Perspectives

Big Tech's AI Boom: Who Pays the Electricity Bill?

AI data centres are pushing up power prices and grid costs, and in November voters across America said no. The industry says it pays its own way. Here is who really carries the cost.

Big Tech's AI Boom: Who Pays the Electricity Bill? — an essay by Bianca Sophia for HustleMemo.
Cover: HustleMemo
In this article9 sections · 0% read

Who pays for the electricity behind Big Tech's AI boom? Increasingly, the honest answer is: everyone on the grid. AI data centres are driving up wholesale power prices and the cost of reserving future capacity, and in shared grids those costs flow into household bills. That is why, in November 2026, voters in town after town across the United States said no to new data centres, and why a technology that promises to make life cheaper has become one of the most expensive political liabilities in American business.

The companies building these campuses insist they pay their own way. Some of them do, in some places. But the fight is no longer really about megawatts. It is about who carries the risk, who collects the reward, and whether the communities hosting the AI economy were ever asked.

In brief

Key takeaways

  • Demand is real and rising fast. The International Energy Agency projects global data-centre electricity use will more than double to about 945 TWh by 2030, slightly more than Japan uses today, with AI-optimised facilities growing fastest.
  • The bill is already arriving. A Federal Reserve Bank of Dallas working paper estimates data centres had raised US wholesale power prices by 6.9% nationally by 2025, and 15% in northern markets.
  • PJM shows the mechanism. The capacity price in the grid serving 65 million people from Illinois to Virginia jumped from $28.92 to $333.44 per megawatt-day in three auctions; PJM attributes nearly all of the latest demand increase to data centres.
  • Voters have noticed. Across about 40 local data-centre ballot measures in November 2026, opponents won in all but one.
  • There is another side. In Loudoun County, Virginia, data centres paid $1.2 billion in property taxes in fiscal 2026, about 39% of the county budget, and regulators in Ohio and Virginia now make the biggest users pay minimum bills.

1. Why are AI data centres raising electricity bills?

Electricity grids are shared. When a single new customer arrives wanting as much power as a small city, three things happen, and households feel all of them.

First, wholesale prices rise. Power is priced at the margin: when demand climbs, more expensive plants run more often, and everyone pays the higher price. The Dallas Fed paper, Processing Power, modelled every hour of every major US wholesale market and found that data centres had already raised wholesale prices by 6.9% nationally by 2025, and by 15% in northern markets. If the projects in the pipeline are built and run hard, it estimates wholesale prices could rise by about 50% by 2028; a more moderate build-out still means around 20%.

Second, grids must reserve capacity for the future. In regions with capacity markets, utilities pay generators years in advance to be available at peak. When forecast peak demand jumps, so does that price, and it is passed through to every customer's bill.

Third, someone has to pay for the wires. New substations, transmission lines and grid upgrades are traditionally spread across all customers. When the reason for building them is a handful of hyperscale campuses, households are effectively co-financing the AI build-out.

None of this is hidden. The US Energy Information Administration expects the average residential electricity price to reach about 18.2 cents per kilowatt-hour in 2026, up nearly 5% on 2025, with the biggest increases on the East Coast, exactly where data-centre construction is most intense. Data centres are not the only reason prices are rising, but they are the newest and the fastest-growing one.

2. PJM: where the AI bill became visible

If you want to see the AI economy on a household bill, look at PJM, the grid operator that runs the power market for 13 states and Washington, DC, including Virginia's "Data Center Alley".

The price of keeping the lights on in PJMCapacity auction clearing price for the main PJM region, $ per megawatt-day, by delivery year

Prices since 2026/27 sit at or near a regulator-approved cap. PJM says nearly all of the demand increase in the 2027/28 forecast came from data centres.

Source: PJM Interconnection, Reliability Pricing Model auction results

View the data
2024/25$28.92
2025/26$269.92
2026/27$329.17
2027/28$333.44
2028/29$325.00

In the space of three auctions, the price of reserving capacity rose more than elevenfold. The December 2025 auction, for 2027/28, cleared at the cap of $333.44 per megawatt-day, a total cost of about $16.4 billion that begins flowing into bills in June 2027. PJM said the forecast peak for that year had risen by about 5,250 megawatts compared with the year before, and that nearly 5,100 megawatts of that increase came from data centres. Without the cap, PJM estimates the price would have been about 60% higher.

The independent market monitor for PJM has gone further, estimating that data centres accounted for about $29.4 billion, or 46%, of capacity charges across the last four auctions. Even allowing for the fact that not every customer pays the clearing price, the direction is unmistakable: a large share of the cost of keeping the grid reliable is now driven by computing, and it is being spread across homes and small businesses.

When the cost of reliability is driven by a few customers and paid by everyone, that is not a market. It is a subsidy with a power cable attached.

The AI power bill in numbers
projected global data-centre electricity use by 2030, more than double today
945 TWh
IEA, Energy and AI
rise in US wholesale power prices caused by data centres by 2025
6.9%
Federal Reserve Bank of Dallas
increase in PJM capacity prices between the 2024/25 and 2027/28 auctions
11x
PJM Interconnection
of Loudoun County, Virginia’s FY2026 budget paid by data centres
39%
Loudoun County

3. Water, land and the local bargain

Electricity is the biggest cost, but it is not the only one communities weigh.

Large data centres use water for cooling, and the numbers are no longer small. Google's 2026 environmental report shows its data centres consumed about 10.5 billion gallons of water in 2025, part of a company total of about 10.9 billion gallons, roughly a third higher than the year before. Google says it replenished 78% of its freshwater consumption in 2025 and aims for 120% by 2030. That is a serious commitment; it is also a reminder that "replenished" water is often returned somewhere other than the drought-prone county where it was used.

Then there is land and noise: hundreds of acres of farmland or forest, round-the-clock cooling fans, high-voltage lines across residential neighbourhoods, and backup diesel generators. In return, a typical hyperscale campus employs relatively few people once construction ends. For many residents, the bargain feels lopsided: the costs are local and permanent, the benefits national and abstract.

4. How did data centres become a political liability?

For most of the last decade, data centres were the kind of development politicians loved to announce: billions in investment, ribbon-cuttings and no traffic jams. That changed when electricity bills started to rise.

In the November 2026 elections, about 40 local ballot measures across 10 states asked voters about data centres: outright bans, moratoriums, limits on energy use, requirements for voter approval and advisory votes. Opponents of development won in all but one; the exception was Harwood, North Dakota, a town of about 850 people outside Fargo. Ohio had the most measures. Alhambra, California, and Edgewater, Florida, were among the places that moved to prohibit data centres outright.

Before the vote, the research group Data Center Watch counted at least 45 projects worth about $68 billion disrupted by local opposition in the second quarter of 2026 alone. Candidates in both parties ran against data centres, and in Texas, exit polling found three-quarters of voters opposed building one in their own community.

The interesting detail is that AI itself barely registered as a voting issue in post-election surveys. Electricity bills did. Voters were not rejecting artificial intelligence. They were rejecting the idea that they should pay for it.

For a sense of how much is riding on that bet, see Bianca's earlier piece on the Jensen Huang Nvidia scandal and the fight over who controls AI. The chips are only half the AI story; the other half is the power plant behind every one of them.

5. The case for data centres, fairly stated

It would be easy to write this as a morality tale. It isn't one, and the industry has real arguments.

They pay enormous taxes. Loudoun County, Virginia, the world's densest data-centre market, says data centres generated $1.2 billion in real and personal property tax revenue in fiscal 2026, about 39% of its entire budget, while occupying around 4% of its commercial land. That revenue funds schools and has let the county hold its residential property tax rate steady. For many rural counties, a single campus can transform the local budget.

Regulators are making them pay more of their own way. In July 2025, Ohio's utility regulator approved an AEP Ohio tariff requiring new data centres of 25 MW or more to pay for at least 85% of the capacity they reserve for up to 12 years, whether or not they use it. In November 2025, Virginia's State Corporation Commission created a separate rate class, GS-5, for customers of 25 MW or more, with minimum-demand charges, 14-year contracts and exit fees, taking effect in 2027. These rules are designed precisely so that households are not left holding stranded costs if the AI boom cools.

Some companies bring their own power. Hyperscalers have signed long-term deals for nuclear, wind, solar and storage, and some are paying to restart or extend existing plants. Where those deals add new supply rather than buying up existing output, they can ease pressure on the grid.

Efficiency keeps improving. Each generation of chips does more computing per watt, and the IEA's own analysis notes that data centres will account for only around a tenth of global electricity demand growth to 2030, less than air conditioning, industrial motors or electric vehicles.

All of that is true. None of it fully answers the question of who pays in the meantime.

6. What a fair deal would look like

I don't think the answer is to ban data centres. AI is going to need computing power, and some of it should be built close to the people and businesses who use it. But the current settlement, where the profits are private and the grid costs are socialised, cannot last politically, and November proved it.

A fair deal would rest on a few principles:

  • Pay for what you cause. Large loads should carry the cost of the capacity, transmission and grid upgrades they trigger, through tariffs like Ohio's and Virginia's, not through averages spread across household bills.
  • Bring new supply, not just contracts. Credit should go to companies that add generation and storage to the grid, not to those that simply buy existing power and push everyone else to the back of the queue.
  • Be honest about water. Report water use by site and by watershed, not only as global replenishment totals.
  • Share the upside locally. Community benefit agreements, local hiring commitments and tax arrangements that visibly lower residents' bills would do more to win consent than any glossy announcement.
  • Ask first. The places that voted no were often places that felt projects had been agreed before residents knew they existed.

Big Tech has spent the past two years telling investors that AI is the most important technology of our time. If that is true, it can afford to pay its own electricity bill.

Conclusion

The question that remains

The AI boom was sold as an engine of abundance: cheaper services, faster science, more productive work. For the households now paying more for power, it has so far delivered the opposite. Until the companies racing to build the AI economy can show the people who live next to it that they are not paying for it, every new data centre will carry a political price as well as an electrical one.

The question is not whether we build the infrastructure for AI. It is whether the people who profit from it will pay for it.

Questions readers ask

Frequently asked questions

Do AI data centres raise electricity bills? Yes, in shared grids they can. A 2026 Federal Reserve Bank of Dallas working paper estimates data centres raised US wholesale electricity prices by 6.9% nationally by 2025 and by 15% in northern markets. Higher wholesale and capacity costs are passed through to household bills over time.

How much electricity do data centres use? The International Energy Agency projects global data-centre electricity consumption will more than double to about 945 TWh by 2030, slightly more than Japan's total electricity use today, with demand from AI-optimised data centres growing fastest.

Why did PJM capacity prices rise so much? Forecast peak demand jumped, driven mostly by data centres. PJM's capacity price rose from $28.92 per megawatt-day for 2024/25 to the capped $333.44 for 2027/28, and PJM said nearly 5,100 of the 5,250 megawatts of added forecast demand came from data centres.

How much water does an AI data centre use? It varies widely by cooling design and climate. Google, one of the few companies to publish detailed figures, reported its data centres consumed about 10.5 billion gallons of water in 2025.

Why are communities voting against data centres? Mostly because of electricity bills, along with water use, land, noise and environmental concerns. In the November 2026 US elections, opponents won all but one of about 40 local data-centre ballot measures.

Do data centres pay their fair share? Increasingly, regulators are making them. Ohio and Virginia now require large data centres to pay minimum demand charges under long-term contracts, and data centres pay large property taxes, about 39% of Loudoun County's budget. Critics argue grid and capacity costs still fall partly on households.

Sources

Sources

Editor's note: This is an opinion piece; views are the author's own. Figures are drawn from the primary sources and reports listed above. Estimates of how much of any household's bill is caused by data centres vary by region, utility and method.

Share this article

About the company: Amazon →, Google (Alphabet) →, Nvidia →