OpenAI locked in 3.2 gigawatts of Georgia power through 2032
Project Camellia is OpenAI's first data center built with OpenAI itself as principal designer, a $20 billion campus in Effingham County backed by a 3.2 gigawatt supply deal with Georgia Power.
OpenAI just became its own data center builder, backed by 3.2 gigawatts of rural Georgia power.
OpenAI signed a 3.2 gigawatt power supply agreement with Georgia Power for Project Camellia, a data center campus in Effingham County, Georgia, reported July 22, 2026. It is the first time OpenAI has acted as the principal designer and builder of its own data center, instead of leasing capacity inside someone else's, like Oracle's or AWS's.
The shift from tenant to builder
Until now, OpenAI's compute buildout ran through other companies' data centers, leasing capacity from partners like Oracle and AWS rather than owning the physical build. Project Camellia flips that: OpenAI is the principal designer and builder on this site, which means it carries the construction risk and the schedule risk directly instead of inheriting someone else's timeline. That is a meaningfully different bet than a lease, and it is the first time OpenAI has made it at this scale.
The site and the money
Project Camellia sits on 1,400 acres of a 2,600 acre site called the Savannah Gateway Industrial Hub, about 12 miles from the Port of Savannah. OpenAI's own investment in the campus is reported at $20 billion. Power arrives in phases from 2028 through 2032, under a 25 year underlying supply agreement with Georgia Power, the investor owned utility that serves most of the state.
Who actually pays for it
Under rules set by Georgia's Public Service Commission, the state regulator, the cost of serving a large new load like this one cannot be passed on to Georgia Power's existing residential and business customers. OpenAI is paying for 100 percent of the infrastructure and power costs itself. The PSC separately approved a large load agreement it projects will save a typical residential customer about $102 a year starting in 2029, a byproduct of spreading fixed grid costs across a much bigger base of demand.
The community pledges attached to the deal
OpenAI is putting up $80 million in direct community benefits plus up to $71 million in Codex education credits for the area, alongside a closed loop water cooling system and a commitment to an independent annual audit. That kind of up front package has become close to standard for hyperscale data center developers trying to get ahead of local opposition before construction starts, rather than responding to it afterward.
Why a build studio cares
OpenAI choosing to design and build its own data center, instead of renting inside Oracle's or AWS's footprint, is a signal about where compute capacity and pricing are headed over the next several years, and Project Camellia is one of the clearest data points yet on what that actually costs and how long it takes to come online. Compute availability and price are direct inputs into what we can promise on an AI workflow build, so a multi year, multi phase supply deal like this is worth tracking for the timeline it sets, 2028 through 2032, more than for the politics around it.
Next step: read OpenAI's own announcement and The Current's reporting from Effingham County. If compute cost and availability factor into your AI build's roadmap, write to us at hello@gattyworks.com.