The woman who took down Pacific Gas & Electric now has Big Tech in her crosshairs, and she's asking the same question that brought down PG&E: what are you hiding?
The Summary
- Erin Brockovich has joined the fight against AI data center expansion, targeting the lack of transparency in how projects get approved at the local level
- Communities are angry because projects are being "shoved down their throat in secrecy," often with local officials gagged by NDAs or projects disguised as ordinary warehouses
- The backlash centers on water drain, electricity costs, and quality of life concerns, but the real accelerant is the secrecy itself
- When residents can't get straight answers from their own elected officials, anger turns into organized resistance
The Signal
The AI infrastructure buildout is running into the same wall that stopped pipelines and power plants: local communities who weren't consulted and don't appreciate finding out after the fact. Brockovich told CNN that residents are discovering data center proposals only to find their local officials can't discuss details because of NDAs, or worse, that projects were pitched as generic warehouses until permits were already moving through approval processes.
This isn't nimbyism. This is pattern recognition. Communities have seen this playbook before with fracking, chemical plants, and waste facilities. Promise jobs and tax revenue, downplay resource consumption, lock local officials into confidentiality agreements, and move fast before opposition can organize. The concerns are concrete: water supply drain, surging electricity costs, and declining quality of life in areas where data centers cluster.
"There's a lot of secrecy and NDAs at a very proposal stage."
The NDA strategy is backfiring spectacularly. In Brockovich's original case against PG&E, the cover-up was worse than the crime. Chromium-6 contamination was a health crisis, but what turned a community against the utility was the systematic lying about it. The same dynamic is playing out now. Tech companies might have legitimate answers about water recycling, renewable energy commitments, and economic benefits. But when those answers come wrapped in legal agreements that prevent elected officials from speaking freely, residents assume the worst.
The timing matters. Brockovich is entering this fight as data center construction is accelerating to meet AI compute demand. Every frontier model iteration requires exponentially more processing power. Meta, Microsoft, Google, and Amazon are all racing to build capacity before their competitors do. That urgency is creating shortcuts in community engagement.
Here's what the tech companies are missing:
- Brockovich built her reputation on making technical environmental issues accessible to regular people
- She knows how to turn local anger into national stories
- Her involvement signals to other communities that resistance is possible and legitimate
The warehouse mislabeling is particularly damaging. When a project is presented as logistics or light industrial and turns out to be a multi-megawatt compute facility with cooling tower vapor plumes visible for miles, that's not a communication issue. That's deception. And it hands community organizers a ready-made narrative about dishonest tech companies buying off local officials.
The Implication
If you're building AI infrastructure, the NDA strategy just became radioactive. Brockovich's involvement will embolden other communities to demand transparency and potentially inspire legal challenges based on procedural failures in the approval process. Expect delays, higher community benefit payments, and much more scrutiny on water and power consumption claims.
For the agent economy, this is an infrastructure bottleneck with teeth. You can't run Web4 on compute you can't build. The companies that figure out genuine community partnership, not performative engagement wrapped in legal secrecy, will have a regulatory moat. The ones that keep using the industrial facility playbook are going to spend the next three years in local zoning battles instead of training models.