Summary
- OpenAI is planning a proposed 3.2GW data centre campus in Effingham County, Georgia.
- Power delivery is expected in phases from 2028 to 2032.
- The project shows how AI data centre planning is being driven by grid access, water use, community agreements, and long construction timelines.
OpenAI is planning a proposed 3.2GW data centre campus in Effingham County, Georgia, adding another multi-gigawatt project to the expanding AI infrastructure pipeline.
The project, known as Project Camellia, is expected to receive power in phases from 2028 to 2032. The development includes commitments around electricity costs, water use, and community funding, reflecting the public pressure now attached to very large compute campuses.
A 3.2GW data centre campus would be an industrial load of exceptional size. Delivering that capacity would require utility coordination, transmission planning, substations, long-lead electrical equipment, cooling infrastructure, backup power, roads, construction labour, and operating staff across several phases.
AI site selection is becoming power selection
Traditional data centre markets grew around connectivity, enterprise demand, cloud regions, tax incentives, and availability of suitable land. Those factors still matter, but AI training and large-scale accelerator clusters are changing the hierarchy. Where workloads are less latency-sensitive, developers can move closer to power and land rather than staying tied to established metropolitan hubs.
That shift favours regions able to organise power delivery over several years. A phased power schedule running from 2028 to 2032 illustrates the gap between AI demand and infrastructure delivery. Model development moves quickly; substations, transmission lines, transformers, and power contracts do not.
Community commitments are becoming part of the project architecture. Large data centres can raise concerns over electricity bills, water use, noise, construction traffic, tax arrangements, and the share of local benefit. Developers increasingly have to address those questions before opposition hardens around the project.
Water is likely to remain one of the more sensitive issues. Depending on the cooling design, local climate, and workload density, a large AI campus can place pressure on water resources or require alternative cooling approaches. Mechanical systems that reduce water demand can increase energy use or capital cost, leaving developers to balance environmental, operational, and commercial trade-offs.
Europe’s version will be more constrained
European markets face the same AI demand curve, but fewer locations can absorb multi-gigawatt campuses without severe planning and grid pressure. The UK has strategic demand connection reforms and a crowded transmission queue. Ireland has had prolonged debate over data centre electricity demand. The Netherlands has used planning controls in parts of the market, while Germany, France, Italy, Spain, and the Nordics each have their own mix of power, land, water, and permitting constraints.
A US campus model cannot be lifted directly into Europe. Sites are smaller, permitting can be slower, community scrutiny can be more intense, and sustainability reporting is becoming more demanding. Europe may therefore see more distributed AI capacity, more public-sector AI infrastructure, and more pressure to locate projects where renewable power, grid headroom, and political support line up.
The Georgia proposal also highlights the long horizon attached to very large AI capacity. Even if announced today, the useful compute arrives only when power, buildings, cooling, networking, security, and operations are ready. That creates a planning mismatch for customers trying to secure capacity quickly and for governments seeking domestic AI infrastructure.
OpenAI’s proposed campus adds another signal that AI demand is being translated into power-led development. The next competitive divide will not sit only between model companies or cloud platforms. It will sit between regions that can turn energy, land, permitting, and construction into operating capacity — and those where the queue stays longer than the opportunity.
