9 Policies for Addressing AI’s Energy Costs
It’s hard to watch a news broadcast or pick up a newspaper without hearing about rising electricity bills—and the forces behind the sticker shock that many Americans are feeling when they open their Pepco, ConEd or Duke Energy envelopes. The data centers that power artificial intelligence are popping up rapidly across the United States, with one estimate claiming that more than 3,000 new centers are already planned or under construction. These data centers require a colossal amount of energy, with a single hyperscale data center able to use more electricity than 80,000 households, according to the U.S. Energy Information Administration.
And yet, the only thing rising faster than household energy bills is Americans’ frustration with the sky-high cost of utilities. Electricity bills are a “major” source of stress for more than one-third of adults, according to an October 2025 Associated Press-NORC Center for Public Affairs Research poll. Rising utility bills were a salient issue in the 2025 elections and are on track to feature prominently as a political issue in this year’s midterms, as politicians from both major parties hone their arguments on affordability. Similarly, policymakers at all levels of government have grown increasingly concerned about the resilience of the electrical grid, given not just rising demand from data centers, but the impacts from aging infrastructure and extreme weather.
In a new report published today, Matthew McHale and Hannah Wiseman propose nine ideas for state and federal policymakers to protect households from increased costs of electricity and to ensure a reliable, resilient grid.
McHale and Wiseman propose six policies to ensure that new and proposed data centers do not drive up rates for residential, commercial, and smaller industrial customers. These include:
Requiring utilities to allocate the costs of grid expansions from data centers to the data centers themselves, rather than other consumers
Requiring data centers to pay for additional local infrastructure costs
Ensuring tax breaks and other benefits are transparent and provide net benefits to the community
They also suggest avoiding overbuilding, ensuring publicly available rates and terms of service, and tracking disconnections to monitor affordability. The authors also propose three policies for ensuring grid reliability and resilience.
As the authors write, “With these tools, the public can better monitor the impact of AI data centers and other large loads, and can meaningfully participate in the ongoing policy project to harness the opportunity of AI while minimizing its dangers to society.”
The report builds on VPA’s research on AI and data centers, including VPA Director of AI and Technology Asad Ramzanali’s paper on how to regulate cloud computing and an essay in Commonplace on state subsidies for data centers by Ramzanali and VPA Director Ganesh Sitaraman.