Inside OpenAI’s finance team, the spreadsheet keeps getting bigger. The company has raised its planned spending on AI infrastructure through 2030 to roughly $750 billion, up from about $600 billion it floated earlier this year, according to a Wall Street Journal report on Wednesday. The revised figure, shared with clients and partners in recent days, covers data centers, graphics processors and the energy needed to run them, and it marks the second time in six months that the company has restated its ambitions upward.
The new number comes with a centerpiece project. OpenAI is preparing to build Project Camellia, a $20 billion data center campus in Effingham County, Georgia, that the company will design and build itself rather than lease from a cloud partner, a first for a firm that has depended on Microsoft, Oracle and Amazon for most of its computing. The campus sits on about 1,400 acres northwest of Savannah, and OpenAI has contracted with Georgia Power for 3.2 gigawatts of electricity under a 25-year agreement, with the first gigawatt expected in the first half of 2027 and the balance arriving between 2028 and 2032.
Power, not silicon, is the binding constraint. One gigawatt is roughly the output of a large nuclear plant, enough to supply hundreds of thousands of homes, and OpenAI’s commitment to pay for the infrastructure and electric service itself is designed to keep the project moving through regulatory review. The Georgia Power buildout that would serve the campus includes new gas generation that still requires regulatory approval, and people familiar with the matter said the timeline could slip.
The spending plan sits apart from the roughly $1.4 trillion in commitments that Chief Executive Sam Altman has referenced in other contexts, and from about $1.15 trillion in hardware and cloud agreements OpenAI has signed for delivery through 2035. None of the figures are audited or contractually locked; they are the company’s own projections. OpenAI raised $122 billion in March at a valuation of about $852 billion, and an initial public offering is expected within the year, a prospect that has made the scale of capital spending a boardroom issue.
The pace of spending has become a source of tension inside the company, according to people familiar with the matter. Altman and Chief Financial Officer Sarah Friar have clashed over how quickly OpenAI should commit capital, with Friar pressing for more discipline ahead of the IPO and Altman arguing that delay risks falling behind in a race where compute is the entry ticket. The $750 billion figure, which emerged as the Journal was reporting the Camellia plans, effectively settles the argument in Altman’s favor, at least for now.
On the same day, OpenAI said it would introduce Presence, an enterprise software product aimed at teams that want to organize work around AI assistants, part of a push to build recurring software revenue beyond the API business that still accounts for the bulk of sales. The product line, which the company has been testing with select corporate customers, is intended to show investors that OpenAI can sell seats as well as tokens.
For the cloud providers that built their strategies around OpenAI, the revision is a backlog signal. Microsoft has committed $250 billion in Azure spending under its restructured partnership, Oracle has contracted roughly six gigawatts of capacity across Stargate sites, and Amazon has committed $138 billion for compute including its own Trainium chips. Analysts said the higher OpenAI figure supports order books at each of these companies, even as OpenAI’s move to build its own campuses shifts a growing share of the work in-house.
The numbers also frame the debate over whether AI spending can be sustained. OpenAI’s $750 billion plan is larger than the annual capital budgets of most of the world’s largest companies, and it rests on the assumption that demand for compute will keep growing at a rate with no precedent in the industry’s history. Supporters point to the constraint chain, from power to chips to data centers, as evidence that supply, not demand, is the bottleneck. Critics say the company is pricing in a future that even the most optimistic forecasts do not support, and that the real test will come when OpenAI’s own infrastructure competes with the partners it still relies on.
Skeptics focus on the execution risk. OpenAI remains deeply unprofitable, posting a $3.7 billion loss in the first quarter of 2026, and building data centers is a different business from training models, one with construction schedules, power contracts and permitting risk that software companies rarely manage well. The company’s own history with Stargate, which has faced delays, shows how easily such projects slip.
The bet, in the end, is that demand for compute grows faster than the cost of supplying it. OpenAI has made that bet twice in six months, each time at a higher price, and the market has so far rewarded the ambition rather than punished the risk. Whether the $750 billion figure survives contact with reality is a question for later; the message to investors, regulators and rivals is that OpenAI intends to be the largest buyer of computing on earth. For the cloud providers that have built their own plans around OpenAI’s ambitions, the figure is also a promise that the spending will flow, one way or another, for years.
