OpenAI used to compete with better algorithms. Now it competes for transmission lines and water rights.
The company’s Project Camellia, planned for Effingham County, Georgia, makes that shift impossible to ignore. This is not a modest server farm tucked into an office park. It is a sprawling data center development anchored by a 3.2-gigawatt power agreement with Georgia Power, with electricity flowing in strategic phases between 2028 and 2032. To understand the scale, consider that one gigawatt already represents the output of a large power plant. Multiply that by three and change, then dedicate the entire stream to training and running artificial intelligence models. That is the physical footprint of OpenAI’s next chapter.
Securing City-Scale Power
The Georgia Power deal stands out for more than its size. OpenAI has committed to covering all project costs independently, a concession designed to neutralize the political poison that often accompanies massive industrial energy loads. Across the United States, residents have watched utility bills climb after data center developers pushed grid expansions without paying full freight. OpenAI appears determined to avoid that fight in Georgia. By guaranteeing that locals will not subsidize the infrastructure through higher electricity rates, the company is buying more than copper and silicon. It is purchasing social license.
The phased delivery between 2028 and 2032 also tells a story. Three-point-two gigawatts do not appear overnight. Transformers, substations, and high-voltage lines require years of permitting and construction. The timeline suggests OpenAI is planning for frontier-model generations that do not yet exist, betting that today’s construction drought will match tomorrow’s compute famine. It is a long-range wager that demand for training and inference will grow fast enough to consume every electron the grid can eventually deliver.
Water, Heat, and Local Scrutiny
Electricity is only half the resource equation. Data centers generate extraordinary heat, and shedding it traditionally requires water. Lots of it. Evaporative cooling towers, still common in the industry, function like elaborate swamp coolers. They continuously draw fresh water and vent vapor into the atmosphere. In a region where aquifers and river systems face real pressure, that approach courts trouble.
OpenAI says Project Camellia will rely on a closed-loop water cooling system. In this setup, coolant circulates through server racks and heat exchangers repeatedly rather than being dumped after a single pass. The loops still require some makeup water and energy to run pumps, but the overall draw drops sharply compared with open-loop evaporative methods. The choice signals that OpenAI has absorbed the criticism aimed at the broader industry. Regulators and community groups are increasingly unwilling to treat resource consumption as an externality. Closed-loop cooling is quickly becoming the minimum standard for any hyperscale project that wants to survive a public hearing.
Still, the company has offered few hard numbers on total water or power efficiency. Without public meter data, the commitment remains a directional promise rather than an audited guarantee. The engineering is credible. The proof will be in the utility bills.
Buying Goodwill and Building a Developer Pipeline
Power and plumbing handle the hardware. Community relations handle the politics. OpenAI has pledged $80 million toward schools, healthcare, and housing initiatives in Effingham County. For a rural county with a population measured in tens of thousands, that is transformative money. It funds teachers, clinic equipment, and affordable housing units that might otherwise wait decades for state