OpenAI Project Camellia: a 3.2 GW data center in Georgia

OpenAI has announced Project Camellia — a data-center campus in Georgia with 3.2 gigawatts of capacity. It is another signal that the real constraint on AI’s growth today is energy and infrastructure, not just algorithms. We stick to the figures given in the announcement.

The confirmed numbers

• Capacity: 3.2 gigawatts.
• Location: 1,400 acres in Effingham County, Georgia.
• Spending: reported outlays exceeding $30 billion.
• Timeline: power delivery in phases from 2028 to 2032.
• Power partner: Georgia Power.
• Additional commitments: $80 million in community benefits, $71 million in Codex credits for Georgia students, and a closed-loop cooling system.

What these numbers really say

Three observations matter more than the sheer scale:

Energy is the bottleneck. A campus measured in gigawatts and a partnership with a power utility show that access to power, not just chips, now sets the pace of AI. It is the same thread we raised at the Nvidia Vera Rubin token-cost-and-energy launch and in the gigawatt contracts from AMD’s Advancing AI 2026.

The horizon is multi-year. Power delivery planned for 2028–2032 is an investment measured in years, not quarters. Today’s “boom” is being built for capacity that does not yet exist.

Scale raises dependency questions. An ever-larger share of AI capability depends on a handful of enormous campuses. For companies that means concentration of risk on the provider side — an argument for architectural flexibility and a plan to change providers.

A practical takeaway for companies

This is infrastructure news, not product news — on its own it changes nothing in your deployment. But it is a reminder of two things: the cost and availability of power will shape AI service prices, and infrastructure concentration is a vendor risk worth factoring into planning. Design so you are not hostage to a single compute provider.

The bottom line

Project Camellia is a picture of the scale at which AI’s foundation is now being built: gigawatts, billions, and a horizon to 2032. For companies, the key indirect takeaway is that energy and infrastructure concentration are real risk and cost factors. If you are planning an AI strategy and want to discuss its resilience, let’s talk.

Sources: Build Fast with AI — 23 July 2026. All figures for Project Camellia come from OpenAI’s announcement and press reporting.