In September, Greg Abbott put a moratorium on new data-center permits in Texas. On Tuesday he lost to Gina Hinojosa, who promised a moratorium on data-center construction in a state that holds more than a quarter of planned US capacity for 2030. Texas chose between two kinds of no.
Ben Thompson reads that race, and the Democratic sweep around it, as a verdict on AI in AI on the Ballot. Wisconsin's campaign ads ran against data centers from both parties. Nearly every local ballot measure opposing one won or is winning; North Dakota is the lone exception so far. His thesis is that the last constraint on AI isn't compute, memory or power. It's politics. And he explains why it bites now when it never did for the last economic shock:
when companies were closing down American factories and laying off workers and moving operations to China, none of the affected towns or workers had a say. … AI, however, is the opposite: building data centers requires permission, which is to say that people actually have a say.Stratechery
The scoreboard is right. The analogy runs backward. Voters won a veto over a building, and a data center isn't a factory. A factory's output has to be made somewhere and trucked out. A data center's output leaves on fiber. The claims queue, the support tickets, the first draft of a junior analyst's model all get processed for a firm in Milwaukee just as well from a rack three states away, or in another country, or, in Thompson's own long-run scenario, in orbit.
Thompson supplies the evidence himself. His ten takes on a permanently impaired ability to build include an edge for China, which has plenty of power and the ability to build more, and a stronger case for data centers in space, because demand may grow until "there eventually is nowhere left to build." Both are the same move: capacity collecting wherever permission is cheap. That isn't the opposite of offshoring. It's offshoring, with permits and megawatts standing in for wages.
Voters won a veto over a building, and a data center isn't a factory.
So the question to ask of any local no is what it's a no to. Some of the opposition is about the site: the power bill, the water, the substation, the hum at 2 a.m. For those complaints a veto is the right tool, and it works. But Thompson's diagnosis is that this wave is about AI, not property values. Local governments used to compete for internet-era data centers. What changed is the fear that the technology inside will take their jobs. That complaint the veto doesn't reach. Stopping a campus outside Milwaukee doesn't decide whether a Milwaukee claims adjuster's job gets automated. It decides where the tokens come from and who collects the property tax on the racks.
The best objection is Thompson's own: scarcity is the point. If enough places say no, compute stays expensive, the frontier gets paced, and people get more time to adapt. He calls it "the break Americans wanted." Grant it. Enough local vetoes do slow a national build-out, and slower does buy time. But the time arrives as a higher price per token, paid first by the startups and everyday users he lists among the losers. It doesn't arrive as any decision about work. A scarcity that lands hardest on the cheap end of the market is a strange way to protect a paycheck.
Factory towns never got a ballot on offshoring. This time they got one, and it asked about land use. Texas chose between two kinds of no, and neither one answered the question underneath the ads.