Skip to content

AI’s Power Consumption Could Put the Grid — and Energy Regulators — to the Test

The story of artificial intelligence (“AI”) is one of technological promise and societal challenge, and its impact on the U.S. electric power grid is fast becoming a pivotal chapter.

The promise is compelling. AI could revolutionize the electric industry, enhancing grid planning, permitting and siting, operations and reliability, and resilience, while helping accelerate progress toward climate goals.1

Yet powering this revolution will be a massive challenge. AI data centers consume eight times the power of non-AI-powered data centers.2 And while modern data centers have become more efficient, the rate of power efficiency gains is decelerating.3 Rising consumption will drive significant cost increases — stemming from demand growth for power as a commodity and from demand for the electric grid to deliver power to data centers.

In 2022, data centers accounted for about 2.5 percent of U.S. electric demand.4 By 2030, that figure could rise to 20 percent, with AI data centers accounting for three-quarters of the demand.5 If forecasts hold, this growth will come alongside skyrocketing demand for new transmission facilities, as the electric grid faces growing demand and changing needs from robust U.S. manufacturing, electric vehicle charging, and the shifting generation mix, among other strains.6

National, Regional, and Local Concern

Like the power grid itself, AI’s power consumption is at once an issue of national, regional, and local concern. And as with the internet revolution before it, Northern Virginia sits at the heart of the issue.

Supported by generous state tax incentives,7 Northern Virginia hosts the largest data center market in the United States — larger than the next five U.S. markets combined.8 Also driving demand in the region are local electricity rates that are nearly 30 percent below the national average.9

But major challenges lie ahead. Power demand from data centers is projected to increase fourfold over the next 15 years, rising to 40 percent of total demand for the local electric utility, Dominion Energy.10 In addition, Dominion’s peak demand — a leading driver for generation planning — is expected to grow at least 5 percent every year for the foreseeable future.11

This unprecedented load growth comes with new transmission and generation needs, which must be paid for, typically through customer rates.12 Such price pressures will increasingly be felt nationwide. It was only a matter of time before these dynamics fell before state and federal regulators — and they have.

FERC into the Fray

The Federal Energy Regulatory Commission (FERC) has already been pulled into the fray. In Northern Virginia, reliability impacts are addressed (and paid for) through a regional process administered by the regional transmission organization, PJM Interconnection (PJM), under tariff provisions filed with FERC. Facing costs that are allocated regionally, neighboring ratepayers in Maryland cried foul, claiming that Virginia’s tax incentives had driven the state’s massive data center growth, and that Marylanders were unfairly sharing the transmission costs of the data centers while Virginians enjoyed nearly all the benefits.

In February, the Maryland Office of Peoples Counsel (“Maryland OPC”) took up the cause, challenging the cost allocation via PJM’s regional process to address reliability violations on the transmission grid. The Maryland OPC argued that the regional process for addressing reliability, which results in regional cost allocation, had been misapplied, contending that the costs for new transmission projects should instead be allocated based on the rules for policy-driven projects — so-called Multi-Driver Projects under the PJM tariff.13

But in April 2024, FERC overruled the Maryland OPC and others’ protests, finding that the transmission project at issue was not a Multi-Driver Project, and thus does not justify more localized cost allocation.14 Recognizing that many states are pushing energy policies that affect neighboring ratepayers, Commissioner Allison Clements concurred, characterizing as unnecessary and impractical the task of attempting to assign transmission costs by determining which of countless public policies contribute to the need for new transmission.15

This dispute is among the first centering on power-hungry AI data centers and regionalized cost sharing. But as states continue to pursue divergent economic and energy policies, it will not be the last.

Beyond Rate Impacts

The sharp growth of AI data centers could also affect the future of fossil fuel-fired generation.16 Yes, the AI sector in general is benefitting from the green promises of the technology. And the U.S. Department of Energy recently highlighted the “promising opportunities for AI to accelerate decarbonization” and advance the broader transition to a clean energy economy,17 consistent with the current administration’s goals.18

Yet a stark reality is also sinking in: The AI revolution requires major energy inputs at a time when the closure of coal-, gas-, and oil-fired power plants is shrinking the baseload generation fleet, a trend that is expected to continue under (and in fact is a goal of) the Environmental Protection Agency’s Clean Power Plan 2.0.19 So, it is far from clear where new AI data centers will get the power they need, as the supply and demand curve must always cross. The math is the math.

The need for fossil fuel-fired generation to power data center growth is also at odds with the internal policies of major tech companies like Google and Amazon,20 which are championing their ability to deliver AI’s power without tarnishing their green credentials. Companies are racing to secure power for data centers through power purchase agreements favoring renewable resources, but the options for doing so are growing fewer by the day.21

Even if new, renewable generation can be planned, permitted, sited, and constructed in time to meet the needs of the AI data centers, the current electric grid might not be able to handle the power. The existing grid was generally built to move power to cities from remote coal- and natural gas-fired generation located near fuel sources. While some data centers are targeting locations close to existing generation sites,22 there are simply not enough prime locations to accommodate the growth.23

As data centers continue to demand renewable energy, most of their power supply will likely have to be moved long distances across the transmission grid from new generation locations to new load centers. Doing so would require a massive build-out of the grid, which is simply not growing fast enough to handle the influx of new load from data centers.24 There is a three-year backlog on key transformers, for example.25 The quest to build AI-powered data centers near reliable power sources has sparked a race to places where renewable energy can be collocated or the grid is resilient enough to accommodate the enormous power needs, as reflected in initiatives like Amazon’s $10 billion data center project in Mississippi.26

Reshaping the Energy Landscape

AI holds enormous potential to reshape the energy landscape and will challenge regulators. Potential harms from AI, like data privacy and cybersecurity threats, are likely to keep regulators up at night.27 Other applications, like using AI-powered cameras to provide around-the-clock fire monitoring and safety alerts, are less controversial.28

The applications for AI to enhance energy infrastructure are many: AI-accelerated power grid models for capacity and transmission studies, large language models to enable compliance with federal permitting, advanced AI to forecast renewable energy production for grid operators, smart grid applications of AI to enhance resilience, and even AI-powered methane detection from pipelines.29

AI will also bolster the capacity of regulators to process information and monitor markets. FERC’s fiscal year 2025 Budget Request to Congress, for example, includes funding to harness the generative potential of AI in its operations.30

Looking ahead, one thing is clear: As data centers are built and large language models are trained, AI’s impact on the power grid — and energy regulation — will continue to grow. Indeed, major government initiatives are pushing the growth of AI to help modernize critical energy infrastructure, which serves as the backbone of the national economy. Regulators will increasingly confront issues around the safe, reliable, and cost-efficient deployment of AI throughout the energy economy. This is only the beginning.

1U.S. Department of Energy, “AI for Energy, Opportunities for a Modern Grid and Clean Energy Economy” (April 2024), available at (“DOE Report”).

2See Dominion Energy, Business review investor meeting, at 49 (Mar. 1, 2024) (describing rack power density), available at (“Dominion Investor Presentation”).

3Goldman Sachs, “Generation growth; AI, data centers and the coming US power demand surge (Apr. 28, 2024), available at

4DiLallo, Matt, AI Could Power Huge Demand for This Fuel by 2030, The Motley Fool, Apr. 20, 2024, available at


6See Sisson, Patrick, “AI Frenzy Complicates Efforts to Keep Power-Hungry Data Sites Green,” New York Times (Feb. 29, 2024), available at

7Dominion Investor Presentation at 48.

8Dominion Investor Presentation at 47.

9Dominion Investor Presentation at 48.

10Dominion Investor Presentation at 50.

11Dominion Investor Presentation at 45-46.

12Hiller, Jennifer and Patterson, Scott, “How Big Data Centers Are Slowing the Shift to Clean Energy,” Wall Street Journal (Apr. 29, 2024), available at

13PJM Interconnection, LLC, Motion for Leave to File Answer and Answer of the Maryland Office of People’s Counsel, FERC Docket No. ER24-843 (filed Mar. 20, 2024).

14PJM Interconnection, L.L.C., “Order on Cost Allocation Report and Tariff Revisions Order,” 187 FERC ¶ 61,012 at 3–7 (Apr. 8, 2024) (citing PJM Open Access Transmission Tariff).

15Id., Concurrence of A. Clements at 5.

16See Hiller and Patterson, supra.

17DOE Report at vi–vii, 26–36.

18The White House, “Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence” (Oct. 30, 2023), available at‌10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/.

19See Environmental Protection Agency, New Source Performance Standards for Greenhouse Gas Emissions From New, Modified, and Reconstructed Fossil Fuel-Fired Electric Generating Units; Emission Guidelines for Greenhouse Gas Emissions From Existing Fossil Fuel-Fired Electric Generating Units; and Repeal of the Affordable Clean Energy Rule, 89 Fed. Reg. 39,798 (May 9, 2024).

20Hiller and Patterson, supra n12; Sisson, supra n4.


22See Sisson, Patrick, “AI Frenzy Complicates Efforts to Keep Power-Hungry Data Sites Green,” New York Times (Feb. 29, 2024), available at

23See Halper, Evan, “Amid Explosive Demand, America Is Running Out of Power,” Washington Post (Mar. 7, 2024), available at

24See Sisson, Patrick, “AI Frenzy Complicates Efforts to Keep Power-Hungry Data Sites Green,” New York Times (Feb. 29, 2024), available at; Walton, Robert, Granholm Tells Congress ‘Adjustments Have Been Made’ to Distribution Transformer Proposal, Utility Dive (Mar. 21, 2024), available at


26Mississippi Development Authority, Amazon Web Services Plans to Invest $10 Billion, Creating 1,000 Jobs to Establish Data Center Complexes in Mississippi (Jan. 25, 2024) available at

27See DOE Report at 24–25, 39–41, identifying risks from the integration of AI on the power grid.

28See Nevada Power Company and Sierra Pacific Power Company, 2023 Form 10-K, at 28 (Apr. 16, 2024), available at

29DOE Report at 5–8.

30Federal Energy Regulatory Commission, FY 2025 Congressional Justification at 10, 83 (Mar. 11, 2024), available at‌Congressional%‌20Justification_3-11-2024_post.pdf.

This information is provided by Vinson & Elkins LLP for educational and informational purposes only and is not intended, nor should it be construed, as legal advice.