AI Is Showing Up on Your Electricity Bill

50% of US electricity demand growth in 2025 came from data centers

In January 2026, John Steinbach opened his electricity bill in northern Virginia. The total: $281, up from roughly $100 the month before. He hadn't bought an electric car. He hadn't installed a hot tub. The Loudoun County area where he lives has nearly 600 data centers, more than any other region on the planet, and they don't pay his bill. He does.

Steinbach's story made it into a Consumer Reports article in March 2026, but it isn't unusual. According to a January 2026 PowerLines analysis cited by Fortune, US power bills have risen 40% since 2021, and electric and gas utilities filed more than $30 billion in rate increase requests during 2025 alone. Roughly 81 million Americans are exposed to those requests. The chart you don't see in any AI keynote is the chart of household power bills, and right now it's bending the wrong way.

This is the most tangible consequence of the AI boom that no one running it wants to talk about. AI keeps showing up on stage as a productivity story, an automation story, a reasoning-model story. Behind those stories sit data centers, and behind data centers sits a power grid that wasn't built for what's being asked of it.

On April 16, 2026, the International Energy Agency put numbers on the shift. Global data center electricity consumption surged 17% in 2025, against just 3% growth for global electricity demand overall. In the United States specifically, data centers accounted for roughly half of all electricity demand growth in 2025, according to the IEA's analysis published April 20 by Fortune. That ratio is projected to hold through 2030. It's hard to overstate how much that bends the math of how a country plans, prices, and pays for power.

Goldman Sachs Research forecasts global data center power demand will rise 165% by the end of the decade compared with 2023, with more recent updates putting the figure as high as 220%. Goldman estimates meeting that demand will require about $720 billion in new grid spending by 2030. The Department of Energy and Lawrence Berkeley National Laboratory project data centers could consume up to 12% of all US electricity by 2028. None of that capacity is free. The question is who pays for it, and the answer is starting to show up in mailboxes.

What 50% of Demand Growth Actually Means

Half of all new US electricity demand in 2025 came from data centers. Most utility regulators and grid operators model their five and ten year plans assuming residential, commercial, and industrial demand grow on their long-term curves. AI broke that assumption inside two years. Nobody was preparing for a doubling of any single category by 2030, but that's now the IEA's central case for data centers.

Building new generation, transmission, and substation capacity is slow. Permits for transmission lines often take five to ten years. Natural gas turbines have backlogs running into 2028. Nuclear is the longest cycle of all. Demand can show up in a year, but supply takes a decade. The squeeze in between gets passed to ratepayers under the standard utility model: when a utility makes a capital investment, regulators allow it to recover the cost from customers plus an authorized return.

In practice, this means the Virginia household paying $281 in January is partially financing the substation upgrades that let a hyperscaler open a new training cluster down the road. Bloomberg's analysis cited in Consumer Reports found that areas with high concentrations of data centers saw electricity prices jump 267% over the past five years. In Virginia specifically, data centers consumed almost 40% of all state electricity in 2024.

The Public Already Sees It

A Consumer Reports survey of 2,146 US adults in November 2025 found that 78% are somewhat or very concerned that new data centers will raise their energy bills. A March 2026 Pew survey found Americans more likely to hold negative views on data centers' environmental costs than positive views on jobs and tax benefits. More than half of respondents told Pew they expect AI to do more harm than good in the long run.

This is unusual. Most major technology shifts in the past 30 years arrived with public optimism running ahead of public skepticism. Cloud, smartphones, broadband, even social media in its early years polled positively until well after their second-order effects landed. AI is starting from skepticism, and the electricity bill is making that skepticism concrete.

That's why this issue moves quickly through state legislatures. According to MultiState's April 14, 2026 analysis, states have considered hundreds of data center bills in the past year. California enacted SB 57. Ohio enacted SB 103. Utah enacted HB 507. Maine is on track to become the first state to enact a moratorium pausing new data center construction until November 2027 to study community impacts. New Jersey's Bill A796 would require data centers to pay their own electricity costs rather than pass them to households.

The Counter-Narrative

The industry isn't ignoring the politics. On March 4, 2026, several major data center developers signed the White House's Ratepayer Protection Pledge, committing to cover the full cost of new electric generation resources their facilities require. That is, in principle, the right answer to the rate-shifting concern. If a hyperscaler builds a $5 billion campus in Texas, it pays for the substation upgrade and the long-term power contract, not the family in Lubbock.

Goldman Sachs estimates 40% of the increase in data center demand will be met by renewables, with a modest amount of nuclear capacity targeted specifically for AI workloads. Some regions, North Dakota among them, have seen rates decrease in places where data centers fund their own infrastructure rather than relying on shared utility build-outs. The pledge signers are betting that voluntary cost coverage plus transparent power purchase agreements get them ahead of the regulatory wave.

Whether the pledge holds in practice is the open question. Voluntary commitments without enforcement mechanisms have an uneven track record in this industry. The same data center developers can sign a pledge in Washington and sit in PJM rate cases the next quarter where the cost allocation is anything but clear. Watch the dockets, not the press releases.

Why This Is the Story That Lasts

A lot of AI coverage rolls past in 48 hours. Model launches, benchmark wins, executive resignations, a viral demo. Most of it goes back into the noise floor within a week. The electricity story doesn't, because it shows up in your mailbox every month and doesn't depend on anyone's narrative.

It's also the story most likely to drive regulatory and political action that materially reshapes how AI gets built in the United States. Bipartisan coalitions are unusual in 2026. Senator Bernie Sanders and Governor Ron DeSantis voiced concern about AI data center electricity prices in the same week earlier this year. When Sanders and DeSantis worry about the same thing, the policy response stops being theoretical.

Globally, Ireland is the canary. Data centers there now consume 21% of national electricity, more than any country in the world. Irish regulators have already paused new data center connections in some regions. The same dynamic is playing out at the regional level in Virginia, in parts of Texas, and increasingly in suburban Phoenix. The pattern repeats: load arrives faster than the grid can adapt, the local rate base ends up subsidizing the gap, and the political backlash follows.

What This Means for You

If you're in any role that touches AI procurement or strategy, the electricity story is now part of your stack. Three concrete implications.

First, the cost of compute is going to keep rising, and not just because GPUs are expensive. Power and grid-interconnection costs are baked into every cloud bill, and the regions with cheap, abundant electricity are getting saturated. Vendor pitches that quote a price per token today are pricing inputs that are about to revalue. Ask, when does this contract reprice, and what's your hedge against power costs.

Second, there's a regulatory horizon you need to track. Maine's moratorium, NJ A796, and the Ratepayer Protection Pledge are early markers. By late 2026 or 2027, many US states will likely have rules requiring data center developers to either pay separately for grid upgrades or accept locational caps. If you're building something that depends on a particular region's data center capacity, that capacity is no longer a given.

Third, for the rest of us, this is the most visible argument against thinking of AI as something happening "elsewhere." The bill arriving at your address is the AI economy's pricing signal made personal. You don't need to be a technologist to read it, and politicians on both sides of the aisle have already noticed. The companies that get ahead of this, the ones building or buying AI tools, will be the ones that can answer "what are you doing about the power bill" with something more substantive than a logo on a press release.