OpenAI's planned spending on computing infrastructure has now reached $750 billion through 2030. That number has moved around quite a bit: CEO Sam Altman previously cited $1.4 trillion, CFO Sarah Friar stepped in earlier this year to correct the record at $600 billion, and now new cloud deals have pushed it back up by another $150 billion. The constant revisions reflect a company still figuring out how fast it can actually grow.
The core problem is straightforward. Running advanced AI systems requires enormous amounts of computing power, and computing power requires physical buildings, specialized chips, and a staggering amount of electricity. OpenAI currently buys most of that capacity from other companies, like Amazon, Oracle, and Microsoft. Project Camellia in Georgia is its first attempt to build and own that infrastructure itself.
The campus will sit on 1,400 acres northwest of Savannah and will draw 3.2 gigawatts of power from the regional utility, Georgia Power. One gigawatt is roughly enough to power a mid-sized city. OpenAI has contracted for more than three of those, making this one of the largest single power commitments by any company in the United States. Power will come online in phases between 2028 and 2032.
OpenAI says it will cover all electricity and infrastructure costs itself, so existing Georgia Power customers will not see higher bills. The company is also receiving a 50 percent property tax reduction from Effingham County for 15 years, and has pledged $80 million toward local schools, healthcare, and housing. Even with the tax break, it is projected to become the county's largest taxpayer.
The project is expected to create up to 1,000 permanent jobs and 400 of those roles are projected to pay six-figure salaries when the first buildings open in 2028. For a mostly rural county, that is significant. But community opposition to data centers is growing across the United States, and local residents have already raised concerns about resource use, noise, and the reality that large data centers employ relatively few permanent workers for their size.
The environmental picture is complicated. Based on Georgia Power's own filings with the state regulator, the new electricity capacity it approved late last year will come mostly from natural gas: roughly 5.8 gigawatts of new gas generation, more than doubling the utility's existing gas fleet. Neither OpenAI nor Georgia Power has disclosed what specific energy sources will power Project Camellia.
The man OpenAI hired to lead construction adds another wrinkle. He previously oversaw xAI's Colossus data center in Memphis, which was built at record speed. That facility is now at the center of a federal lawsuit: the NAACP sued xAI in April 2026, alleging the company operated 27 gas turbines without environmental permits, potentially emitting over 1,700 tons of smog-forming pollutants per year into communities that are already failing federal air quality standards. OpenAI has said it will conduct annual independent audits and use a closed-loop water system at the Georgia site, but the fuel source question is unanswered.
Meanwhile, the original Stargate project, announced with great fanfare at the White House in January 2025, has largely stalled. Disputes between OpenAI, Oracle, and SoftBank over who would own and control the data centers meant little was actually built under the joint venture. OpenAI has since pivoted toward a mix of bilateral deals with individual cloud providers and, now, its own first-party construction. The Georgia project is the clearest sign yet of that shift.
For businesses that depend on AI services, this spending race matters in one concrete way: the more computing capacity gets built and locked up by OpenAI, the more stable and capable those services are likely to be over the next decade. The risk is that the spending outpaces the revenue, which OpenAI's own CFO has privately flagged as a concern. The company earned $13 billion in 2025 and is projecting $280 billion by 2030, a trajectory that requires everything to go right for a very long time.