# Our Intelligence Troubles **By:** Will Manidis (@WillManidis) **Published:** 2026-02-22T22:51:24.000Z **Source:** [X Article](https://x.com/willmanidis/status/2025923396148621522) --- # Our Intelligence Troubles **Will Manidis (@WillManidis)** ![](https://pbs.twimg.com/media/HBs-g-tXUAAMr8w.jpg) On Wednesday evening, the New Brunswick City Council voted unanimously to kill a proposed twenty-seven-thousand-square-foot AI data center. Hundreds packed City Hall. Hundreds more stood outside. The building was full twenty minutes before the meeting started. Within an hour the project was pulled from the docket, the land reverted to a public park, and residents were chanting in the street. A local organizer grabbed a megaphone and shouted what has become, as far as I can tell, the emerging consensus of the American public: "a big fuck you to Big Tech." The New Brunswick vote happened while the CEOs of the major foundation model companies were eleven time zones away, presenting at the India AI Impact Summit in New Delhi. Two days later, at that same forum, Sam Altman was asked about the growing concern over AI's power consumption. His answer: > "People talk about how much energy it takes to train an AI model, but it also takes a lot of energy to train a human. It takes like twenty years of life and all the food you eat during that time before you get smart." New Brunswick is not an outlier. Between May 2024 and June 2025, an estimated one hundred and sixty-two billion dollars in US data center projects were blocked or delayed by organized community opposition. One hundred and eighty-eight groups across more than two dozen states are now coordinating legal strategies, expert testimony, and messaging against data center development. Two-thirds of tracked projects under protest were blocked or delayed. The number is growing every quarter. It is my belief — and I say this having worked in AI my entire career — that we should expect widespread asymmetric violence against AI infrastructure in the United States in the near future. I do not say this happily. I am not rooting for it. I condemn violence in its fullest extent. The document that follows is not a manual for committing this kind of violence, but a warning of how easy it would be for dedicated groups to grind the American AI industry to a halt. I want to demonstrate three things: 1. First, AI is far less popular than its advocates believe. The opposition is broad, bipartisan, accelerating, and structurally different from any prior technology backlash. 2. Second, stopping AI — or imposing catastrophic costs on the industry — would be a trivial activity for a small number of motivated actors, and it would be structurally difficult for the state to prevent. 3. Third, if the industry does not change how it communicates with and operates within the communities that host its infrastructure, the violence I describe is not a risk to be managed. It is a certainty to be endured. In this post we will cover the first section: popular AI sentiment. Later posts will cover exhaustive wargaming of anti-AI popular action, state response, and infrastructure hardening. ## I. AI Labs Are Much More Unpopular Than We Give Credit For In June of 2025, the Pew Research Center surveyed five thousand and twenty-three American adults about AI. The headline finding was that fifty percent said they were more concerned than excited about the increased use of AI in daily life. This was up from thirty-seven percent in 2021. Thirteen points in four years. Fifty-seven percent rate the societal risks of AI as high. Only twenty-five percent say the benefits are high. Fifty-three percent believe AI is making people worse at thinking creatively. Fifty percent say it is making people worse at forming meaningful relationships. Five percent — five — say AI is improving human relationships. The ratio is ten to one. YouGov ran the same question at four intervals between December 2024 and June 2025. The share of Americans who believe AI's impact on society will be negative went: thirty-four, forty, forty-one, forty-seven. Thirteen points in six months. Only six percent say the effect will be "very positive." Forty-three percent of the American public is at least somewhat concerned that artificial intelligence will cause the end of the human race. Up from thirty-seven percent three months earlier. It's not easy to write off this anti-AI sentiment as a fringe, narrow issue anymore. A December 2025 Navigator Research poll found top-line AI favorability at 49%. But as with many averages, this one covers up something much deeper. Men skew favorable by 17 points. Women skew unfavorable by 7. Men under 25: plus 26. Women over 55: minus 9. The spread between the most favorable demographic and the least favorable is 35 points. The partisan split is also worth looking at—because there isn't one. Republicans are net positive 17. Democrats are net positive 2. Independents are net negative 8. In a political environment where the parties are evenly matched and neatly diverged on virtually everything by 20 points or more, AI opposition does not map onto tribal lines. It's not a red issue or a blue issue. It's an issue where the majority of each party is closer to the other party than to the industry pushing the thing. Employment numbers capture the fear most viscerally. 71% of respondents to a Reuters poll said they were concerned about AI putting "too many people out of work permanently." 64% told Pew they expect AI to lead to fewer jobs over the next 20 years. 61% of white-collar workers believe AI could replace their specific job within the next couple of years. It's easy to see where these fears are coming from. Approximately 55,000 layoffs were attributed directly to AI in 2025, according to the firm Challenger, Gray & Christmas. 40% of employers surveyed in the World Economic Forum's Future of Jobs Report said they expect to cut staff because of AI. Dario Amodei, the CEO of Anthropic, publicly said that AI could probably eliminate 50% of entry-level white-collar jobs. When you ask everyday Americans what they want done about AI, the consistency is almost eerie. 72% of voters want to slow down AI development. 82% do not trust technology executives to regulate AI—a level of distrust that puts AI CEOs somewhere between Congress and used-car dealers. 75% of Democrats and 75% of Republicans prefer a careful, considered approach to AI development. 75 and 75. I am not a big follower of American polling, but I can't imagine another issue where both parties agree at that level. 64% of registered voters want federal regulation. Americans do not want the AI industry to regulate itself. The response our industry has come up with is the China argument. And let me be clear: the China argument does not work. 80% of Americans told Axios that they prefer cautious AI implementation even if it means letting China get ahead. Our industry has been betting its future on a messianic fantasy of a coming war with China, and everyday Americans simply do not care. They say slow down anyway. 61% favor taxing AI companies' high energy use to support the electrical grid. This is a direct policy proposal. On data centers specifically, only 40% of Americans would welcome one in their community. To put that number in context, data centers are less popular than gas-fired power plants, less popular than wind farms, and less popular than nuclear facilities. By the end of 2025, data centers were a campaign issue in at least a dozen states. In Texas, candidates ran against them and won. Thousands signed petitions to block data center projects across the state. Strategy meetings drew state legislators. A Republican candidate for state senate ran explicitly as the voice for rural data center opposition. In Indiana, an organizer for the Citizens Action Coalition called data center opposition by far the largest kind of local pushback he's ever seen—and he's been organizing for 16 years. In Georgia, eight municipalities enacted moratoriums. In Virginia, a county board chair accused a large tech company of hiring investigators to dig up dirt on local opponents and declared the county ready to go to war. In Maryland, 20,000 people signed a petition against a single proposal to convert a dead mall into a data center. In Arizona, more than 100 people show up at the state capitol to protest data centers each week according to a security guard I talked to. Big Tech responded by increasing lobbying and advertising expenditures to what observers described as unprecedented levels. The track record of lobbying as a tool for defusing populist opposition is, historically, not great. ## II. AI Labs Are Unpopular in a Way Other Technologies Have Never Been It's easy to write off these numbers as the kind of thing that happens whenever a new technology is introduced. We've certainly seen protests over the cost of transformative technological change many times throughout our history. But I think these numbers matter beyond the obvious, for a very specific reason: every technology that has faced broad public opposition survived that opposition because it had an institutional and popular constituency fighting on its behalf. Nuclear power, although brutalized, had the defense establishment, the utilities, the construction unions, and a significant share of the environmental movement. The fight over nuclear power was a fight between two large, well-organized coalitions, and it was a tight one. GMOs had industrial agriculture—one of the most politically powerful sectors in America—as their constituency. Importantly, they had big ag, which was powerful before the GMO wave arrived. Fracking had local jobs, royalty payments, cheap natural gas, and energy independence overlaid on the ongoing wars in the Middle East. The communities that bore the environmental costs of fracking were often the same communities that benefited from the economic activity. The fight over fracking was ultimately a fight within communities, not between them. AI has none of this. AI's constituency is the people who build it, the people who invest in it, and the people who earn enough to believe they'll come out ahead. These people are concentrated in literally a handful of zip codes. They are disproportionately male, young, college-educated, and high-income. They are, in demographic terms, niche. They also largely don't have pre-existing political power or coalitions. Many of them are encountering Washington for the first time. You can tell by how they tie their ties. Further, unlike crypto, there is no sense in what an everyday American could participate in the windfall of artificial intelligence. The largest AI names are now publicly traded and the ones that are already priced in transformative AI change. The labs are not hiring, on the margin, from outside of small networks of already existing tech employees. And unlike crypto, there is no retail trade that will take you to the moon. Everyone else—the majority of the American population—is somewhere between uncertain and hostile, and the trend line is moving toward hostile at roughly 10 points a year, with no indication of reversal no matter how much AI progress we make. The structural problem is the cost-benefit distribution. AI's costs are concentrated. The community that hosts the data center pays in higher electricity bills, water issues, construction disruption, lost farmland, and noise. The workers who lose their jobs pay with their livelihoods. These costs are specific, tangible, and personal. AI's benefits are diffuse, abstract, and largely captured by a small number of firms and their shareholders. This is the inverse of the structure that allows controversial technologies to survive. When costs are diffuse and benefits are local, the technology has a natural constituency that will fight for it. When costs are local and benefits are diffuse, no one shows up to the public hearing to defend it. The opposition has the field to itself. No major technology in American history has entered its scaling phase—the phase where you deploy trillions of dollars into physical plant, into real communities, drawing real resources—with this demographic profile of opposition. AI is attempting to do something without precedent, and it's attempting to do so without noticing. If you listen to conversation inside the industry, you wouldn't hear any of these numbers being discussed. The discourse is about scaling laws and token budgets and capability curves and the race to AGI and China. To the extent that anyone has articulated these concerns, the response is that amorphous benefits—productivity gains, curing cancer, transformative tech bio—will turn people around once they see undeniable evidence that something good is occurring here. This assumption is backed by no data. The data shows the opposite. The more people learn about AI, the more they use it, the more they oppose its unchecked development. The trend lines are unambiguous. The AI industry is not merely unpopular. It is unpopular in the specific way that precedes organized political action — in the way that generates ballot initiatives and moratoriums and candidates and movements. It is unpopular in a way that, historically, precedes something worse than politics. ## III. AI Leaders cannot stop shooting themselves in the foot. It would be one thing if the industry were just unpopular and quiet about it. Many industries survive unpopularity by keeping their heads down, making concessions, and waiting for the cycle to turn. Exxon did this for decades. Monsanto still does in certain jurisdictions. The AI industry has chosen a different strategy, which is to get on stage every two weeks and tell the American public that their jobs are ending and their children will be cognitively inferior to a piece of software. Here are just some recent examples that showed up in a five-minute scroll of my Twitter feed: 1. Sam Altman, at a Federal Reserve event in July 2025: customer support jobs will be "totally, totally gone." 2. Altman, at OpenAI's DevDay in October 2025: some of the jobs AI will replace may not have been "real work" anyway. 3. Altman, in a separate interview: a child born in 2025 is "unlikely ever to be as smart as artificial intelligence." 4. Altman: "It'll be very hard to outwork a GPU." 5. Altman: OpenAI believes it may be "only a couple of years away from early versions of true superintelligence." 6. Dario Amodei, CEO of Anthropic: AI could eliminate fifty percent of entry-level white-collar jobs within five years. 7. Mustafa Suleyman, Microsoft's AI chief: white-collar work has "a year to eighteen months." 8. Sebastian Siemiatkowski, CEO of Klarna: cutting his workforce by a third by 2030, citing AI. These are non-editorialized quotes. Read them from the perspective of someone who answers phones for a living, processes insurance claims, does data entry, or works any of the hundreds of white-collar occupations that these men are announcing the elimination of. Fear is the only rational response. The core issue is that the industry is caught in a contradiction it can't resolve. In order to raise the money necessary to fund massive training runs, investors and enterprise customers must hear the CEO stand on stage and explain how many human tasks the technology can now perform, how much cheaper it will be than the humans, how much better it will be by next quarter. This is the revenue case. It's what the market rewards. It's what every earnings call is built around. The pitch to the public then requires that same CEO to promise that AI will create new jobs, that the transition will be managed, and that no one will be left behind. This is at best a political survival argument. It's what a continued social license to operate demands. The problem is that these two claims cannot coexist. The market pitch wins because that's where the money is. No one particularly cares what happens to the people left behind, and everyone can tell. In February 2026, Altman acknowledged that "AI washing" is real—companies are blaming AI for layoffs that have nothing to do with AI. In that very same interview, he said that AI-caused job displacement will begin to be "very palpable." He needs the displacement to be simultaneously exaggerated by others to deflect blame and real when he describes it to justify the product. This is not a needle that can be threaded with the current comms regime. The standard reassurance—AI will create new kinds of jobs, just like every other technological revolution—is probably true at the macro level. Every major technology transition has eventually produced more employment than it destroyed, and I have no reason to believe this one will be different in the aggregate. In fact, it might create many more jobs than previous ones have. But that is irrelevant at the individual level. The new AI jobs are in San Francisco, Seattle, and New York. The displaced workers are in Phoenix, Indianapolis, and rural Georgia. The gap between these places is not a retraining program. It's thousands of miles, a different life, a different community. This is the same thing that happened to the communities that were promised retraining when manufacturing was relentlessly offshored and stolen from American workers in the 1990s. Those workers, if they are still alive at all, are still waiting for those retraining programs. Everyone who has ever lived in one of these affected communities knows this. It is a promise that has been made before and not kept. ## IV. The Water and Power Libels The AI industry does not just have a messaging problem. It has a resource problem. Depending on who you talk to, AI data centers either use a drop of water per year or enough water to repopulate the Sahara. It's beyond the scope of my work here to determine which thesis is correct, but it's useful to understand that to the American public, AI is a thing that increases your power bill, destroys your water table, and takes money out of your pocket. That means from the point of view of the American consumer, the same executives who are stealing your jobs are also stealing your electricity and water. From the perspective of a community hosting a data center, your electricity bill will go up, your water table will drop, and the company will pay less in local taxes than the dry cleaner on Main Street—because it negotiated an abatement package with your city council, which approved it before anyone understood what they were agreeing to. And the CEO of the company that built that data center just went on television and told you your kid will be unemployed forever because of his software. This leads to a particularly interesting political environment for anti-AI direct action. If the issue was purely left wing coded, anti-AI activists could count on the complex liberal NGO apparatus to pay their bail or at least promise some representation. If the issue was purely right wing coded, an anti-AI activist could count on right-of-center everyday Americans concerned about the energy usage to acquit them in front of a jury. While in later parts we will dig into the legitimacy of state power in response to anti-AI direct action, this bipartisan understanding of the issue makes it difficult at a purely threshold level. The industry's response to the political opposition this generates is lobbying. In California, a bill to separate data center electricity rates from residential rates—to shield households from cost increases—was killed by industry lobbying. A separate bill requiring data centers to disclose their water usage was vetoed by the governor. What survived the legislative session was a requirement for regulators to produce a study on data center energy impacts, due in 2027. The findings will not be available in time for the 2026 session. When we look at other industries that have been in the same place, the pattern is not mysterious. Tobacco knew the harm and chose to deny it publicly. It lost public trust irrevocably and became the only industry in America that cannot legally advertise its own product. The fossil fuel industry had a window to lead the energy transition but chose to fund climate denial instead. It is now a permanent political target in every jurisdiction where there is a coherent left-wing movement, and the reputational damage will outlast the free cash flow of the companies themselves. Social media had a window to self-regulate on child safety and misinformation and chose growth instead. It is now under regulatory siege on multiple continents, although not the US. Every proposed law is punitive rather than collaborative, because the legislature no longer trusts the industry to act in good faith. In every case, the industry had a window, and the window involved getting ahead of the opposition through honesty and conversation. Even when these industries chose short-term optimization—denial, lobbying, growth—they still had a window. But once the window closed, it did not reopen. The relationship between industry and public became permanently adversarial. AI leaders seem committed to slamming the window shut at full speed. They've spent the last five years telling the public that their jobs are worthless, their children are inferior, and their electricity bills are a small price to pay for the future. The public has drawn its conclusions. Up Next: Over the coming weeks, I'll publish a series of follow-ups to this piece detailing wargaming scenarios for what I believe is an inevitable wave of violence against AI data centers. I have shared these privately with individuals across AI labs and in government to ensure that the risks are addressed before any public disclosure. If you have not received a draft of those sections and work in a role where they would be useful to you, please reach out and I'd be happy to provide them. To be clear: my goal is to identify systemic vulnerabilities facing the AI industry, not to provide a guidebook for action against it. Responsible disclosure to the parties involved is the only ethical path — and to date, many of the firms that need to take these risks seriously have not. [Open the X Article](https://x.com/i/article/2025280848187035648)