On July 22, OpenAI announced Project Camellia — a 3.2-gigawatt data center campus in Effingham County, Georgia that will cost more than $30 billion at full build-out. It is the first data center OpenAI will design, build, and operate itself.

That last detail matters more than the price tag.

Since its founding, OpenAI has run on Microsoft’s Azure cloud infrastructure. That arrangement made sense when OpenAI was a research lab with unpredictable compute needs. It makes less sense when you are a company planning to serve hundreds of millions of users with AI agents that run around the clock. Project Camellia is OpenAI stepping out from behind its hyperscaler partner and planting its own flag in the ground.

What Is Being Built

The campus sits on 1,400 acres near Savannah — a four-building complex roughly 30 minutes inland from the coast. OpenAI has contracted with Georgia Power for 3.2 gigawatts of electricity under a 25-year agreement, with power delivered in phases between 2028 and 2032.

The capital investment stands at $20 billion minimum. According to Bloomberg, OpenAI’s VP of Compute Strategy Sachin Katti put the total cost at more than $30 billion when the campus reaches full scale.

Phase one opens in 2028 with 400 permanent on-site jobs. By 2032, the campus is expected to employ approximately 1,000 people.

OpenAI negotiated a 50% property tax abatement with Effingham County for 15 years. Even at half the standard rate, the company is expected to become the county’s largest taxpayer — generating enough new revenue that the county has launched an initiative to reduce average homeowner property taxes by 40% as early as next year. OpenAI is also committing $80 million in community investment over the life of the project, directed toward schools, public safety, healthcare, housing, and workforce training.

The facility uses a closed-loop water cooling system that avoids drawing from or warming local water sources, a notable design choice given the scale of water consumption at AI data centers.

Why OpenAI Is Building Its Own Infrastructure

The shift from renting compute to owning it follows a pattern that every industry at scale has gone through. At the start of the internet era, companies rented server space. Over time, the largest platforms built their own data centers because the cost curve and control requirements demanded it. AWS, Google Cloud, and Azure were built precisely because Amazon, Google, and Microsoft could not afford to keep renting.

OpenAI now faces the same pressure from the other direction. The company is paying Microsoft for the compute it needs to train models and serve millions of users, with pricing set by a commercial agreement rather than direct infrastructure cost. Building your own hardware is expensive upfront but dramatically cheaper per token at scale.

There is also a strategic dimension. OpenAI’s model roadmap — increasingly oriented around agents that run autonomously over extended periods — requires infrastructure designed for agentic workloads: high-throughput inference, low-latency memory access, and persistent compute that can handle tasks taking hours or days, not seconds. Hyperscaler infrastructure is general-purpose. Project Camellia can be purpose-built.

The Larger Pattern

Project Camellia is the most visible example of a trend that has been building for 18 months: AI companies are moving from cloud-native to infrastructure-native. Anthropic has begun its own data center partnerships. Meta has been public about building AI-specific facilities. The hyperscalers themselves are investing at a pace that would have seemed impossible three years ago.

The underlying driver is demand that is not decelerating. Enterprise AI adoption has moved from pilot to production faster than anyone in the industry predicted. Each new deployment of AI agents — handling support, processing documents, managing workflows — generates inference demand that runs continuously rather than in bursts. That constant demand justifies infrastructure at a scale that was previously only justified by consumer applications.

What This Means for Business

The practical implication for organisations building on AI is straightforward: the foundation is getting more solid, not less.

When a single company is committing $30 billion to own its own compute — not to rent it, not to partner with someone who owns it, but to build it from scratch — you are not looking at a technology in its speculative phase. You are looking at a technology that has crossed the threshold from experiment to infrastructure.

The comparison that fits is the buildout of mobile networks in the early 2000s. When carriers started building tower networks at national scale, the question was no longer whether mobile would matter — it was what you were going to build on top of it.

AI infrastructure is at that inflection point. The business risk has shifted: it is no longer the risk of adopting a technology that might not last. It is the risk of being late to a capability that your competitors are already building into their operations.

Enterprise DNA works with businesses to identify where AI creates real operational advantage — not as a future experiment, but as a current capability ready to deploy. If you are still in planning mode while your industry moves into production mode, book a discovery call to map out where to start.