$750 billion. That is what OpenAI has told backers it now plans to spend on computing infrastructure through 2030, reportedly 25 percent above the estimate the company gave earlier this year, according to a Wall Street Journal report cited by TechCrunch on July 22. The increase lands the same week work began on the plan’s first named project, and it arrives as OpenAI’s original Stargate campus has stalled.
The first installment is Project Camellia, a $20 billion data center in Effingham County, Georgia. The campus footprint is 1,400 acres, sited northwest of Savannah. OpenAI is committed to draw a minimum of 3.2 gigawatts from Georgia Power, the local utility, with generation available sometime between 2028 and 2032. OpenAI has said it will pay “the full cost of the infrastructure and electric-service costs,” a commitment the Georgia Public Service Commission has required of any user drawing more than 100 megawatts since a rule change last year. Effingham County is granting OpenAI a property-tax abatement of 50 percent for the next 15 years.
Georgia Power’s own filings show where the power is coming from, even if OpenAI has not said how it will use it. The utility secured approval in December to add 9,885 megawatts of capacity and expects it fully committed before 2027; OpenAI’s project accounts for roughly a third of that total. Most of the new supply, nearly 6 gigawatts of new natural gas capacity, will more than double the utility’s existing gas fleet. AI Insiders previously covered OpenAI’s Ohio data center lease, backed by Nvidia financing commitments. This Georgia disclosure is different: it is the first time OpenAI has put one total dollar figure on its infrastructure ambition through the decade’s end, and that figure now dwarfs any revenue number the company has ever disclosed.
That gap is the real story. OpenAI has never published audited annual revenue anywhere near $750 billion, and its own public statements have put annualized revenue in the tens of billions, not hundreds. A company spending at this scale cannot fund it from operating income alone. It funds it through vendor financing: chipmakers and cloud partners extending credit, taking equity, or prepaying for capacity in exchange for guaranteed future purchases.
That is the same mechanism running through two other stories in this news cycle. AMD’s deal with Anthropic ties chip supply to an equity stake rather than a straight cash sale. Separately, lenders have begun extending credit against secondhand GPUs as collateral, treating depreciating silicon as a bankable asset. Nvidia, AMD, and Oracle each sit on both sides of OpenAI’s ledger: as vendors billing for chips and capacity, and as financiers whose capital or guarantees make that billing possible. When the customer, the lender, and the equity holder overlap this heavily, a slowdown in AI demand will not stay contained to one balance sheet.
None of this makes the $750 billion figure fiction. Georgia Power’s regulatory filings, the tax abatement, and the 3.2 gigawatt commitment are concrete and already moving through permitting. What remains undisclosed is the financing structure behind the number: OpenAI has not said how Project Camellia, let alone the full $750 billion, gets paid for beyond cost pass-through language aimed at ratepayers and utilities.
Anyone modeling OpenAI’s path to an IPO or estimating 2027 AI capital expenditure should treat $750 billion as a spending commitment, not a funded balance sheet. The gap between the two is now the largest unresolved variable in AI infrastructure finance, and it will surface first in the credit terms Nvidia, AMD, and Oracle extend to OpenAI over the next two quarters.
Reported by TechCrunch on July 22, 2026.
