OpenAI has raised its projected compute spending through 2030 to roughly $750 billion, a 25 percent increase over the $600 billion figure the company set earlier this year, the Wall Street Journal reported on July 22. The revision adds $150 billion to the largest compute budget in the industry just five months after that budget was drawn up.
The company also committed $20 billion to Project Camellia, a four-building data center campus in Effingham County near Savannah, Georgia, and the first facility OpenAI is developing itself rather than leasing. Georgia Power will supply the site with 3.2 gigawatts of electricity, delivered in phases from 2028 to 2032, and total costs at full scale could exceed $30 billion. Until now OpenAI has rented capacity through agreements with Microsoft, Oracle, Amazon Web Services, and CoreWeave, including an Oracle cloud contract worth about $300 billion that begins in 2027. Sachin Katti, OpenAI’s vice president of compute, said the company is “stepping up to lead the design and the entire process” at the Georgia site, though it is still seeking partners to help finance and build the campus. OpenAI has hired Brent Mayo, who built xAI’s Colossus campus in Memphis, to run data center construction and delivery, and has discussed standardizing its designs and potentially open-sourcing them.
The $750 billion covers compute expenditure specifically and is separate from the roughly $1.4 trillion in total infrastructure commitments that chief executive Sam Altman has described. OpenAI closed a $122 billion funding round in February at an $852 billion post-money valuation, with monthly revenue of about $2 billion reported at the time. Spending at the new projected rate would still far outrun that revenue, which is the gap the company’s fundraising has been built to cover.
The timing gives the number extra weight. Moonshot’s release of Kimi K3 days earlier triggered a selloff in AI and semiconductor stocks on the theory that near-frontier capability was getting dramatically cheaper, and that the capital expenditure plans of US labs and hyperscalers might therefore be oversized. OpenAI, which has more visibility into frontier training and serving costs than any outside analyst, responded to that environment by raising its estimate rather than trimming it.
The self-build decision also changes OpenAI’s position in the data center market. As a tenant, the company has been the anchor demand behind leases and buildouts across the specialist operator tier. As a builder, it starts to compete with its own landlords for land, power, and construction capacity, in a market where hyperscalers were already projected to spend $600 billion in 2026 against binding power constraints. A $20 billion first project is modest next to the decade budget, and its function is partly instructional: Katti said the company wants to apply what it has learned from watching its partners build.
Sources: Reuters, MarketScreener, TradingKey
–
By the Control Plane Editorial Team