The same neighborhoods that bear the scars of 20th-century industry are being asked to power 21st-century AI, and they're saying no.
The Summary
- Philadelphia officials proposed data center construction in a neighborhood still dealing with fallout from a defunct oil refinery, sparking national backlash against AI infrastructure siting
- Despite calls to slow AI development, existing model usage will continue driving robust computing demand, creating pressure to build regardless of local opposition
- The pattern repeating across the US: tech companies need land, power, and water, and they're looking at the same communities heavy industry already extracted from
The Signal
The AI buildout has a geography problem. Philadelphia's proposal to site data centers near a shuttered refinery isn't an isolated case of bad planning. It's the leading edge of a conflict that will define where Web4 infrastructure gets built and who pays the price.
These neighborhoods already know the deal. They hosted refineries, steel mills, and chemical plants. They got the jobs, then watched them disappear. They kept the contaminated groundwater, the elevated asthma rates, and the property values that never recovered. Now tech is knocking, promising clean digital jobs while asking for the same things: cheap land, massive power hookups, and lots of water for cooling.
"The use of existing AI models will fuel demand for computing power, regardless of whether new model development slows."
Bloomberg notes that even if AI development pauses, the infrastructure demand won't. Every ChatGPT query, every AI coding assistant, every automated customer service interaction requires compute. The models are already trained. The agents are already deployed. The data centers have to exist somewhere.
But "somewhere" is turning into "not here" in community after community. The pattern: tech scouts find industrial land with existing power infrastructure. Perfect for data centers. Also perfect indicators of environmental justice communities that have been on the receiving end of extraction for generations.
The economics push one way. Retrofitting old industrial sites is faster and cheaper than greenfield development. The power lines are already there. The zoning allows industrial use. But the political reality is pushing back harder than the industry expected.
Key tensions emerging:
- Data centers need 24/7 power and water, creating sustained demand unlike other industrial users
- Local opposition increasingly organized and connected across cities
- Climate goals colliding with AI infrastructure build requirements
This isn't just NIMBY resistance. These communities have receipts. They can point to the last time they were told industrial development would bring prosperity and show you the Superfund sites instead. The AI industry needs social license to build at scale, and it's trying to earn that license from people who've heard these promises before.
The Implication
If you're building AI infrastructure, your site selection process now requires a political strategy, not just an engineering one. Community opposition can delay projects by years or kill them entirely. The companies that win the buildout race will be the ones that either find genuinely willing host communities or figure out how to distribute smaller data centers across more locations.
Watch for the emergence of "data center deserts," regions where opposition is so entrenched that no amount of economic incentive can overcome it. Also watch for opportunistic regions that position themselves as AI-friendly alternatives, potentially creating new geographic advantages in the agent economy. Where the infrastructure goes determines who can build on it and who benefits from it.