The News
Every data center ban on Ohio's ballot passed on Tuesday, all 11 of them, according to unofficial results. Kansas voters approved bans in Emporia, El Dorado and Ottawa, and Edgewater, Florida and Alhambra, California passed their own. The two questions that would have welcomed data centers, in Monroeville, Ohio and Joliet, Illinois, lost badly. Our results page has the details.
The View
Look at the calendar. On Monday, Anthropic's amended IPO filing put its compute commitments at about $895 billion. The same day Oracle priced $10 billion of bonds at up to 8.65% and put $10 billion of stock up for sale to keep building. On Tuesday, a few dozen towns and cities voted on whether they wanted any of it.
They said no.
These weren't close. Trenton, Ohio passed its ban with almost 80%. In Ashville, almost 82% voted to reverse a deal their own council had made with EdgeConneX for two data centers. Monroeville's advisory question on allowing data centers lost with about 88% voting no.
I don't think these towns hate technology. Many of the bans have a size limit. Conneaut, Oregon, Sunbury and Trenton drew the line at 25 megawatts, and Emporia and El Dorado at 10 megawatts or 100,000 gallons of water a day. A server room for the local hospital fits under that. An AI campus doesn't. What voters said is that they don't want to be the power plant and the water supply for a company somewhere else.
I have some skin in this. I run small sites that call these companies' AI models, so I want compute to be cheap and plentiful like everybody else in this business. But I get the town's side too. A campus brings a huge building, a new substation and a lot more load on the local grid, and once it's built it hires few people. Where's the upside for the folks who live next to it?
The industry's answer has mostly been to go around them. Boulder City, Nevada turned down a data center, so the developer moved the project to neighboring federal land, and the city is appealing the federal approval. In Emporia, city officials took the citizens' petition to court before a judge ordered it onto the ballot. It passed with about 60%.
That works once or twice. It doesn't work when the opposition shows up on dozens of ballots in one day. And the companies know it. Anthropic's own filing says opposition to data center development "has also become more organized and coordinated across jurisdictions." That's a risk factor, written by lawyers, in a document meant to raise money.
Tuesday was the proof.
So what would work? Perry, Ohio had the best idea on the ballot. It didn't ban anything. It said developers who want tax breaks have to post financial guarantees, and about 87% of voters agreed. That's a deal. A wall would have been a ban. If the industry offered terms like that up front, with guarantees, real tax payments and the cost of its own grid upgrades, I think a lot of these towns would take it.
The money side should be paying attention too. Lenders are pricing these builders by balance sheet right now, from Microsoft near 5% to Oracle near 8% and the smaller builders in double digits. None of that pricing seems to account for a town voting the site down after the money is committed.
Room for Disagreement
These are mostly small towns, and the biggest campuses go to counties that court them with tax breaks. The statewide Ohio ban didn't even make the ballot. A developer who loses Sunbury can look at the next county over, and plenty of places still want the tax base. Maybe Tuesday was a few dozen local fights and nothing more.
Bottom Line
If you build or finance AI capacity, treat a local vote as a real risk to the schedule and price it in before the money is committed. The cheapest fix on Tuesday's ballot was Perry's: guarantees in exchange for tax breaks, offered before anyone has to circulate a petition.
— Hank