OpenAI has raised its stated compute-infrastructure budget to $750 billion through 2030 — a quarter more than the estimate it gave earlier the same year. The scale shows in a single line item: Project Camellia, a $20 billion data center campus in Georgia with a power demand of at least 3.2 gigawatts.
Key takeaways
- Total stated infrastructure spending: $750 billion through 2030, up 25% from the earlier 2026 estimate.
- Project Camellia: $20 billion, 1,400 acres northwest of Savannah in Effingham County.
- Power demand: at least 3.2 GW from Georgia Power, availability across a 2028–2032 window.
- OpenAI is to pay the full cost of infrastructure and electricity and to cut its draw by up to 1 GW during peak grid demand.
- OpenAI hired Brett Mayo, previously at xAI, to lead data center construction.
A quarter more in a single year
As TechCrunch reports, OpenAI has raised its stated infrastructure commitments to $750 billion spread through 2030. That is 25% above the figure it gave earlier in 2026. The amount covers data centers, compute deals and energy — the physical backbone needed to train and serve the next generations of models.
The number defies ordinary market comparisons. $750 billion is a larger commitment than the annual capital budgets of most top-ten technology firms. For a company that is still unprofitable and funds its expansion through capital rounds and partner deals, it is a bet that demand for compute will grow faster than its cost.
Project Camellia: $20B and 3.2 GW in Georgia
The most concrete part of the plan is Project Camellia — a data center campus on 1,400 acres northwest of Savannah, in Effingham County. The $20 billion project is set to draw at least 3.2 gigawatts from Georgia Power, with availability spread across 2028 to 2032. For context, a single large power unit is usually on the order of 1 GW — Camellia needs power comparable to several plants.
According to filings described by DataCenterDynamics, OpenAI is contracting energy directly from Georgia Power, with the first gigawatt due in the first half of 2027. OpenAI has committed to pay the full cost of the infrastructure and electric service, and to reduce its own draw by up to 1 GW during peak grid demand. Effingham County granted the project a 50% property tax abatement for 15 years.
A power problem, not just a chip problem
Georgia Power won regulatory approval to add 9,885 megawatts of capacity — the OpenAI deal accounts for roughly one-third of that pool. The new capacity is to come largely from natural gas: about 5.8 GW, more than doubling the utility's existing gas fleet. Part will come from simple-cycle turbines?Simple-cycle turbine: A gas turbine with no waste-heat recovery — cheaper and faster to deploy, but less efficient and more polluting than a combined-cycle setup., which are more polluting, with grid-scale batteries and solar making up the rest.
This marks a shift in the bottleneck for building AI. Not long ago the main constraint was accelerator supply and the HBM?HBM: High Bandwidth Memory — stacked memory critical for AI accelerators. memory chain. Increasingly it is the power grid and the pace of connecting new capacity that determine how fast the next data centers can go up.
Against the backdrop of Stargate's troubles
The new figure lands as OpenAI's flagship infrastructure program, Stargate?Stargate: OpenAI's flagship program to build its own data centers for model training., has faced delays and friction. To lead data center construction, OpenAI hired Brett Mayo, who previously handled similar work at xAI. Bringing in an experienced physical-infrastructure manager signals that the company treats building its own backbone as a distinct competency, not just renting capacity from cloud providers.
Why it matters
Raising the commitment to $750 billion is not just a bigger number — it signals that OpenAI is tying its future to owning or controlling physical infrastructure, not only to models. In practice, competitive advantage is increasingly defined by access to power, land and energy contracts, not model quality alone. The scale of the commitment also carries real risk: the company is still unprofitable, and a figure of this order assumes product revenue will grow fast enough to justify the outlay. If demand cools or the cost of capital rises, some of these obligations could prove expensive. For the wider market the signal is clear: AI is becoming an energy-intensive industry where gigawatt contracts are as strategic as access to the best chips.
What's next?
- The first gigawatt for Project Camellia is due in the first half of 2027, per filings described by DataCenterDynamics.
- The full 3.2 GW availability window from Georgia Power runs across 2028 to 2032.
- Georgia Power's gas buildout (about 5.8 GW) will require further regulatory decisions whose outcome affects the campus timeline.
Sources
- TechCrunch — OpenAI's AI spending spree has ballooned to $750B
- DataCenterDynamics — OpenAI plans 3.2GW data center campus in Effingham County, Georgia
