGPT and Gemini, is expected to increase by 130% by the decade’s end, according to a new study by the International Energy Agency (IEA). Data center power demand has already grown 12% annually between 2015 and 2024, the IEA says. 

So, how will we pay for all of this new energy ushering in the era of AI?

To keep up, utilities plan to significantly expand their electricity generation using a mix of natural gas, solar, and wind.

But with federal clean energy policy under threat in Washington, the deployment of these new renewable resources that are key to powering data centers could be delayed or derailed thanks to potential tax credit cuts. That means the cost of expanding the U.S. power grid may fall directly on households through higher utility bills as electricity rates increase

The GOP’s reconciliation bill, now under consideration in the Senate, will eliminate or drastically accelerate the phase-out of many clean energy tax credits established by the Inflation Reduction Act (IRA). In particular, the residential solar tax credit (Section 25D of the U.S. tax code) is at serious risk of being repealed at the end of this year or early next year—a move that could devastate U.S. solar jobs, weaken energy independence, and raise long-term electricity costs for American families.

While financial incentives for large-scale commercial and utility solar projects may fare slightly better with the bill’s current language, the accelerated phase-out and strict limits on materials sourced from “prohibited foreign entities” would make it much more difficult for new renewable energy developments to qualify for tax credits. 

As a result, clean energy industries will contract, leading to greater strain on the U.S. power grid at a time when electricity demand is already skyrocketing, and resulting in higher electric bills for everyone.

“The need for clean energy extends far beyond data centers,” Daniel Bresette, president of the Environmental and Energy Study Institute (EESI), told EnergySage.“Even without taking into account climate or environmental concerns, which are significant and urgent, the U.S. energy sector has been on a clean-cleaner-cleanest trajectory powered by the cost-competitiveness of renewable energy. The best way to ensure that data centers are powered by clean energy sources is not to let ourselves revert back to dirtier, less-affordable alternatives,” said Bresette.

After a decade of stagnation, U.S. data center electricity consumption grew 12% annually from 2015 to 2024. This trend is expected to accelerate dramatically, with consumption levels projected to increase by 130% by 2030, according to the IEA report.

Energy consumption of data centers

It’s clear that AI is driving the energy demands of data centers. The training required to teach AI models uses massive amounts of computing power—the ability of a computer system to process data and execute tasks—on top of the energy it takes the model to answer user queries. For example, GPT-4 used around 30 megawatts (MW) of power just for training purposes alone. 

The United States leads globally in per-capita data center consumption. By the decade's end, IEA forecasts that U.S. per-capita data center energy usage could amount to approximately 10% of total annual household electricity consumption. By 2030, the U.S. economy is anticipated to consume more electricity for data processing than for all manufacturing of energy-intensive goods combined, including aluminum, steel, cement, and chemicals.

According to the IEA, conventional data centers typically require 10 to 15 megawatts (MW) of power, while AI-focused hyperscale centers can demand 100 MW or more. As AI models and services proliferate, data centers are expanding rapidly to keep pace with rising energy needs.

AI-induced demand spikes are regionally concentrated

One of the biggest challenges when it comes to data centers is that they’re often located near cities and are highly concentrated in certain areas, creating regional consumption spikes. This disproportionate demand ends up straining local grids, often causing households in the same service area to bear the cost through higher electricity rates.

U.S. data center hubs

To address the anticipated surge in electricity demand, many utilities plan to build new power plants while also expanding renewable energy resources—but these upgrades can take years to complete. New or expanding data centers in congested markets often face multi-year connection delays.

“Renewable energy has been booming in recent years, but we need even more, as well as grid improvements, to accommodate new-generation assets coming online,” Bresette said. “All of this needs to happen yesterday. Renewable energy offers many benefits in terms of job creation and sustainability, and it also happens to be exceedingly cost-competitive with fossil resources. We just need to get more of it on the grid.”

The IEA report indicates that natural gas is currently the leading source of electricity for U.S. data centers, supplying over 40% of their power—a situation that isn’t sustainable long-term. Renewables—particularly solar and wind—also play a critical role in meeting the growing electricity demand from AI.

As of the first quarter of  2025, plans show that U.S. utilities intend to add 94 gigawatts (GW) of new natural gas capacity and 262 GW of additional solar capacity by 2035. The IEA projects that solar and wind will remain the second-largest source of electricity for data centers, contributing an additional 110 terawatt-hours (TWh) between 2024 and 2030.

Utilities cite the low cost of natural gas and the high reliability requirements of data centers—which need uninterrupted power 24/7—as key reasons for expanding natural gas generation. According to the IEA report, major data center operators have already partnered with utilities and energy developers on new gas infrastructure. For example, Entergy Louisiana plans to build over 2 GW of additional natural gas generation to power Meta’s data centers.

While natural gas is seen as an affordable energy source, it’s a fossil fuel that produces harmful carbon emissions. Additionally, the construction of new grid resources always comes at a price, no matter the source. Those costs are ultimately passed on to homeowners.

Alternatively, power purchase agreements (PPAs) and co-locating data centers with renewable power plants help align AI energy use with clean energy deployment while lessening the impact on homeowners’ bills. 

Power purchase agreements (PPAs)

Many data center operators use PPAs to help meet their clean energy targets. While structures vary, a PPA is essentially a financial contract in which the data center agrees to buy a set amount of electricity from a renewable energy project—typically enough to match its annual electricity use.

In some cases, PPAs deliver renewable electricity directly to the facility. However, most are tied to off-site projects that don’t physically power the data center. Instead, the renewable energy is sent to the broader grid, serving as a clean energy offset that accounts for the data center’s electricity consumption.

Co-location

Some companies are choosing to co-locate data centers with renewable energy projects to reduce dependence on grid-supplied electricity and directly source clean power, bypassing the indirect nature of most PPAs. Developing a co-located renewable facility can also be faster than waiting for utility companies to build new grid infrastructure. However, this approach often comes with higher upfront costs, more complex permitting requirements, and increased responsibilities for maintenance and operations.

Artificial intelligence is fueling massive electricity demand, and clean energy is critical to meeting that demand affordably and sustainably, but ongoing policy threats could shrink the clean energy pipeline just when it's needed most.

In its current form, the reconciliation bill will accelerate the sunsetting of clean energy tax credits over the next six months. This phase-out effectively strips away the business certainty that injects confidence into the future of clean energy projects—like those needed to power AI data centers at scale.

If the GOP’s budget bill passes, these measures could drastically slow the pace of new solar, wind, and battery storage projects, leaving utility companies to fill the energy gap—likely with fossil fuels—and pass the cost of new infrastructure directly to American households. 

“The bill would end the tax incentives that established a decade of certainty underpinning generational levels of investments in on-shore manufacturing and clean energy technologies,”  Bresette said. “Ending these tax incentives would make keeping pace with increasing electricity demand from data centers costlier and much less sustainable.”

Clean energy isn't just a climate solution—it’s a cost-control mechanism. As AI adoption expands, weakening clean energy policies will make clean tech’s benefits harder to afford for everyone else.

The good news is that the bill isn’t law yet. If you're concerned about how the current version of the GOP bill could impact your electricity bills, contact your representatives and make your voice heard.