OpenAI’s compute spending is set to surge to $750 billion by 2030, but revenue is stuck at the $25 billion mark

OpenAI has raised its compute spending plan through 2030 to about $750 billion, about $150 billion more than the target set in February this year. Some of this money goes to Oracle’s $300 billion cloud contract, and some is invested in its self-built $20 billion Project Camellia data centers.
(Background recap: Why did the $300 billion deal between Oracle and OpenAI ring alarm bells about the AI bubble?)
(Additional context: Burn money to buy growth! Leaked OpenAI files reveal a “major operating loss of $20.9 billion,” with profitability not expected until 2030)

Table of contents

Toggle

• What this $750 billion is buying
• Can it be paid for?
• Who ultimately foots the bill

Compute power is the real bargaining chip in this AI arms race. According to reports from foreign media, OpenAI has raised its plan for cloud and compute spending through 2030 from the target set in February this year to about $750 billion, about $150 billion higher than the previous figure of about $600 billion—an increase of about 25%. The rationale is straightforward: by signing new deals with cloud providers, it locks in compute capacity—so it can train and run the next generation of even larger models.

What this $750 billion is buying

Where exactly is the money going? The list of suppliers has already been laid out: Microsoft, Oracle, Amazon Web Services (AWS), and CoreWeave all received contracts. The biggest single deal is Oracle’s cloud contract, worth about $300 billion, with a five-year term starting in 2027.

In May this year, OpenAI co-chair Greg Brockman revealed in testimony in court that the company expects to spend about $50 billion on compute in a single year in 2026 alone—equivalent to more than $4 billion per month.

In addition to signing contracts with external cloud providers, OpenAI is also building its own infrastructure. Project Camellia, announced this Wednesday (7/22), is a self-built data center with an investment of $20 billion. It is located in Effingham County, Georgia (Northwest of Savannah), covering 1,400 acres. Power will be drawn from Georgia Power for at least 3.2 GW (gigawatts), with production ramping up between 2028 and 2032. In charge of the construction is Brett Mayo, who previously also played the same role at xAI’s Colossus data center in Memphis.

Can it be paid for?

Chief Financial Officer Sarah Friar privately warned management: if revenue growth cannot accelerate, OpenAI may not have the capacity to fulfill these compute and data center contracts in the future.

Currently, the company’s annualized recurring revenue (ARR) has grown from $2 billion in 2023 to more than $20 billion by the end of 2025. In February, it reached about $25 billion (estimated by Sacra), but from February to April it has barely moved higher.

A report by The Wall Street Journal on April 28 also noted that OpenAI had already missed its internally set weekly active users and revenue targets at the time. Put the two numbers side by side: annualized revenue of about $25 billion versus a compute commitment of $750 billion—a gap of nearly 30 times. That is also why Friar previously floated the idea of a “backstop” for chip financing by the government.

But this concept was just proposed and immediately drew public criticism, because it amounts to having taxpayers underwrite the compute bets of a private company.

Who ultimately foots the bill

OpenAI is not the only player raising capital expenditure. AWS’s expected capital expenditures next year are about $200 billion, a large increase from this year’s $132 billion; Google has also raised its 2026 capital expenditure guidance to $175 billion to $185 billion; and Microsoft’s capital expenditures in a single quarter have already reached $37.5 billion. The logic is clear: whoever locks in computing resources first gets the early pass to train the next generation of models, and those who move later may not even be able to get onto the trading table.

The question is: who ultimately buys the ticket? Cloud providers lock in steady cash flow with long-term contracts, OpenAI relies on financing and optimistic commitments in testimony to sustain its scale, and investors rely on the narrative to support valuations. Each party has reasons to believe someone else will pay first, but when the books are laid out, it is always the same name written on the bill.