OpenAI raised its infrastructure spending plan to $750 billion through 2030, representing a massive multi-year capital commitment.
OpenAI has lifted its projected infrastructure spending through 2030 to approximately $750 billion, a 25% increase from the roughly $600 billion the company estimated earlier this year, according to The Wall Street Journal. The figure—equivalent to Sweden's annual GDP—reflects an expanding web of cloud computing contracts and OpenAI's first foray into designing and building its own data centers.
The centerpiece of the new spending push is Project Camellia, a $20 billion data center campus spanning 1,400 acres in Effingham County, Georgia, near Savannah. OpenAI has secured a contract with Georgia Power for 3.2 gigawatts of electricity, with initial capacity expected to come online in 2028 and full buildout extending through 2032.
The escalation comes as OpenAI pursues simultaneous cloud deals with Oracle, Amazon Web Services, and Microsoft Azure worth a combined $688 billion or more, while the company's flagship Stargate joint venture with SoftBank and Oracle has stalled over partner disputes. OpenAI generated $13.07 billion in revenue in 2025 against $38.5 billion in net losses, raising questions about whether it can sustain commitments of this magnitude.
OpenAI's $750 billion projection encompasses cloud computing contracts across three major providers. The largest single deal is a $300 billion commitment with Oracle beginning in 2027, spanning five years and requiring approximately 4.5 gigawatts of computing capacity. In February 2026, OpenAI expanded an existing agreement with Amazon Web Services by $100 billion over eight years—on top of an original $38 billion deal—bringing the AWS total to $138 billion. That arrangement includes 2 gigawatts of capacity running on Amazon's proprietary Trainium AI chips. OpenAI also committed to $250 billion in additional spending on Microsoft Azure, though no specific timeline was disclosed.
The company's compute footprint has scaled rapidly: from roughly 200 megawatts in 2023 to 600 megawatts in 2024 and 1.9 gigawatts by the end of 2025. OpenAI projects $50 billion in computing expenditure for 2026 alone.
Project Camellia marks OpenAI's first venture as the principal designer and builder of its own data center, a departure from its previous model of leasing chip capacity from cloud providers. The Georgia facility will sit at the Savannah Gateway Industrial Hub, about 45 minutes outside Savannah. Georgia Power's contract for 3.2 gigawatts accounts for roughly one-third of the 9,885 megawatts of new capacity the utility received approval to develop in December. The power mix will lean heavily on natural gas—approximately 5.8 gigawatts—with grid-scale batteries and solar supplying the remainder.
To win local approval, OpenAI accepted a 50% property tax abatement for 15 years from the Effingham County Industrial Development Authority and pledged $80 million toward local schools, public safety, healthcare, and workforce training. The company also committed up to $71 million in Codex AI credits for Georgia college students. County officials estimate the project will create up to 1,000 permanent jobs. OpenAI hired Brett Mayo, formerly of Elon Musk's xAI, to lead data center construction and delivery.
The scale of OpenAI's infrastructure commitments stands in stark contrast to its current financial position. The company lost $38.5 billion in 2025 while generating $13.07 billion in revenue, yielding a commitment-to-revenue ratio of roughly 57:1. OpenAI reported $2 billion in monthly revenue when it announced a $122 billion funding round at an $852 billion post-money valuation in February 2026.
CFO Sarah Friar has privately raised concerns with company leadership that OpenAI may struggle to honor future computing contracts if revenue growth does not accelerate, according to The Wall Street Journal. The tension between Friar's fiscal caution and CEO Sam Altman's aggressive buildout ambitions surfaced publicly in April, when the Journal reported OpenAI had fallen short of internal targets for new users and revenue. Altman and Friar pushed back on reports of internal friction, stating: "We are totally aligned on buying as much compute as we can and working hard on it together every day." OpenAI has set an internal revenue target of $280 billion or more by 2030.
The spending increase arrives against the backdrop of OpenAI's troubled Stargate initiative, the $500 billion joint venture with SoftBank and Oracle announced in January 2025 with backing from President Trump. The project has stalled over disagreements among the three partners on control, financing, and operational responsibilities. Oracle and OpenAI scrapped plans to expand the flagship Stargate site in Abilene, Texas, though construction on the initial roughly 1-gigawatt, $3–4 billion campus is proceeding separately under Oracle's oversight, with completion expected by the end of 2026. OpenAI and SoftBank subsequently reached a compromise in which SoftBank owns and develops a 1-gigawatt Texas site while OpenAI controls design and holds a long-term lease. The Stargate complications appear to have accelerated OpenAI's pivot toward bilateral deals and self-built infrastructure like Project Camellia, diversifying its computing supply chain beyond any single partner.
OpenAI's $750 billion projection covers compute spending only and is separate from CEO Altman's broader claim that the company intends to deploy $1.4 trillion in total infrastructure commitments—a figure that Friar subsequently walked back to investors. The gap between the two numbers underscores the fluidity of OpenAI's capital planning as it prepares for a potential IPO.
The company's ability to execute on these commitments hinges on its revenue trajectory, the pace of AI adoption by enterprise customers, and its capacity to raise additional capital. With $50 billion in projected compute spending for 2026 alone and power deliveries from Project Camellia not beginning until 2028, OpenAI faces years of cash burn before its self-built infrastructure comes online.