AI's New Gatekeeper Is a Zoning Board

The model race ended not because someone won but because the model got cheap. OpenAI cut the price of its Luna model by 80%, Anthropic released a high-performance model at half the cost of its top system, and every major U.S. lab followed, all of it aimed at undercutting cheap Chinese models ^[1]. David Sacks, the White House AI czar, put the new reality plainly: Zhipu's GLM-5.2 sits "just a tick below" Anthropic's Opus 4.8, and it costs roughly one-sixth as much ^[2]. The frontier stopped being scarce. Hugging Face data shows 83% of all model downloads go to models with fewer than a billion parameters, while the 100-billion-plus frontier systems account for 1% ^[3]. The value had to go somewhere, and it went downstream. Morgan Stanley now describes cheap model access as a loss leader for the cloud and infrastructure spending that actually makes money ^[4]. Meta, after spending $130 to $145 billion on AI and failing to turn its chatbots into a business, is weighing selling raw compute to competitors ^[5][6]. Amazon raised its AI infrastructure spending to $220 billion ^[7]. The new AI suppliers are power and land companies: Chevron signed a 20-year power deal with Microsoft, Caterpillar sits on a record $72 billion backlog of generators, and warehouse landlord Prologis launched a $25 billion data center arm ^[8][9]. That infrastructure runs on land and power, and someone has to approve both. Pennsylvania's governor made local approval a precondition for any state permit ^[10].

If the local community doesn’t approve a project, the state won’t approve it either. — Josh Shapiro

Indianapolis voted 23-1 to advance a moratorium on new data centers through 2027 ^[11]. Even Loudoun County, Virginia, which collects $1.3 billion a year from 250 data centers, is voting on a pause ^[12]. The wall is not monolithic. West Virginia stripped local governments of zoning authority over data centers to create a uniform state standard ^[13], and Hamilton, Ontario voted 10-6 to reject a moratorium ^[14]. But even in West Virginia the backlash is building, with the state party chair vowing to restore local control ^[13]. Pennsylvania's approach points the other way: formalizing local consent rather than eliminating it. The question of who wins AI has relocated from a lab in San Francisco to a zoning board in a town you've never heard of. The firms that win the next phase won't be the ones with the best model. They'll be the ones that can get a building permit.