OpenAI told investors this year that its compute and infrastructure spending through 2030 would run around $600 billion. A presentation prepared in July, built to win a computing deal, puts that figure at roughly $856 billion instead, an increase of about 43 percent in five months.
The same July materials show a smaller number moving in the opposite direction. OpenAI now projects negative free cash flow of $278 billion for 2026 through 2030, down from about $305 billion forecast in May. Revenue is projected to grow roughly tenfold over four years, from about $36 billion this year to $350 billion in 2030, and that assumption sits underneath every other figure in the deck.
Spending can climb while cash burn falls when the build sits on someone else’s balance sheet. OpenAI does not carry an investment grade credit rating, so much of its infrastructure gets financed through partners instead of paid for directly. Nvidia has reportedly been in talks to guarantee $250 billion of data center debt tied to the buildout, letting lenders underwrite against the chipmaker’s credit rather than OpenAI’s. Oracle is spending more on data centers each quarter than it earns, much of that tied to OpenAI commitments, with the capital expense landing on Oracle’s books. SB Energy took $5.5 billion in OpenAI warrants in exchange for a 20 year lease, trading a cash obligation for equity.
That financing pattern is worth watching on its own. Oracle needed the bond manager PIMCO to anchor $10 billion of a $16.3 billion data center loan after US banks pulled back from the deal, a sign that lenders are already pricing risk carefully around the buildout that underpins OpenAI’s spending.
The revenue projection deserves more scrutiny than either spending figure. The cash burn number is essentially what is left once that tenfold revenue jump gets subtracted from planned spending. If the compute commitments are largely locked in through contracts and leases, a revenue shortfall would not shrink the burn number. It would make it worse.
There is also a timing pressure behind these projections. OpenAI raised $122 billion in March at an $852 billion valuation and is reportedly on pace to spend through that by 2028, two years before this spending window closes. That gap helps explain reports that OpenAI is in talks valuing the company near $1.2 trillion, and why a public listing timeline keeps coming up in coverage of the company.
None of these figures are audited results. They are projections inside a document built to close a specific financing deal, and OpenAI has already revised them twice within the year. The company has also paused a planned Stargate data center site in the United Kingdom over energy costs and copyright disputes, a reminder that announced infrastructure plans do not always survive contact with local conditions.
The Financial Times reported these numbers from the July presentation; The Next Web summarized and analyzed that reporting.
For anyone tracking AI infrastructure exposure, the number to watch is not $856 billion or $278 billion on their own. It is whether the guarantees from Nvidia, Oracle and other partners show up in language OpenAI can actually be held to, since a figure in a sales deck and a figure in a binding contract carry very different consequences if the revenue growth does not arrive on schedule.
Reported by the Financial Times and summarized by The Next Web, September 19, 2026.
