OpenAI is increasing its computing budget to 750 billion dollars while revenues lag behind. Is OpenAI digging a hole it can no longer fill?
No company can keep spending as much money as OpenAI. The Wall Street Journal reports that OpenAI will spend up to 750 billion dollars on computing capacity by 2030. That would be a significant increase compared to earlier estimates that put it at ‘only’ 600 billion dollars.
Such high expenditures are necessary to continue developing new, better, and larger models, and to keep them running. But internally, doubts are growing about whether CEO Sam Altman’s spending spree is sustainable. The revenue OpenAI generates from its AI models is nowhere near enough to cover the costs.
Project Camellia
OpenAI announced earlier this week that it wants to build its own data center in the US state of Georgia. Under the project name ‘Project Camellia’, the company has already reserved 3.2 gigawatts of power, which it will consume between 2028 and 2032. With its own data center, OpenAI wants to become less dependent on hyperscalers. Under that same motto, the company is also working with Broadcom to develop its own AI chips.
Independence is not free. The data center in Georgia has an estimated cost of 20 billion dollars. Furthermore, OpenAI does not want to limit itself to a single data center. Under the banner of the Stargate project, announced with much fanfare in 2025, OpenAI dreamed of a data center network spread across the United States. A year and a half later, that project seems to be making little progress, so OpenAI is taking matters into its own hands.
Billion-dollar contracts with hyperscalers
Until those first data centers are actually in place, OpenAI must continue to turn to external providers. For a long time, Microsoft was the exclusive cloud partner, but OpenAI has put its eggs in several baskets. It has closed agreements with Oracle, AWS, and Google Cloud.
The costs of those cloud contracts are again reaching into the billions of dollars. A contract for 6 gigawatts of computing power was signed with Oracle for 300 billion dollars. The contract with AWS will cost 138 billion dollars over eight years, and an additional 250 billion dollars in extra computing capacity was purchased from main supplier Microsoft. A standard calculator would have long since overheated when calculating these amounts.
Deep red figures
To keep funding all of this, OpenAI relies on generous sponsors and investors. After all, OpenAI is far from profitable—on the contrary. In the context of the planned IPO, OpenAI had to provide an insight into its figures, and they are deep in the red.
While revenues rose to 13 billion dollars in 2025, the company ran an operating loss of 20.9 billion dollars. In one year, that loss increased by twelve billion dollars. R&D alone costs the company about twenty billion dollars annually, ten billion of which it pays to Microsoft. Microsoft certainly wants something in return for years of sponsorship.
Generous sponsors, fragile promises
With figures like these, you would normally not be able to convince a rational investor, but logic does not seem to prevail when it comes to AI companies. Investors are lining up to advance OpenAI’s bills. Microsoft remains the most important investor to date, but that relationship has been under high tension for some time. After Microsoft, Amazon, Nvidia, and the Japanese Softbank are among the main shareholders.
To raise extra cash, OpenAI plans to go public: an IPO from which much is expected. Elon Musk’s SpaceX already showed earlier this year how many billions an AI company can rake in on the stock market. OpenAI’s market value is estimated at one trillion dollars. That high valuation is based on promises that OpenAI will one day make large profits.
As a listed company, OpenAI will be viewed differently. It will not be able to keep investors interested forever with beautiful promises. No company can keep spending money indefinitely that does not come back in, not even OpenAI.
