OpenAI is building a $20 billion AI data center campus near Savannah, Georgia, and it is not treating power as an afterthought. The project, called Project Camellia, is planned to draw 3.2 gigawatts of electricity from Georgia Power in phases between 2028 and 2032.

That 3.2 gigawatts number is the headline, because data centers do not succeed or fail on software alone. They live and die by access to reliable power at scale, with construction timelines that often have to line up with grid capacity, interconnection approvals, and utility planning. Project Camellia is making that bottleneck the center of the story by explicitly tying the campus buildout to phased power draw from Georgia Power over a five-year window.

To understand why this matters beyond one campus, zoom out to what is happening across the AI infrastructure stack. Training and running modern AI systems are electricity-hungry at levels that are hard to compare to earlier internet-era workloads. That creates a new kind of competitive arena where the winners are the teams that can secure power capacity early enough, not just those with the best chips or the shiniest models. When a project like this publicly frames its timeline and megawatts, it effectively tells the market what kind of bottleneck it expects to manage.

The geography is also a signal. Savannah, Georgia is positioned to become an AI infrastructure node, which has second-order effects for regional developers, construction firms, and logistics providers that depend on major industrial projects. It can also reshape local competition for land, permitting bandwidth, and workforce availability, even though the source only gives the power draw, name, and timeline. The point for executives is not that every campus will mirror Savannah. It is that the “where” and “how fast” decisions are increasingly constrained by the “how much power, when” reality.

On the utility side, Georgia Power becomes more than a background supplier. The phrase “draw 3.2 gigawatts of power from Georgia Power in phases” implies a staged approach, not a one-time switch flip. Phasing is usually how big power customers manage risk on both sides: the operator avoids betting everything on a single moment, and the utility can plan upgrades and capacity allocation without committing to the entire load at once. For board members and C-suite leaders at data-center-heavy companies, that is a reminder that long-horizon infrastructure projects are often sold as build schedules, but governed as power contracts and grid timelines.

For investors and operators, the strategic implication is that capex planning now has a hard dependency: electricity procurement. When you see a 2028 to 2032 buildout period connected to 3.2 gigawatts, you should read it as a risk map. Even if hardware availability improves, execution can still get stuck behind power delivery schedules. That shifts how companies should evaluate partnerships and vendor timelines, because the critical path may be the grid.

Now zoom back to OpenAI and the broader ecosystem. A $20 billion campus is the kind of commitment that can influence regional infrastructure decisions and signal long-term demand to the market. Whether the project is ultimately sized, phased, or executed exactly as planned, the act of announcing it with specific power draw and a defined time window helps align expectations among multiple stakeholders, including the utility, regulators, and infrastructure contractors. In a world where AI budgets are competing for electrons, projects that name megawatts and dates set the terms for how others benchmark their own plans.

For peers in similar roles, the stake is simple: if power is the gating factor, then your strategy has to treat electricity like capital. The leaders who can translate “AI compute needs” into “power procurement and delivery” will move faster. The leaders who treat it as a background utility matter will eventually feel the slowest constraint in the system.