OpenAI raised its spending plan to $750 billion. Its own CFO has doubts.

OpenAI has raised its planned infrastructure spending to $750 billion through 2030, up from roughly $600 billion estimated earlier this year, The Wall Street Journal reported on July 22, 2026. New deals with cloud computing providers are driving the increase. This is OpenAI's own near-term planning number, not the same thing as the much larger multi-decade totals attached to its individual vendor contracts, and the two are easy to conflate.

Three numbers that keep getting confused

OpenAI and its vendors have put three separate dollar figures into public view recently, and none of them restates another. The $750 billion is OpenAI's internal planning number for its own infrastructure and compute spend through 2030. Separately, seven vendor deals signed across 2025 and 2026 commit OpenAI to $1.15 trillion through 2035: Broadcom at $350 billion, Oracle at $300 billion, Microsoft at $250 billion, Nvidia at $100 billion, AMD at $90 billion, AWS at $38 billion, and CoreWeave at $22 billion. Those seven figures add up to exactly $1.15 trillion, a decade-long total spread across chips and cloud capacity. Sam Altman has separately referenced a $1.4 trillion figure elsewhere. Each of the three numbers measures something different, on a different timeline, and none of them should stand in for the other two.

What is actually driving the increase

The jump from about $600 billion to $750 billion is attributed to new agreements with cloud computing providers, though the reporting behind it does not fully break out which contracts moved the number or by how much. Signing capacity commitments with multiple providers at once is, mechanically, how a compute budget grows this fast: each new contract adds to the total rather than replacing an old one. The one piece named specifically is Project Camellia, OpenAI's data center project in Georgia. Beyond that, neither OpenAI nor the reporting has detailed what the rest of the $750 billion buys, so the total is confirmed and its full composition is not.

The person most worried about the math is the CFO

The sharper detail in the Journal's reporting is not the total, it is who is reportedly uneasy about it. OpenAI's chief financial officer, Sarah Friar, has privately raised concerns about whether the company's revenue growth can keep pace with the compute commitments it keeps signing. That is worth sitting with: the person whose job is to reconcile spending against revenue is reported to be running exactly that math, and not entirely convinced by what it shows.

Why a build studio cares

We do not sign nine-figure infrastructure contracts, but we buy the same underlying thing at a much smaller scale: inference, running on someone else's compute, at a price we do not set. When a company shaping a large share of that market is planning $750 billion in spend while its own CFO is reported to be questioning whether revenue supports it, that is useful signal for anyone pricing a roadmap on the assumption that model calls keep getting cheaper. They might. A $150 billion jump in planned spending within a single year is a reason to plan for either direction.

Next step: read TechCrunch's report and Yahoo Finance's coverage of the Journal's story. If your product's unit economics depend on inference pricing staying where it is today, write to us at hello@gattyworks.com.