Concentrated Gains, Dispersed Risks: The Politics of AI in the United States

On 29 September 2026, President Donald Trump hosted the heads of the world’s most powerful AI companies for lunch at the White House. Among more than two dozen guests were Elon Musk (xAI, Tesla and SpaceX), Mark Zuckerberg (Meta), Dario Amodei (Anthropic), Sundar Pichai (Google), Jensen Huang (Nvidia) and Jeff Bezos (Amazon). The meeting was called after an AI researcher, Jacob Coxon, quit Anthropic in September 2026 and warned that the industry was gambling with our lives. Trump and six of the guests (Musk, Zuckerberg, Huang, Amodei, Pichai and OpenAI’s president, Greg Brockman) signed a one-page pledge for the industry to police itself, which Trump called “morally binding”. Then they took questions from reporters. The lunch showed how the United States has chosen to govern AI for now: by voluntary agreement with the industry rather than by law.

The mood in the room was confident. Musk told reporters that jobs will change, as they always have, and that the most likely future is “an age of abundance” in which everyone receives a “universal high income”. Trump predicted that data centres “are going to be very popular”.

Outside the room, the mood is very different. In a Fox News poll published in July 2026, 70% of voters opposed building data centres in their area. In a Bentley-Gallup survey of May 2026, 79% of Americans said AI will reduce the number of jobs in the United States over the next ten years. Trust that businesses will use AI responsibly is also falling.

So why are so many people turning against a technology its builders describe as the best thing that could happen to them? This article argues that the answer lies in three mismatches. The first is between how insiders and everyone else actually use AI. The second is between who carries the risk and who collects the reward. The third is between the politics of today and the politics that may follow. The lunch put all three on display.

Mismatch one: a word everyone knows, a tool few use well

Part of the problem starts with the word itself. “AI” now covers a chatbot that suggests a dinner recipe, a model that reads X-rays, a system that writes most of the code at a software company and, since the executive order Trump signed on the same day, “Super Intelligence”. That breadth has helped AI spread quickly. Most people have heard of it, many have tried it, and many feel they understand it.

Knowing the word is a different thing from knowing the tool. The people who get the most out of AI today are technical people, and the gap between how they use it and how everyone else uses it is wide.

Neither use is wrong. But it means the people building AI watch it transform their own work every day, while most of the public sees a handy gadget wrapped in endless talk about changing the world. When the benefit you can feel is small and the warnings you hear are huge, frustration is a natural reaction. The hype starts to sound like a threat.

Some new products are starting to narrow that gap. Meta’s Muse, launched on 8 September 2026, xAI’s Grok Bot and OpenAI’s new dots are personal agents that carry out everyday tasks, such as booking travel, filling in forms or keeping track of appointments, instead of only answering questions. Europe is again at the back of the queue, though. Muse is available only in the United States, and OpenAI has not made dots available in the European Economic Area, Switzerland and the UK, where data-protection rules are strict.

The White House offered its own example on the same day. It launched America.gov, an AI chatbot that answers questions about federal services. According to the US chief design officer, Joe Gebbia, it runs on Google’s Gemini and xAI’s Grok, and from 2027 it is meant to let people complete procedures on the platform and track their progress. It is exactly the kind of everyday use that could make AI feel useful to ordinary people. Its reliability is still an open question: on its first day it gave an answer that contradicted the President, and within hours some of its answers had changed.

Mismatch two: who carries the risk, who collects the reward

On jobs, Musk said at the White House what most people in this debate now accept: “Jobs are going to change. Jobs have always changed.” There is less disagreement here than the noise suggests. CEOs, economists, unions and workers argue about the words. Some say jobs will be replaced, some say tasks will change, others prefer to talk about jobs “evolving”. Strip away the labels and they agree on the core: work is going to change, probably dramatically, and people will need new skills. Those who can’t keep up risk being filtered out of the economy. In a debate this polarised, that is a big thing to agree on.

Some of the best data on where the change will land comes, ironically, from one of the companies driving it. Anthropic regularly publishes studies of how people use its model at work. Its labour market report of 5 March 2026, by its economists Maxim Massenkoff and Peter McCrory, measured how much of each occupation’s work is already being done with AI.

The pattern is clear. The most exposed jobs are desk jobs built on text, data and screens. The least exposed involve hands and physical presence: grounds maintenance, transport, farming, food service and construction. For decades we were told automation would hit the factory floor first. This wave is aimed at the office.

Anthropic also found that unemployment in the exposed fields has not risen yet, but that hiring of younger workers there appears to have slowed. Other data points the same way. According to Indeed’s Hiring Lab, entry-level job postings have been falling since they peaked in 2022, and in May 2026 they were 7.5% lower than a year earlier. Stanford economists Erik Brynjolfsson, Bharat Chandar and Ruyu Chen find that employment of 22- to 25-year-olds in AI-exposed occupations is now 19% below where it would be had it kept pace with less-exposed peers, mainly because fewer young people are being hired. So far AI is mostly not firing people. It is quietly closing the door young people use to get in. A separate Anthropic survey found that early-career workers see the largest share of their own tasks as automatable, and they are the most worried about it.

How much of the rest is speculation? A fair amount. Douglas Elmendorf, a Harvard economist and former director of the Congressional Budget Office, warns that what we have seen so far tells us very little about what comes next. Most of the evidence, however, points in the same direction. In May 2025, Dario Amodei, Anthropic’s chief executive, told Axios that AI could wipe out half of entry-level white-collar jobs within five years. At the White House he sounded more careful than the people around him: “AI offers tremendous benefits, but there are real risks, and how to address them is still under discussion.”

Now look at the other side of the ledger. At the end of 2025, Anthropic’s annualised revenue was about $9 billion. In May 2026 the company said it had passed $47 billion, and by the end of July it had passed $65 billion, according to Bloomberg, which also reported that the company is preparing to list on the stock market. Its valuation went from $61.5 billion in March 2025 to $965 billion in May 2026. Over roughly the same period, Bloomberg reported in August 2026 that OpenAI’s annualised revenue had passed $40 billion, about double its level at the end of 2025, and in September that OpenAI was seeking new funding at a valuation of around $1.4 trillion.

Behind them stand the giants selling the picks and shovels. Nvidia, which makes the chips almost all of this runs on, is the most valuable company in the world, worth about $5.5 trillion at the time of writing. Its data-centre business brought in $193.7 billion in fiscal year 2026, up from $115.2 billion the year before. Microsoft, Alphabet, Amazon and Meta are on course to spend about $725 billion on capital expenditure in 2026, most of it on AI infrastructure.

Put those two stories side by side. The companies building AI have multiplied their revenues within a year. The people whose jobs are most exposed are told to retrain, adapt and keep up. The money is concentrated in a handful of firms and their shareholders. The risk is spread across millions of workers, and it lands first on the young.

Musk’s universal high income may arrive one day. But people live in the present, and in the present the gains show up in share prices while the risks show up in job listings. So it is fair to ask why the public would want this future. A growing share say they don’t.

Trust is falling fastest among the young. In the same Bentley-Gallup survey, 41% of 18- to 29-year-olds said they have no trust at all that businesses will use AI responsibly, up from 29% in 2025. Seen from outside, a lunch at which the industry’s leaders promised abundance and signed a voluntary pledge to oversee themselves does little to change that. If anything, it confirms the suspicion that the future of AI is being decided by the people who profit from it most.

Mismatch three: today’s politics and tomorrow’s

In the short term, Trump’s presidency appears to be the most favourable political setting the tech industry could hope for. He opposes new AI rules, calls for “tremendous self-regulation”, frames everything as a race against China, and invites the industry’s leaders to shape policy over lunch. For now, that is everything the industry could ask for.

In the longer term, it could cut the other way. Had the Democrats been in power, they would probably have had to own the AI boom. They would likely have wanted credit for the growth, the investment and the construction jobs, and would have had to show they wanted America to win the race with China. They might well have spent their energy managing the tension between their voters and their economy, much as they did with globalisation in the 1990s.

Out of power, they have no such problem. For the left, opposing the AI build-out is easy, and it fits their principles: protecting workers, distrusting billionaires, and worrying about energy prices and the environment. It is already happening. In March 2026, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced a bill for a federal moratorium on new AI data centres until workers and consumers are protected.

“We cannot sit back and allow a handful of billionaire Big Tech oligarchs to make decisions that will reshape our economy.” Bernie Sanders, 25 March 2026

The public is listening. Trump says data centres will be very popular, and the executives at the lunch promised to win over the towns where they build. The polls point the other way. In a University of Massachusetts Amherst poll released on 14 September 2026, 65% of Americans opposed building an AI data centre in their community, including 76% of Democrats. On 30 September 2026, the day after the lunch, Senate Democrats blocked a Republican bill on data-centre electricity costs, arguing it was too weak to protect households.

The worry also crosses party lines. In the same UMass poll, 52% of Republicans opposed a data centre in their community, and in a Reuters/Ipsos poll in June 2026, 47% of Republicans said they feared AI could cost them or someone in their household a job. CNBC reports that Trump’s numbers on AI are slipping. If job losses spread from young graduates to the wider middle class, the party that hosted the lunch could be held responsible, and the party that stayed outside could say it had warned everyone. Tying the presidency this closely to the tech industry buys the industry access today. It may also hand its opponents a target for years to come.

Closing the gap

The lunch was meant to show that AI is in safe hands. To the people in the room, it probably did. To many people watching, it showed a small group of executives from the industry’s most valuable companies agreeing among themselves about a technology that is likely to reshape everyone else’s work.

The evidence in this article suggests that a voluntary pledge will not close these mismatches on its own. Public opinion is more likely to shift if people feel the benefits of AI in their own daily lives, see a share of the gains, and believe that someone outside the industry is watching. Until then, many Americans are likely to hear the abundance promised at the White House as a promise made to someone else.