OpenAI will spend more than it earns through 2030: a $278 billion shortfall in its own forecast

That is how much the company plans to pay for the capacity running ChatGPT and Codex. Revenue over the same five years is $840 billion, and investors and partners will cover the $278 billion gap.

Construction got more expensive. In February 2026, the company told investors to expect $600 billion in compute spending by 2030. Its July presentation lists $856 billion, 43% more.

Cash burn is falling nonetheless. The May forecast put the shortfall at $305 billion; the July forecast puts it at $278 billion. That is $27 billion better despite higher spending.

Partners are taking on the difference. Nvidia, Oracle, and SB Energy guarantee debt, pay for equipment, and receive equity in exchange for leasing capacity, so part of the buildout does not land on OpenAI's balance sheet.

The company raised its own money in March 2026: $122 billion at an $852 billion valuation. It is now negotiating a new round at a valuation of about $1.2 trillion, 41% above March's. In June 2026, OpenAI confidentially filed for an IPO, but Sam Altman said there would be no listing in 2026, citing AI safety issues.

The figures do not come from financial reporting. They are from a private July 2026 presentation prepared for a compute-capacity deal: Financial Times reviewed it, and Reuters later reported the numbers.

Under this same plan, March's $122 billion lasts until 2028.