Why Everyone Hates AI (And Why We Can’t Afford to Stop Building It)

Not long ago, artificial intelligence was a bogeyman reserved for anxious Silicon Valley techies. Now, the fear has gone mainstream. Across the West, voters are angry, and AI is climbing the political agenda. While Britain’s flimsy prime-minister-in-waiting, Andy Burnham, has barely said a word about it, and Americans still rank it just 29^th out of 39 election issues, the mood is turning sour fast.

In the United States, grassroots protests against massive data centres have already scuppered nearly $100bn worth of tech projects. In a recent Manhattan congressional race, warring AI megadonors dumped tens of millions into the campaigns. Why? Because around 40% of American voters now tell pollsters they want AI banned from most industries. The anger isn’t confined to the West, either. After chipmaking profits soared in South Korea, workers at Samsung threatened a strike just to secure special AI payouts.

This backlash is only getting started, mostly because the technology is only getting started, too. And the bitter fights over data centres offer a hint of the struggles to come.

If you stood atop the backyard pool slide of a home in rural Ohio last April, you would have looked out over lush farmland, dense forest, and pretty clapboard houses. Today, your view is blocked by six giant, weatherproof military tents (the kind used to house fighter jets in war zones). Very soon, those tents will hold around $30bn worth of cutting-edge semiconductors. Owned by Meta and powered by a noisy cluster of gas turbines, this “Prometheus” site covers an area the size of an airport terminal. When it comes online this year, it will chew through an entire gigawatt (GW) of electricity. That is enough to power one million homes, roughly matching the output of a large nuclear reactor.

The mammoth data centres needed to train the frontier AI models of 2030 will not be tucked away in traditional tech hubs like Virginia or California. Instead, they are invading the “Silicon Heartland” of Michigan, Wisconsin, and Ohio, as well as southern states like Louisiana, Mississippi, and Texas. Tech giants (including Amazon, Google, Meta, Microsoft, and Oracle) are pouring up to $750bn into these places. Between 2026 and 2030, an estimated $3trn will be spent on AI data centres globally, with much of it earmarked for American soil. To support this, US AI computing capacity will skyrocket from just under 12GW today to nearly 60GW by the end of the decade.

Across the country, residents of all political stripes are furious. They hate the ugly boxy buildings, the constant roar of generators and cooling fans, the skeleton army of new transmission towers cutting across the landscape, and the lingering fear of contaminated water. Astonishingly, surveys show Americans would rather live next door to a nuclear power plant than a data centre. Even a plan to build one in the remote Utah desert met with fierce local resistance.

This pushback has real teeth. In the first three months of 2026 alone, local activists forced the cancellation of at least 20 data-centre projects worth $42bn, which would have consumed 3.5GW of power. Over the past three years, $85bn worth of projects have been scrapped, including smaller sites planned by Amazon and Meta. In Cedar Rapids, Iowa, residents are fighting off Google. In Michigan, several townships passed emergency construction bans after OpenAI broke ground on a site in Saline despite bitter local opposition.

Yet this resistance goes deeper than simply NIMBYism (Not In My Back Yard). An April survey by Pew Research found that Americans who live hundreds of miles away and have merely heard of data centres oppose them just as fiercely as people living without five miles of one. Why? Because the tech bosses themselves spent years screaming that an AI job-pocalypse was coming, or warning that a rogue AI might engineer a super-virus and make humans extinct. Philosophers have long worried about a machine given a careless objective function to maximise, like making paperclips, hoovering up all the Earth’s resources and paving the planet with servers just to maximise the production of paperclips. OpenAI’s Sam Altman and Anthropic’s Dario Amodei warned Congress of mass harm. Now, the physical machinery required for that future is turning up on suburban doorsteps looking like a military invasion, and residents are begging local councils to axe the projects.

The industry is sweating. “We need to stay a way ahead of China,” says Chris Wright, America’s energy secretary. He calls leading in AI the overriding goal of his tenure, adding, “We’ve got to enable these data centres to get permitted and built and turned on power.”

Right now, roughly 1 to 2GW of US data-centre capacity is dedicated to training frontier models by big players like Anthropic, OpenAI, Google, Meta, and xAI. Another 10GW is used for “inference,” the day-to-day power needed when customers ask chatbots questions or write code. But when demand soared in early 2026, the available computing oomph feel woefully short. Anthropic had to throttle model usage, OpenAI scrapped a power-hungry video tool, and Microsoft hiked the price of its coding assistant so sharply that some developers went back to writing software by hand.

While big projects underway should add 30GW of capacity by 2028, the power needed to build new models is growing even faster. Anthropic estimated last year that training a single frontier model would require 5GW by 2028. Research firm Epoch AI predicts that figure could hit 16GW by 2030. If true, almost all the new capacity built in the next few years will be swallowed up by training alone.

Some factors might ease the crunch. Chips are getting more efficient, old crypto-mining rigs are being turned into AI labs, and chipmaker bosses like Cerebras’s Andrew Feldman note that daily inference doesn’t need massive, single-site facilities. Yet it might not be enough. So far, only computer coding has been truly shaken up by AI. Law, finance, and media are still barely out of the starting blocks.

Many sites currently being built, like OpenAI’s project in Saline, Michigan, only survived because they were approved before the public lost its temper. The Saline project was voted down by the local council and only squeezed through because the country lacked industrial zoning laws, creating a legal loophole. Even regions that used to love tech build-outs are shutting the doors. In March 2025, Loudoun County, Virginia, dubbed “data-centre alley,” scrapped rules that made building easy, forcing new developers to face public hearing. In Texas, the city of San Marcos passed a flat-out ban on new centres.

Is the grid about to melt down or run out of water?

Much of the local anger is fuelled by genuine concern for the neighbourhood, though it is often mixed with bad fiction. Take the viral rumour that AI data centres are drinking towns dry. This myth took off in 2025 due to a popular book that relied on a massive mathematical error. In reality, a mid-sized data centre uses about as much water annually as two golf courses. If it uses modern water-recycling technology, it drinks much less. Across the whole of America, data centres consume a tiny fraction of the water sprayed onto golf courses.

Electricity, however, is a real problem. Research firm SemiAnalysis calculates there is around one terawatt (1,000GW) of requests waiting to connect to American power grids, almost all from data centres. That equals the entire generating capacity of the US electrical grid, which maxes out at 1,250GW. Normally, US power demand averages 470GW, peaking at 750GW during hot summers. Utility companies like to keep a 15 to 20% safety buffer above that peak, and only about 975GW of power is reliably available on demand.

Voters worry this massive tech demand will cause household electricity bills to spike. So far, there is little hard evidence of this happening. Growing demand actually lets utility companies spread the cost of grid upgrades across more customers. Futhermore, tech operators install massive backup battery systems. As Alistair Speirs, who leads data-centre builds at Microsoft, explains, these batteries let companies “choose when to sip and when to slurp from the grid,” pulling back during severe storms so regular homes keep their lights on.

Even so, America needs to generate vastly more electricity. The Department of Energy expects the country will need an extra 50GW of power for AI by 2030, plus another 50GW for a general manufacturing revival. Energy Secretary Wright is sceptical of wind and solar power because they only work intermittently. To keep the AI boom alive, he has stopped coal-fired power stations from closing, pushed to restart old nuclear plants, and backed new gas-fired power stations. While over a third of data centres plan to generate their own on-site power by 2030, putting gas and nuclear plants next to server farms will only make them a bigger local eyesore.

Ohio has been cleverer than most states at calming ratepayer panics. In July 2025, the state utility commission ruled that large data-centre operators must pay for at least 85% of the power capacity they request each month, even if they never use it. This guarantees that local Ohioans will not get stuck paying the bill for grid upgrades. The idea was so good it was copied in the White House’s “ratepayer protection pledge” signed by tech giants in March, though Ohio’s law is actually binding.

Yet, even clever rules haven’t calmed the hillbilly uprising in the Buckeye State. Three-quarters of Democrats and two-thirds of Republicans in Ohio oppose local data centres. Donald Trump won the state by 11 points in 2024, but Vivek Ramaswamy, a vocal AI enthusiast running for governor, is currently neck and neck in the polls with his Democratic rival, dragged down by voter rage over server farms.

To bypass local zoning headaches, the Trump administration is getting creative. In March, the Department of Energy announced a massive 10GW project on federal land in Piketon, rural Ohio. It will be funded by Japanese conglomerate SoftBank, led by Masayoshi Son, who plans to build a gas power plant to run what will be the world’s largest data centre. The site was once used for America’s 1950s nuclear enrichment programme, so locals are hardly squeamish. State representative Adam Holmes laughed about the groundbreaking ceremony: “Imagine being in an Appalachian farmer’s field with mud on our shoes with Secretary Lutnick, Secretary Wright, Mr Son and me… and all the other hillbillies!”

Despite the laughs, local politicians are having a miserable time on the doorsteps. State senator Shane Wilkin tried to reassure his Ohio constituents using hard facts from a water utility expert. The expert testified that only one data centre in the state even had a water discharge permit, and its only two violations were for late paperwork. Wilkin tells voters that the centres bring their own electricity and don’t dump dirty water. But the voters don’t care about data. They simply fold their arms and say: “Well, I just don’t want it.”

How did AI become the only thing politicians agree to hate?

“Most Americans already know AI could be transformative and dangerous at the same time,” says Alex Bores, a computer scientist and New York state assemblyman. “What they want to know is whether anyone in government is actually going to do something about it.”

Bores made strict AI regulation the core of his campaign for Congress in a crowded New York Democratic primary on June 23^rd. He had co-authored a state law forcing AI firms to report safety incidents and ban models that posed severe risks. He wanted to make it national law. The tech industry went to war over him. One group of tech titans spent millions trying to destroy his campaign, while a rival group of AI firms spent millions backing him, agreeing that national safety guardrails were essential. In the end, Bores narrowly lost to Micah Lasher, a traditional candidate who also promised to stand up to Silicon Valley.

This mess shows just how explosive AI has become. Inside the tech industry, factions are at war. Donald Trump’s adviser, David Sacks, wants to develop AI as fast as humanly possible to stay ahead of China. Others plead for caution. Yet among ordinary voters, there is zero division: everyone is terrified.

Three-quarters of Americans tell YouGov they want stricter AI regulation, with Republicans just as eager as Democrats. A Pew Research poll shows 66% think the technology is moving too quickly, compared to just 2% who think it is moving too slowly. A staggering 45% of Americans genuinely worry AI will cause the extinction of the human race. Roughly 65% believe it will destroy jobs (including 74% of Democrats and 56% of Republicans). Over 70% fret that it will wreck mental health and concentrate wealth in the hands of a few tech billionaires.

American gloom is unique. A 24-country study by Pew found US citizens were by far the most pessimistic in the world. Recent polling shows 61% expect AI to affect society “a lot” and they believe the negative impacts will outweigh the positive ones by 46% to 31%.

For the progressive left, AI is a toxic brew. They fear it will kill unionised jobs, they hate the local disruption of data centres, and they loathe the billionaires funding it, especially the trillionaire who pals around with Donald Trump and wants to put servers in space. Malcolm Kenyatta, vice-chair of the Democratic National Committee, calls today’s tech barons “far worse than the railroad or coal barons of the ‘Great Gatsby’ age.” Igor Volsky of the pressure group Tax the Greedy Billionaires complains that “a small group of very rich, mostly white men” are dictating the future of society.

The left also worries about human connection. Maurice Mitchell, leader of the Working Families Party (which backs socialists like Zohran Mamdani), worries about his ten-year-old child forming an emotional relationship with a chatbot. He notes that without rules, AI apps are being marketed as replacement for real friends and professional therapists.

Strangely, the right echoes these exact fears. Republican senator Josh Hawley is deeply alarmed, predicting AI could wipe out half of all entry-level white-collar jobs before coming for manual factory workers. He told the Teamsters union on June 16^th that mega-corporations have “lost their moral compass.” Hawley thinks using AI to make lorry-driving safer is fine, but he wants to outright ban self-driving lorries. If AI makes a few people rich while throwing everyone else on the scrapheap, Hawley warns,” I don’t think you can sustain a democracy.”

Hawley also frets about young people getting addicted to screens and shunning marriage and parenthood. Grounded in his Protestant faith, he fears tech plutocrats will use AI to become superhumans or try to “live infinitely in the cloud.” His disgust matches the views of the Pope, who recently warned that “transhumanism” could trick the elite into believe ordinary human lives are less worthy.

In statehouses across America, this bipartisan dread has birthed bills in at least ten states to freeze data-centre construction. Brendan Steinhauser, a Republican strategist heading the Alliance for Secure AI lobby group, says the old accelerationist “let them cook” mindset is dying. “People want to slow it down and get it right,” he says.

The trouble is that no one knows what “getting it right” means. Politics moves at a glacial pace compared to software; a new model is trained in the time it take Congress to schedule a single committee hearing. Dozens of bills have been drafted, some hundreds of pages long, but nothing major has passed.

The White House has been a mess of contradictions. Last year, the Trump administration used legislative tricks and executive orders to block state-level AI rules. In March, it released an AI framework that was wonderfully permissive. Then, in June, Trump panicked and banned all foreign nations from accessing the most powerful US AI models, forcing Anthropic to close its top systems to everyone outside America.

To deal with long-term threats like mass unemployment, Congress and the President must find common ground. Economic redistribution might be the unexpected bridge. Left-wing senator Bernie Sanders wants the government to seize a 50% stake in major AI firms and hand the dividends to the public. Trump has also said the public deserves a financial share in AI wealth, though his plans are vague. Giving voters a direct cash stake in tech profits would be wildly popular across party lines.

How to distribute that cash is another fight entirely. Many on the left and in Silicon Valley love Universal Basic Income (UBI). The right hates it. “I hate universal basic income,” says Senator Hawley, arguing that paying people not to work destroys their personal independence. As the midterms loom, voters are terrified of a topic neither party knows how to handle.

While the West argues, is China quietly winning the race?

If America decides to regulate AI into a standstill, Beijing is ready to take the crown. America’s lead over China in AI is currently at its narrowest point in over a year.

We saw a preview of this shock in January 2025, when a Chinese lab released DeepSeek r1. It instantly wiped $1trn off US stock markets. Chipmaker Nvidia shed 17% of its value in hours, and the Nasdaq sank 3.1%. Investors panicked not just because the Chinese model was superb, but because Beijing was giving it away for free.

Now, Chinese labs are causing fresh panic. On June 13^th, 2026, a Beijing lab named Zhipu (also known as Z.ai) released its newest model, glm 5.2. It promised “a step closer to frontier intelligence for everyone.” It’s the most powerful Chinese model ever built, it costs less than a tenth of Anthropic’s new Fable 5 model to run, and Zhipu has publicly released the underlying code and weights. China is now competing on brainpower, bargain prices, and total openness.

The timing was on point. American businesses have been screaming about soaring AI software costs, which often run into thousands of dollars per employee each month. Firms are rationing tokens to save cash. Then, on June 12^th, the day before Zhipu’s launch, the Trump administration banned non-Americans from using Anthropic’s Fable 5. For the first time, access to top-tier AI depended on the whim of a narcissistic US president. Global businesses immediately began scouting for non-American alternatives, and glm 5.2 looked like a cheap, powerful lifeline beyond Trump’s reach.

How smart is it? Research from Artificial Analysis ranked glm 5.2 as the world’s most intelligent open-source model. On their global leaderboards, it took fourth place overall, trailing OpenAI’s ChatGPT 5.5, but beating Google’s Gemini. Elon Musk posted on X that he expects China to match America’s best models by early next year. Zhipu co-founder Tang Jie confidently replied: it “won’t take that long.”

Despite the hype, American markets didn’t crash this time. Why? Because grading Chinese AI has become a tricky business. Artificial Analysis tests models using public exam questions. On those standard tests, America’s Fable 5 is still roughly 17% smarter than glm 5.2, and the Chinese model performs at a level the West reached about four months ago.

However, Havard Tveit Ihle of the Norwegian Defence Research Establishment notes that Chinese labs often “teach to the test,” optimising their software for published exam questions. When tested on private benchmarks where the questions are kept secret, America’s lead doubled to roughly eight to ten months. A US government study in May confirmed this gap. On a secret test called Weirdml (which requires deep logical reasoning), glm 5.2 lagged seven months behind. On SimpleBench (a common-sense test designed to trick computers), it fell a full year behind.

Yet China is catching up in specific areas. On June 19^th, Artificial Analysis released a brand-new exam testing how well AI handles messy office tasks, like sorting conflicting files. Because the test was brand new, Zhipu couldn’t have cheated. Yet glm 5.2 beat OpenAI’s two-month-old ChatGPT 5.5. The results prove America still holds the lead, but the gap is no longer growing.

Chinese software has a distinct personality. It excels at rigid subjects like maths and coding, but stumbles on tasks requiring open-ended judgment. This happens because US export bans on advanced semiconductors have starved Chinese labs of the raw computing power needed to train massive systems from scratch. To survive, Chinese engineers are masters of “post-training,” fine-tuning smaller models, sometimes using data harvested from American software through a technique called distillation.

What about that bargain price tag? DeepSeek v4 charges just $0.87 per million output tokens, compared to a whopping $50 for Anthropic’s Fable 5. In June, invoicing firm Ramp noticed a sharp spike in US companies buying DeepSeek services, and Microsoft is reportedly considering using Chinese code in its Copilot software.

But the cheap price is often an illusion. Because Chinese models lack raw reasoning power, they’re more inefficient when solving a problem. A study by Georgia Tech researcher Du Zheng showed that DeepSeek required 23 times more tokens than OpenAI to finish the exact same coding task. When you measure the total cost of solving a software engineering test, China’s glm 5.2 actually cost more to use than systems from Anthropic or OpenAI.

Reliability is another headache. When Tang Jie released glm 5.2 at 5:21pm Beijing time on June 13^th, he boasted of “radical openness” and slammed US trade blockades. Because open-source software can be downloaded onto local computer hard drives, it cannot be turned off by politicians. However, US congressional committees are investigating American firms that use Chinese code. Furthermore, China’s hardware shortage means its AI services suffer frequent blackouts and traffic jams. Most tellingly, the Beijing government has not bothered to regulate its local AI labs heavily yet, the clearest proof possible that Chinese officials know their tech is still running behind America’s.

So what should we do?

The stakes are admittedly high. If the West caves in to local NIMBYism and voter fear, it risks ceding the AI frontier (and its future military and cyber dominance) to an authoritarian regime in Beijing. Europe and Canada are even more risk-averse than the US. If they choke off their computing power while the rest of the world races ahead, their economic losses will be permanent. Two centuries after the Industrial Revolution, very few nations that missed the initial jump have ever managed to catch up.

Yet the public backlash itself is dangerous. AI promises to improve the world much like electricity or the steam engineer did. It could supercharge stagnant economic growth, boost worker wages, invent cures for untreatable diseases, and transform education and green energy. If governments regulate it into uselessness, those miracles vanish. We already saw this happen with medical research: advances in mRNA vaccines were delayed by public backlash during the COVID-19 pandemic.

Grand political speeches about a “new social contract” are useless. The future is too unpredictable for master plans. Leaders should instead copy the famous advice of Chinese leader Deng Xiaoping during the 1980s, when China’s economy was booming at 10% a year: “cross the river by feeling the stones.” We must move forward step-by-step, plan for trouble, and stay flexible.

To survive the AI age without a revolution, politicians and tech bosses should follow four rules:

• First, share the wealth. People who block construction need to see a direct financial reward for getting out of the way. Smart data-centre operators are already offering grants to nearby towns. This idea must expand nationwide. Governments should create wage-insurance policies to protect workers whose jobs are disrupted, ensuring ordinary citizens have a financial stake in AI’s growth. Only shared prosperity can cure the toxic politics of winners and losers.
• Second, regulate the real dangers ruthlessly. The threat of AI-enabled cyber-attacks and engineered bioterrorism is not taken seriously enough. Cracking down hard on genuine national security threats (ideally through national treaties) will strip away the arguments used by activists to ban general technology outright.
• Third, measure everything. The widespread belief that AI is currently causing mass layoffs and spiking residential electric bills is mostly wrong, but we lack the official statistics to prove it. Without hard data, viral myths about water usage spread unchecked. Facts alone won’t cure misinformation, but a lack of facts guarantees it. Other nations should copy Britain, which wisely set up an AI Security Institute and a new AI Economics Institute to track the real-world numbers.
• Fourth, use AI to fix broken government services. Why should only corporations benefit? Filing taxes should be instant and effortless. State-run healthcare systems should share medical records seamlessly to save lives, and local school should use AI tutors to help struggling children. Technology could even help voters track what their politicians are doing behind closed doors.

Voters are entirely right to care about how AI will reshape their communities. The years ahead will be strange, messy, and disruptive. But persuading the public that this disruption serves their family’s best interests in just as vital as building smarter computers. If our leaders fail to make that case, the pitchforks will come out, and humanity will destroy its brightest opportunity for the future.