When Sam Altman ran Y Combinator, the start-up incubator’s motto for aspiring founders was to “make something people want”. As CEO of OpenAI, however, Altman has not followed his own advice.

Last week, voters in 10 US states decided 41 local ballot measures relating to artificial intelligence ( AI ) data centres. The AI industry lost everywhere except Harwood, North Dakota. In Urbana, Ohio, voters banned data centres and recalled their mayor. Data centres were a central issue up and down the ballot. Two days later, the tech-dominated Nasdaq fell by 5.5% as investors began pricing in a Democratic Congress and new governors who had campaigned against data centres.

The midterm elections confirmed what polls have shown for months: AI is extraordinarily unpopular. A New York Times/Siena poll found that 61% of likely voters oppose building data centres. This is a stunning reversal for an industry whose flagship product, ChatGPT, was the fastest-growing consumer application in history. But the industry has since squandered that goodwill by asking the public to bear the costs of its product while treating consent as optional.

Silicon Valley has long cultivated the myth of the entrepreneur-king: the visionary who moves fast and breaks things, who seeks forgiveness rather than permission. That may have worked for taxi medallions and hotel licences but clearly doesn’t work for resource-hungry data centres in people’s backyards.

Rather than make their case to local communities, hyperscalers rely on cloak-and-dagger tactics. In Beaver Dam, Wisconsin, the local economic development agency signed a nondisclosure agreement ( NDA ) with an anonymous shell company. Over the next year, officials approved a development agreement and a tax-increment financing arrangement to help fund the project without ever revealing that it was a data centre. To power its Colossus supercomputer in South Memphis, Elon Musk’s xAI ran dozens of unpermitted gas turbines next to a majority-Black neighbourhood where the cancer risk from air pollution was already four times the US average.

These tactics violate the basic premise of representative democracy: elected officials are supposed to act on behalf of their constituents. When bound by an NDA with a trillion-dollar company, officials can’t tell constituents what they’re voting on. Beaver Dam residents couldn’t object to a data centre nobody told them was coming. In Memphis, xAI didn’t even bother asking.

The arrogance of AI industry leaders hasn’t helped. In a recent podcast interview, Altman said it’s “quite terrible” that so much of the world’s supply of dynamic random-access memory is sitting idle in ordinary people’s PCs and devices rather than in his data centres. A month earlier, he argued that “the world should accept some bad things happening for the benefits of this technology.”

So far, industry leaders’ pitch to the American people has been: “We will automate your job, build a data centre in your backyard and raise your electric bill, all while warning that AI poses an ‘existential risk’ to humanity. And if the buildout ends in a financial bust, you will be left holding the bag.” Is it any wonder that the American people have slammed the door in their face?

Even those using the technology feel steamrolled. When OpenAI claimed its models had cracked a famous math problem, an NYU mathematician accused the company of drawing on his unpublished work, which he’d been checking with OpenAI’s coding tool. OpenAI denied the allegation and said it was “impossible” that his inputs had influenced or been used to train its models.

Consider how earlier technologies won over a sceptical public. Boeing’s 707, which ushered in the Jet Age, promised speed, quiet, and glamour, making air travel something ordinary families could aspire to. Steve Jobs sold the iPhone as a glimpse into a future people actually wanted to live in. Even FTX fraudster Sam Bankman-Fried tried to sell the public on the idea that cryptocurrencies would make the world a better place. By contrast, the AI industry’s message to the world has been: “This is happening whether you like it or not.”

This is a spectacularly bad strategy. As major oil companies learned in the 1960s and 1970s, the billions of dollars sunk into wells and pipelines are at the mercy of local governments. With Anthropic’s IPO prospectus warning that the data centre backlash threatens its business, investors are beginning to question the pace of capital spending on the AI buildout.

Now that American voters have spoken, the industry has suddenly changed its tune. OpenAI has pledged to stop using NDAs in its dealings with public officials and cover the power costs of future data centres, calling the way it builds them an early test of “whether the benefits of AI will be broadly shared”. That’s a clear admission that it could have sought consent and paid its bills along.

The incoming Democratic Party-controlled Congress has an opportunity to enact meaningful AI regulation. Opposition to data centres crosses party lines and could even overcome a presidential veto. California, where governor Gavin Newsom’s 2024 veto of SB 1047 scuttled liability rules that could have imposed real penalties on AI developers for security incidents like the Hugging Face hack, should serve as a cautionary tale.

Other recent incidents, such as the discovery of the PHP2Shell exploit, make regulatory action all the more urgent. Lawmakers should treat AI like every other federally regulated technology and establish enforceable rules rather than relying on voluntary pledges.

The AI industry may eventually learn to make something people want. But American voters have made clear that they are not willing to wait and find out. The people they elected shouldn’t wait, either.

Brian Judge is the research director of the Programme on Finance and Democracy at the University of California, Berkeley.

Copyright: Project Syndicaate