OpenAI just wrote the playbook for how to sell a power-hungry data center to a community that doesn't want one.
The Summary
- OpenAI announced Project Camellia, a $20+ billion data center near Savannah requiring 3.2 gigawatts of power, with promises locals won't see higher electricity bills, water use will be "minimal," and students get token credits
- The timing matters: new data centers built through 2033 could consume as much electricity as India uses today, with total consumption expected to quadruple by 2035
- OpenAI is holding public meetings *before* breaking ground, a notable shift from the tech industry's usual "ask forgiveness" approach
- The company is essentially pre-empting every NIMBYism with sweeteners and promises, revealing just how aware Big AI is that their infrastructure needs are becoming politically toxic
The Signal
OpenAI isn't just building a data center in rural Georgia. They're running a charm offensive. Public open houses. Meetings with "community leaders, schools, and other stakeholders." A press release that reads like a preemptive response to every complaint data centers have faced from California to Virginia.
The promises are specific and telling. No higher electricity bills for residents. Minimal water use. Token credits for eligible students. These aren't random perks. They're responses to the three biggest community objections that have killed or delayed data center projects elsewhere: grid strain, water depletion, and "what's in it for us?"
"We believe showing up early is the right way to build trust."
The context makes this strategy make sense. Data centers are on track to consume four times more electricity by 2035, with facilities built through 2033 alone using as much power as India does today. That's not a rounding error. That's a geopolitical resource allocation problem dressed up as infrastructure planning.
OpenAI's 3.2 gigawatt ask for a single campus isn't an outlier anymore. It's the new normal for frontier AI companies. But the political and social license to build at this scale is eroding faster than the AI companies expected. Community pushback has moved from fringe environmentalists to county commissioners asking "the tough questions," as Effingham County's Jamie DeLoach put it.
Key details that matter:
- This is OpenAI's play *after* their Abilene, Texas facility, meaning they learned something from round one
- $20 billion is a floor, not a ceiling, and OpenAI admits "significant work remains" on financing
- The "operating model" is still undefined, hinting at potential partnerships or novel ownership structures
The token credits for students are the most interesting detail. It's a small-scale version of what some Web3 advocates have proposed: local communities getting ownership stakes or direct benefit streams from infrastructure built on their land and power. Except here, it's not ownership. It's access credits to a proprietary product. It's Uber giving drivers promo codes, not equity.
This matters because the agent economy OpenAI is building *requires* this kind of infrastructure. Every API call, every enterprise deployment, every agent running in the background of someone's workday needs chips running hot in a warehouse somewhere. The bottleneck to Web4 isn't just algorithmic. It's physical. It's political. And increasingly, it's local.
The Implication
Watch how other AI infrastructure players respond to OpenAI's community engagement playbook. If it works in rural Georgia, expect Meta, Microsoft, and Google to copy it everywhere else they're trying to build. If it fails, or if locals still organize against it despite the sweeteners, the entire AI scaling roadmap gets a lot more complicated.
For founders building agent companies: your compute costs are about to reflect the political cost of building this infrastructure. The days of abstracting away "the cloud" are over. Where your models run, and who's happy or angry about it, will become part of your operational risk.
For everyone else: the data center siting wars are just starting. The companies building the agent economy need power measured in gigawatts, and they're running out of places to put it without pissing people off. Tokens for students won't solve that. But it might delay the reckoning.