Americans Have Turned Against AI

Kind: captions Language: en Millions of Americans will open chat GPT in the next 10 minutes and most of them believe it's making the world worse. Half the country uses AI now. Yet only 16% of Americans think it'll be good for society. It is the lowest level of optimism any major technology has launched with. The people building AI call it humanity's greatest breakthrough. The people using it describe something closer to a hostage situation. So, how did the fastest adopted technology of our lifetime also become one of the most hated? You might expect that it's the older generations that are pushing back on the new AI revolution. After all, every new technology has faced resistance from people who didn't grow up with it. But that's not what's happening. The biggest users of AI are Americans under 30. They use it for school work, job applications, coding, brainstorming, and as a substitute for Google. By every indicator, they should be AI's biggest fans. Instead, they are some of its biggest critics. A Gallup study found the share of Gen Z who say AI makes them feel angry climbed from 22% to 31% by early 2026. And according to Pew Research, 40% of young Americans said it would make the country worse than better. But this isn't just a Gen Z story. Trust in AI is collapsing all across the US. Twothirds of people think it's being developed too quickly. 59% don't trust the companies building it. And most don't believe the government can keep it under control. Yet people keep using it. Chat GPT alone is used by 44% of American adults, more than double its share in 2023. That is the AI paradox. So why does adoption keep accelerating if confidence keeps falling? For most people, it wasn't really adopted. Workplaces began implementing it in their workflow. Schools added it to assignments. Apps and search engines displayed it as a main feature. People had no other choice. It was forced upon them. AI didn't earn their trust. They have no other option. That shift happened so gradually that most people barely noticed it until suddenly it was everywhere. The moment you open Google and you search for something, you're hit with a block of AI written text answering the question before you even reach the search results. The company that's supposed to help you find information is now deciding what you see first. And it works. 60% of Americans say they read the summaries regularly because there's no way to avoid them. The moment Google started answering the question itself, the experience changed. And the more the internet degraded, the less people trusted it. In 2021, 37% of Americans said the spread of AI worried them more than it excited them. By 2025, that number had climbed to 50%. Even though most people hadn't been personally affected by AI in a professional capacity. That would soon change. In April 2025, Toby Lutkkey, Shopify's chief executive, posted a memo to his staff. Before any team could ask for a bigger budget or another employee, they had to prove that a machine couldn't do the work first. Lucky insisted that reflexive AI use was the new baseline expectation. Shopify wasn't the only company going allin on AI. Language learning platform Duolingo was next. CEO Lewis Vonan told his staff the company was going AI first. It would rely on AI before hiring, reduce the number of contractors, and weigh AI use in performance reviews. Within a week, 148 new courses were shipped. It was work that would have taken staff years to develop. The same AI implementation turned up at Box, at Fiverr, and at other firms. The wording barely changed. Run the software before you ask for new hires or extra money. Then the push back happened. Longtime Dualingo users filmed themselves deleting the app in protest. Staff started asking if they had to use AI well or just use it. Almost a year later, Luis Vonan walked his decision back. He acknowledged that the AI written code was hard to debug and the lessons weren't reliable. But the damage had been done and it was just the beginning. The fear of AI taking jobs is real. So in the summer of 2025, a Stanford economist named Eric Brinolson decided to look into it. His team at the Digital Economy Lab pulled the records from ADP, the firm that processes the payroll for millions of American workers, and studied who still has a job month after month. The AI impact was real, just not where people expected it. The results showed that workers aged 22 to 25 in software and customer support had dropped about 13% since late 2022. Workers over 30 in the same roles remained stable or actually grew. It was the people just walking in the door that were at risk of an AI coal. Some don't even get in the door at all. Revelio Labs, which tracks hiring across the economy, counts roughly 35% fewer entry-level postings since the start of 2023. Once employers saw what the best models could do, many stopped graduate hiring. That freeze only surfaced in the payroll data a year or two later. Bin Olson's team found that when a company used AI to take over a task completely, employment among younger workers fell. When AI was used to make a person faster, employment remained stable and sometimes grew. Same technology, opposite result. Everything comes down to one decision. Does a company use AI to replace a worker or to back one up. Near the bottom of the pay scale, replacement is winning. Salesforce froze much of its junior hiring for a stretch and credited its own AI for work graduates do. Goldman Sachs economists found the results. The big names weren't even pretending anymore. But not every bet paid off. Clara, the buy now pay later company, spent a year bragging that its chatbot did the work of hundreds of agents and that it had stopped hiring people. They later admitted that the bot gave lower quality service and started hiring humans back. But some companies didn't see that as a reason to slow down. They saw it as a reason to aim lower and the contractor was in the crosshairs. An MIT report on AI in businesses found that AI is wiping out outsourced and contract roles, not full-time staff. In mid 2025, the data labeling firm Scale AI laid off 200 staff and cut ties with 500 contractors. That same report put the share of jobs AI can already do at roughly 3% today and saw it climbing over the long run. Graduates weren't losing out to people with a better resume. They were losing out, period. There was not a job. By early 2026, half of employed Gen Z said AI's risks at work now outweigh its benefits. A jump of 11 percentage points in just one year. And when they're handed the same piece of work, 69% overwhelmingly trust the human version versus the AI one. An entire generation now spends its days training the technology that is replacing the jobs they are trying to build careers in. The lost jobs are easy enough to measure, but the resentment isn't just about money. In July 2024, all it took was a power surge across Northern Virginia for nearly 60 data centers to drop off the grid. Roughly, 1500 megawatts of demand vanished instantly. Operators scrambled to get the surge under control before it took down anything else. For a few seconds, the densest data center hub in the country just wasn't there. Virginia holds more of those buildings than anywhere else. And by 2023, they were already drawing about 26% of the state's electricity. That's why one voltage swing could shock the whole grid. Each of these buildings runs hot. One question to a chatbot burns about 2.9 W hours. That's roughly 10 times a normal web search. On its own, it's nothing. But nobody asks just one question. Multiply it by billions of questions a day every day in warehouses that never switch off and that becomes 183 terowatt hours a year. That's about 4% of all the power the country burns. That is already 540 kwatt hours running in your name every year whether you've ever touched a chatbot or not. And the forecasts are only going up as high as 17% of the nation's electricity by 2030. More than 40% of the power fueling the data centers comes from natural gas. Before the decade is out, those buildings are set to burn more electricity than the entire country uses to make its steel, cement, aluminum, and chemicals combined. A high voltage line takes up to 8 years to build, meaning a fifth of these data center projects are about to run straight into a power shortage. To cover the expense of those extra lines, utilities are going to regulators to demand more money from regular consumers. This means your monthly expenses are going up by 2030 just to keep the lights on at a server farm down the road. You didn't leave your lights on or crank up your AC, but your money is bleeding out anyway to power the massive cooling fans of a tech giant. Researchers at Carnegie Melon estimate it could push the average American power bill up by about 8% by 2030 and in the worst hit parts of Virginia north of 25%. There simply is not enough energy or money to fund it. So some companies got tired of waiting. Instead of waiting for years for the power grid to catch up, they built their own power plants beside the data centers. And one company took that idea to the extreme to power Colossus, the supercomputer training Elon Musk's Gro Chatbot. XAI deployed a fleet of gas turbines in South Memphis. The entire site was up and running in under 122 days, a project that would have normally taken years. By the time the Southern Environmental Law Center flew thermal drones overhead, it counted around 35 turbines, most of them without air permits. Just downwind sits Boxtown, a neighborhood that spent decades surrounded by refineries and chemical plants. Residents began to show up at public hearings describing a rotten egg smell in the air and children wheezing through the night. Memphis already earns an F for ozone pollution from the American Lung Association, and Boxtown's cancer risk is about four times the national average. The researchers at the University of Tennessee analyzed satellite data for time and found peak nitrogen dioxide levels near the site had jumped roughly 79% compared with before XAI arrived. The turbines were labeled as temporary, a workaround that allowed them to keep on running and bypassed the regulations. Lawsuits followed, but by the time the case moved forward, the service had already been running for a year. Power isn't the only resource these data centers consume. In Newton County, Georgia, about an hour and a half east of Atlanta, Beverly and Jeff Morris live 1,000 ft from a Meta data center. Their house runs on a private well, which means their drinking water comes straight out of the ground. It worked fine for years, right up until Meta broke ground next door in 2018. A few months after construction began, the water pressure slowed to a trickle and then it stopped altogether. When it eventually came back, it was brown and murky. Beverly Morris told the New York Times that she was afraid to drink the water coming from her own tap. The neighboring data center uses roughly 500,000 gallons every day, nearly a tenth of all the water Newton County pumps. County projections show demand outpacing supply by 2030. While residents face water bills expected to jump 33% in just two years compared with the usual 2% increase, a metaf funded study said the company's operations probably aren't to blame and has invested more than $4.5 million in local schools and nonprofits. Private wells like the Morrises fall outside the federal safe drinking water act. So Meta didn't break any laws. The Environmental Protection Agency is powerless. So Meta keeps on pulling water to cool the chips. And those hubs are insatiable. A 2023 study by UC Riverside and the University of Texas found that training one model GPT3 ran through about 5.4 million L of fresh water before the public ever even typed a word. The same researchers found that a short back and forth around 100 words consumes close to a full bottle. One Google data center in Pilion, Nebraska went through around 1 and a half billion liters in 2024, more than 600 Olympic pools, and it is one site among thousands. That daily draw dwarfs the one-time cost of training. Nationally, data centers used over 60 billion LERs in 2023, and the World Resources Institute expects AI to pass a trillion gallons a year by 2030. In August 2025, Tucson City Council voted 7 to zero to reject Project Blue, an Amazon linked data center campus that would have drained millions of gallons of city water in the middle of a desert. Residents had packed the chamber for weeks under a banner that read no desert data center. For one night, it looked like a regular city could tell a tech giant no and actually make it stick. It didn't end there. The developer already owned the land just outside of city limits and kept building anyway. Then in the spring of 2026, the contractor was caught drawing enough Tucson water to supply five households for a year. The city shut it off. Mayor Regina Romero has been writing ordinances to guard against the next data center because there's always a next one. Tucson wasn't alone. In the Dallas, Oregon, a Google plant ended up using close to a quarter of the entire town's water. The company fought to keep those numbers hidden from the public through court orders. Newer facilities can cool with air or closed loop systems that use almost no local water. Microsoft has converted two Arizona sites to do exactly that. But the biggest water users aren't the new builds. They're the older data centers, the ones still sucking up millions of gallons instead of returning it to the river. So while families shorten their showers and they wait until dark to water the lawn, the data center down the road keeps pulling water at full capacity every hour every night. jobs, electricity, water. Three separate battles happening in different parts of the country, but they all lead back to the exact same business model. In 2024 alone, Amazon, Microsoft, Google, and Meta spent more than $200 billion building data centers behind AI, up over 60% in a single year. Management consultancy firm McKenzie expects $5.2 trillion to be funneled into them by 2030. That money buys the land, the power, and the water. But there is one more thing it buys. Something nobody is signing off on. The writing, the art, and the code scraped off the open internet to train the models in the first place. To the companies, that's an investment in the platform of the century. To the creatives whose work fed it, it looks different. Their work has been harvested for free and then rented back to everyone by the month. Not one of them was asked. Not one of them got paid. Publishers and authors have sued in waves, arguing their work was stolen and used to train AI without permission. The courts still haven't decided where the legal line is. And so far, none of it has been ruled illegal. But AI is built on other people's writing, powered by other people's electricity, cooled with other people's water, and then sold back to those same people as progress. Now, Americans are starting to notice. The companies insist the costs are temporary. Google and Microsoft both promise they'll return more water than they use by 2030 and keep shifting to cleaner energy. But if you've already lost your job, your power bill just went up, or your town's water is running low, a promise four years from now doesn't feel like much of an answer. Between May 2024 in March 2025, that anger blocked or delayed about 64 billion in data center projects nationwide, according to Data Center Watch. In Virginia alone, 42 separate groups were working to slow or stop new ones. What nobody agrees on is what to do about it. No one is demanding an outright ban. No one is cheering it on. There's just a country caught between needing the tool and resenting it. Most of these fights end the same way. A council denies the water the way Tucson did. The county next door sells the land. A state commission signs off on the power. The building goes up anyway, trading its thirst for water for a heavier pull on the grid. Residents keep learning the same lesson. The project can almost always route around them. This was never about whether AI would arrive. That was inevitable. It was about whether anyone still had the power to stop it. By the time we started asking that question, the answer was already no. It's not just ordinary Americans who are losing faith in AI. The people building it are too. Watch why AI researchers are quitting and panicking on the way out to find out why the industry is collapsing. Or click on this video.