Summary
- OpenAI says it is contracting with Georgia Power for 3.2GW of power for Project Camellia in Effingham County, delivered in phases from 2028 to 2032.
- The 3.2GW is a future supply plan, not commissioned IT load or an operating data centre today.
- OpenAI says it will pay the project’s infrastructure and service costs so residential customers do not subsidise it, and plans to reduce load before residential curtailment; these are company representations.
- It separates $80 million of community benefits over the project life from up to $71 million of Codex credits; projected state and local taxes are hundreds of millions, not taxes already paid.
- OpenAI promises a closed-loop water system and annual independent audit, while financing, final design, detailed phasing and operating model remain unresolved.
Project Camellia begins with a number large enough to sound like a completed asset: 3.2GW. In OpenAI’s disclosure, however, it is the capacity the company says it is contracting to receive from Georgia Power in stages between 2028 and 2032.
That difference places execution risk in the foreground. Electricity supply, transmission, substations, construction and computing equipment must arrive in a coordinated sequence. A contract can reserve a path without proving that every stage is financed, permitted, built and energised.
The power promise has two sides of risk
OpenAI says Project Camellia will be entirely privately funded and that it will pay the full infrastructure and service costs of its electricity requirements. It says Georgia residents will not subsidise the project through their power bills.
Those are material commitments, but they are OpenAI’s representations. The disclosure is not an independent regulatory guarantee that every future allocation, rate treatment or system cost has already been audited.
The company also says the project will reduce its electricity use before residential customers are asked to curtail. That turns computing load into a proposed grid-flexibility resource. Its value will depend on operating rules: how quickly load can fall, for how long, under whose dispatch signal and with what effect on computing commitments.
Three financial labels must remain separate
OpenAI commits $80 million to community benefits over the life of the project. Separately, it offers up to $71 million in Codex credits. Credits are not cash paid to the county, and “up to” is not a guaranteed full draw.
The company also projects hundreds of millions of dollars in state and local tax revenue. A projection is neither a payment already received nor a minimum contractual tax amount. It depends on construction, property, operating structure and future assessments.
Adding $80 million and $71 million into $151 million of cash paid now would therefore be wrong. The amounts have different recipients, instruments and conditions. The tax projection belongs to a third category.
Closed-loop water does not mean zero water
OpenAI says the data centre will use a closed-loop water system. Closed loops can recirculate water and reduce continuing withdrawals relative to once-through systems. They still require initial fill, maintenance and treatment, and the disclosure does not establish zero water use.
The same precision applies to employment. The project may create construction and operating work, but OpenAI has not announced thousands of permanent jobs as a contractual minimum in this disclosure.
Independent audit will matter after the promises become measurable
OpenAI promises an annual independent audit covering its commitments. That creates a future accountability mechanism if the scope, benchmark and findings are public enough to test ratepayer costs, community spending, water design and demand response.
An audit promise is not an audit result. The project is in early development. Financing, final design, exact timing of each phase and the operating model remain unfinished, as do the permits and construction milestones needed for operation.
No official $20 billion project value appears in OpenAI’s announcement, so that secondary estimate should not be presented as the company’s valuation. What can be monitored is more concrete: contract milestones, grid approvals, private-cost allocation, construction, energisation, community disbursements and annual audit findings.
The 3.2GW figure establishes ambition. The 2028–2032 sequence establishes that delivery risk is distributed across years. Project Camellia becomes infrastructure only as each promised layer moves from company statement to independently observable fact.