OpenAI faces 69 local data‑center bans

AI data centers are basically warehouses full of power-hungry computers. OpenAI needs a lot of them, fast, if Stargate is going to become the backbone for future model training. But the hard part is no longer just chips, land, or financing. It’s local politics. A new round of reporting this week put a number on that resistance — and it helps explain why OpenAI’s infrastructure strategy has started to look more flexible, more distributed, and a lot less straightforward. ### What happened here? Tom’s Hardware highlighted a count of 69 U.S. jurisdictions blocking new data-center builds, with four of those moves treated as effectively permanent. That count came from the U.S. Data Center Moratorium Tracker, which has become a rough scoreboard for the backlash against these projects. The point is not that every ban targets OpenAI specifically. It’s that the same kind of giant campus Stargate wants is now running into organized opposition across the country. (finance.yahoo.com) ### Why are towns saying no? The complaints are pretty concrete. Residents worry about electricity rates, water demand, diesel backup generators, land use, tax subsidies, and the weird mismatch between the size of these projects and the number of permanent jobs they create. Harvard’s Ben Green put it bluntly in April: communities are getting more educated about the tradeoffs, and many have become effective at stopping projects. That’s the shift — this is no longer obscure zoning drama. (finance.yahoo.com) It’s a repeatable local political movement. ### Why does this hit OpenAI harder? Because OpenAI’s roadmap depends on very large, very power-dense sites. Stargate was announced on January 21, 2025 as a plan to invest $500 billion in U.S. AI infrastructure, with $100 billion to start immediately. By September 2025, OpenAI, Oracle, and SoftBank said they had nearly 7 gigawatts of planned capacity across Abilene, Texas, plus five additional U.S. sites, after reviewing more than 300 proposals from over 30 states. A project that large can’t just hide in the background — every substation, water permit, and zoning hearing becomes a political choke point. (news.harvard.edu) ### Is Stargate already feeling it? Looks like yes. Tom’s Hardware reported in late April that OpenAI had effectively backed away from relying mainly on first-party Stargate campuses and was leaning more on flexible infrastructure deals instead. The same outlet also noted that a planned 600-megawatt expansion near Abilene was scrapped even while a larger 4.5-gigawatt Oracle agreement stayed alive. That doesn’t prove local opposition caused every change — cost and execution matter too — but it fits the pattern that huge bespoke campuses are the hardest version of the buildout. (openai.com) ### What does “hardest version” mean? Think of it like trying to build an airport instead of leasing gates. Owning the whole thing gives more control, but every piece becomes your problem — land, transmission, permits, neighbors, and lawsuits. OpenAI’s newer announcements reflect that reality. Some Stargate sites are Oracle-led. Some are SoftBank-led. Some are partnerships with existing infrastructure players. That looks less like a pure owner-operator model and more like a way to route around local friction where possible. (tomshardware.com) ### Are there real examples beyond Texas? Yes — Michigan is one of the clearest. Regulators approved power arrangements for a planned Saline Township Stargate-related facility tied to OpenAI and Oracle, with 1.4 gigawatts of power service. Local anger there has centered on governance, scale, and who bears the cost. Planet Detroit noted that the facility would consume more electricity than 1 million homes. That kind of number turns an abstract AI story into a neighborhood fight fast. (openai.com) ### Is this just a few noisy fights? Probably not. The broader tracker now shows 1,074 actions tied to data-center buildout fights and legislation, and Data Center Watch said 20 projects representing $98 billion were blocked or delayed in Q2 2025 alone. Even if you treat those databases as directional rather than perfect, the trend is obvious — resistance is scaling alongside AI demand. ### Bottom line? OpenAI still has money, partners, and a huge appetite for compute. (gizmodo.com) But the bottleneck is moving outward — from GPUs inside the building to communities outside it. If 2025 was the year AI companies learned they needed more power, 2026 looks like the year they learned they also need permission. (openai.com) (datacentertracker.org)