OpenAI announces Project Camellia, a data center in Effingham County, Georgia, with a 3.2 gigawatt contract signed with Georgia Power to be delivered in phases between 2028 and 2032. OpenAI covers all infrastructure costs, so local utility bills won’t rise. Closed-loop cooling to limit water consumption, $80 million for schools, healthcare, and housing in the county, plus up to $71 million in Codex credits for Georgia students.

For anyone building with AI, data centers are the physical substrate underlying every API call and every agent session. More compute capacity means prices that hold steady or drop. The interesting part is the $71 million in Codex credits for students: it’s community investment and user acquisition in a single move. We’ve seen Codex grow from 700k to 7 million users in six months, and free credits for Georgia students expand that user base before they even enter the job market.

The commitments on energy, water, and local investment should also be read against the backdrop of New York’s data center moratorium. U.S. communities see more costs than benefits in data centers, and some states are beginning to block them. OpenAI’s response is to pay down friction: utility bills that don’t rise, funded schools, water saved.

In Detail

Project Camellia is the first data center OpenAI announces with an energy contract of this scale and a structured package of community commitments. The 3.2 gigawatts need context: a typical data center consumes between 50 and 100 megawatts. 3.2 GW equals roughly 30-60 conventional data centers, approximately the electrical consumption of a mid-sized city. Phased delivery between 2028 and 2032 means capacity arrives gradually, not all at once.

The Georgia Power contract. OpenAI pays all infrastructure costs, so county residents see no increase in utility bills. This is politically sensitive: in other locations, data centers have strained local grids, raising costs for residents. Closed-loop cooling should limit water consumption, but OpenAI hasn’t yet released specific figures on how much water the system will use at full capacity.

The $71 million in Codex credits for students is the most relevant part for those tracking AI as a tool. Codex is OpenAI’s coding agent, included in ChatGPT subscriptions. Giving free credits to Georgia students means letting a generation of future developers try the tool before they can compare it to alternatives. Anthropic and other AI companies do the same with their education programs: it’s an established pattern in the industry, and it works because early adoption tends to create lock-in.

The context of local reaction. Protests against data centers are growing across the U.S. Residents see more downsides than benefits: heavy energy and water consumption, noise, few permanent jobs. New York’s moratorium was the first legislative signal. Project Camellia is the first data center where an AI company responds to this criticism with an explicit financial package, not just statements of intent. If the model works, other labs will copy it.

What remains unclear. OpenAI hasn’t disclosed compute capacity in GPU or flop terms. It hasn’t published estimates on permanent jobs the data center will create, which is a recurring resident complaint. Water consumption data remain vague. The delivery timeline, 2028-2032, is long and political and economic conditions could shift along the way.

For readers, the signal is twofold: the race for compute capacity isn’t slowing down, and those leading it are learning that buying local community consent, not just megawatts, is essential.