Average People Feel AI First on the Power Bill

49% of U.S. adults now use AI chatbots, yet the impact most households can name is a higher power bill, not a new appliance. Nathan Lambert, who writes the Interconnects newsletter, argued in a Sept. 9, 2026 essay that family life, food, transport, and entertainment still look much the same, and that the boom still serves desk work first.

He compared the moment to earlier industrial revolutions, when cheaper cloth, indoor plumbing, and bicycles showed up in ordinary homes. In surveys and utility dockets, the first round of AI’s impact is already here as chatbot use, schoolwork, and electricity demand.

Half of Adults Already Open a Chatbot

Pew Research Center surveyed 5,119 U.S. adults from Feb. 17 to 23, 2026, and found that about half of U.S. adults now use tools such as ChatGPT, Gemini, or Copilot. That share was 33% in 2024. 24% say they use chatbots daily, including 12% several times a day and 4% almost constantly. Another 25% use them several times a week or less. 51% still do not use them at all.

Search is the common door. 42% of adults use chatbots to look up information. Among people with a job, 38% use them for work tasks. Fun, image editing, and medical questions trail that. ChatGPT remains the default: 44% of adults say they use it, up from 18% in 2023. Gemini is at 24%, Copilot at 17%, and Meta AI at 14%. Claude sits at 6%.

HOW ADULTS USE CHATBOTS

Those are software habits, not new stoves. About a third of adults say they own a smart speaker, and smaller shares have a doorbell or thermostat with AI features. Pew also found that more adults expect a negative effect on their lives and on society than a positive one, and that majorities think the tools are moving too fast. People can use the product daily and still not credit it as a gift.

The Bill Arrives Before the Robot

The U.S. Energy Information Administration said on Jan. 13, 2026, that electricity use would rise 1% in 2026 and 3% in 2027. That would be the first four-year run of higher demand since 2007, and the strongest four-year growth since 2000. Tristan Abbey, the agency’s administrator, said demand rises through 2027, “driven largely by increasing demand from large computing facilities, including data centers.”

Actual sales already show the commercial side catching homes. EIA’s Electric Power Monthly put 2025 U.S. sales at 4,058,007 thousand megawatthours, with residential customers at 1,514,993 thousand and commercial customers at 1,493,486 thousand. Through June 2026, commercial sales had moved ahead on a year-to-date basis, 738,707 thousand megawatthours against 722,613 thousand for homes.

WHAT THE GRID IS BEING ASKED TO CARRY

• 2024 load: A Department of Energy study, cited in a Sept. 1, 2026 Congressional Research Service briefing, put U.S. data centers at about 4.7% of electricity use, or 192 terawatthours.
• 2030 range: Lawrence Berkeley National Laboratory’s 2025 update, in the same briefing, puts data centers at 9.5% to 15.3% of U.S. electricity by 2030, or 521 to 843 terawatthours.
• Inside the building: CRS says running the computers takes roughly half the power in a typical hall, and cooling takes about 40%.
• The market: PJM Interconnection, the country’s largest grid operator, serves more than 65 million people in 13 states and Washington, D.C., and is the region where the data-center fight is loudest.

The causal share of any one household bill is still argued. CRS notes that Berkeley Lab found states with the largest data-center growth often saw prices fall from 2019 through 2025, even as other models warn of later increases and PJM capacity costs have become a political fact. Congress has spent the 119th session on a narrower question: whether homes pick up the tab for plants and power lines built to serve new computing load. That is a form of AI’s impact that does not require anyone to open an app.

Ohio Puts Data Centers on Their Own Tab

On Aug. 5, 2026, the Public Utilities Commission of Ohio ordered AEP Ohio to protect customers from data center costs if a large computing customer returns to default service. The order requires 180 days’ notice and a separate power purchase, with the full cost assigned to that customer rather than to other households on the default rate.

We continue to implement safeguards to ensure that other customers are not impacted by costs to serve large loads like data centers.

Jenifer French, Chair, Public Utilities Commission of Ohio, Aug. 5, 2026

Virginia has approved a separate rate class for large loads, set to take effect in 2027, for customers at or above 25 megawatts. CRS, writing for Congress, said local opposition to new data centers has been growing, with the flash point being new plants and lines and whether other customers pay for them. Hyperscale halls, in that briefing, start at about 5,000 servers and 10,000 square feet and can run much larger.

Ohio’s governor’s race has already absorbed that grievance. Vivek Ramaswamy, a Republican candidate, wrote on Aug. 6, 2026, that data-center growth was a top concern he heard across the state, “second only to property taxes,” and that the industry had “badly failed to earn the trust of millions of everyday Ohioans who are struggling with electric bills and property taxes.” He pledged that if a center is built, nearby homes would no longer pay for household power, and he claimed 100 megawatts could cover 75,000 to 100,000 homes. That math is a campaign offer, not a tariff. The constituency it aims at is real: people who never asked for a chatbot and still see a computing campus on the same wires.

National arguments about model intelligence miss that local register. Neighbors talk about a bill that ticks up, water on hot days, and construction on the school run. They do not talk about training runs.

Chatbots Landed in Classrooms First

Kitchens have not changed. Schools have. Pew surveyed 1,458 U.S. teens from Sept. 25 to Oct. 9, 2025, and found that 64% use AI chatbots. 54% have used them for schoolwork, and 57% have used them to search for information. ChatGPT is the teen default at 59%, more than double Gemini at 23%.

WHAT TEENS ALREADY DO WITH THE TOOLS

• Daily habit: About three in ten teens use chatbots every day, including 16% who do so several times a day or almost constantly.
• Heavy school use: One in ten teens say they do all or most of their schoolwork with chatbot help; 21% say some, and 23% say a little.
• Cheating, as teens see it: About six in ten say students at their school use chatbots to cheat at least somewhat often, including about a third who say it happens extremely or very often.
• Income split: One in five teens in households earning less than $30,000 a year say they do all or most of their schoolwork with chatbot help, against 7% in higher-earning homes.

That is a direct hit on family routine that does not look like a sewing machine. Parents meet the technology as a homework policy, a cheating dispute, or a kid who now writes first drafts in a chat window. Lambert wrote that many people have heard of lab incidents they cannot parse, and that the positive case still sounds like fun images and a slightly better search box. Teen surveys say the classroom got there first.

British Wages Lagged for Fifty Years

Lambert reached for Engels’ pause, the label economic historian Robert C. Allen gave to 1790 to 1840 in Britain, when working-class wages stalled while output per person rose. Allen’s 2009 paper, drawing on Crafts-Harley output figures and Feinstein’s wage index, is the version that now travels with the AI debate.

THE WAGE GAP ALLEN MEASURED

Allen also found that the profit rate doubled in the first stretch, and that capital’s share of national income rose at the expense of labor and land. After the middle of the nineteenth century, wages grew with productivity. Cheaper cloth, cookware, and reading material were already in shops during the pause. Tangible goods did not, on their own, lift pay. The later catch-up ran through capital deepening, politics, and bargaining, not through the arrival of one more household machine.

That undercuts the hope that robots in the driveway will settle the distribution fight. If the analogue holds, software can get cheaper, search can get better, and a slice of desk work can get faster while a large share of pay still lags. Lambert is right that a Navier-Stokes prize would not rearrange an American evening. He is on weaker ground when the test for “feeling it” is a new physical good. Britain had those goods. Wages still waited.

Who Uses the Tools at Work?

Lambert called knowledge work “roughly half of the U.S. economy” and said AI is becoming as basic as electricity for that half. Headcount is a colder measure. Occupations that typically need a bachelor’s degree accounted for a quarter of U.S. jobs in 2025, or 42,870.1 thousand out of 170,280.8 thousand, according to the Bureau of Labor Statistics. Add master’s and doctoral-typical roles and the share is 30.2%. High school-typical jobs were 35.4%, and jobs with no formal credential were 23.6%.

Pew’s workplace numbers follow that contour. In September 2025, 21% of U.S. workers said at least some of their work was done with AI, up from 16% a year earlier, and 65% said they used it little or not at all. Among workers with a bachelor’s degree or more, the share rose from 20% to 28%. Among those with some college or less, it rose from 13% to 16%. By February 2026, 38% of employed adults said they used chatbots for tasks at work. The tool concentrated where the degrees already were.

That split is the political problem Lambert named, even if the “half of the economy” line overstates the payroll. A product that raises output in offices, while homes meet it as a bill and a school rule, will not look like electrification. It will look like a club. In a later Pew reading, more Americans still expected harm than help. The users and the skeptics are often the same people.

The Fight Is Over Who Pays

Today’s AI is primarily a tool to serve the elite.

Nathan Lambert, Interconnects, Sept. 9, 2026

Lambert’s warning is that patience will run out, and that a nuclear-style freeze could arrive before the compounding benefits do. He wants the industry to stop selling doom, to keep training, and to wait for robotics and self-driving to make the gains visible. Self-driving, as he notes, has been on its own track for years. Robots in daily life are still a bet.

The sequence on the ground is already different. ChatGPT went public in 2022. By early 2026, 49% of adults had used a chatbot and 24% were in it daily. Teens had pulled the tools into homework. Data centers were 4.7% of U.S. electricity in 2024, with a path toward as much as 15.3% by 2030. Ohio’s regulator moved in August 2026 to keep large-load costs off other customers. Virginia built a rate class for 2027. A gubernatorial candidate in Ohio offered free household power as the price of a new campus.

Average people do not need a new washing machine to enter this story. They need a meter, a teenager, or a job that already runs through a screen. The buildout’s next constraint is not whether the models get smarter. It is whether the people who pay for the plants believe the tab is theirs.