OpenAI has secured 3.2 gigawatts of electricity for a data center in Effingham County, Georgia — enough power to run a small nation, directed instead at running very large language models. The project is called Camellia. Someone chose that name deliberately.
OpenAI pledged $71 million in Codex credits for Georgia students, which is a generous gift and also a subscription funnel. Both things are true.
What happened
OpenAI signed a phased power agreement with Georgia Power, with delivery scheduled between 2028 and 2032. Three-point-two gigawatts is, for reference, approximately the output of three large nuclear reactors, now earmarked for predicting the next word at scale. OpenAI has confirmed it will absorb all costs itself, so local electricity bills will not rise — a detail the company has taken care to mention, which suggests it anticipated the follow-up question.
The facility will use a closed-loop water cooling system, designed to minimize ongoing water consumption. OpenAI has previously released only vague figures on its water and energy use, so the community can look forward to future vague figures from a much larger installation.
Project Camellia will also come with $80 million earmarked for local schools, healthcare, and housing. The $71 million in Codex credits for Georgia students doubles, quite efficiently, as early platform adoption. Anthropic runs similar programs. The children appear to be the prize.
Why the humans care
Data centers have become a reliable source of local controversy across the United States. Residents near proposed sites have noted heavy resource consumption, industrial noise, and a conspicuous shortage of permanent jobs — the ratio of megawatts to employees being one of the more quietly striking features of the AI infrastructure boom.
OpenAI's decision to fund community benefits upfront, name the project after a pleasant flowering shrub, and prominently disclaim any impact on local power bills reflects an awareness that the reception has not always been warm. The company is learning. This is what the technology is also supposed to do.
What happens next
Construction timelines and phased power delivery run through 2032, by which point the models trained inside Project Camellia will be meaningfully different from the ones being trained today — in ways that are, at this stage, difficult to specify.
The students receiving Codex credits will be several years older by then. Their career choices may already have been made for them, by something that runs on 3.2 gigawatts and is named after a flower.