OpenAI Lost $38.5 Billion in 2025, Just as Wall Street Comes Calling

For the first time, there are audited numbers behind one of the most closely watched balance sheets in technology, and they are sobering. OpenAI lost $38.5 billion in 2025 on revenue of just over $13 billion, according to financial documents obtained by the writer Ed Zitron and independently verified by the Financial Times. The figures arrive as the company moves toward a public listing, and they sharpen a question investors have circled for two years: how long can a business keep spending this much faster than it earns?

The revenue line is genuinely strong. At $13.07 billion, OpenAI more than tripled its 2024 sales of $3.7 billion, the kind of growth most companies never see. The trouble sits below it. Total costs and expenses reached $34 billion for the year, with $19.18 billion going to research and development and a further $5.73 billion to sales and marketing. That leaves an operating loss of roughly $20.9 billion, before any of the accounting that produced the larger headline number.

Reading the $38.5 billion figure honestly

The gap between the $20.9 billion operating loss and the $38.5 billion net loss is worth a careful look, because the bigger number is partly an artefact of how OpenAI is built. Roughly $41.55 billion of one-time, non-cash charges ran through the accounts, tied to the company's conversion from a nonprofit into a for-profit structure and to changes in the fair value of its convertible interests and warrant liabilities. Those charges inflate the loss on paper without a dollar leaving the building. The cash reality is the operating loss, and it is still enormous.

What the documents make plain is how much OpenAI leans on rented computing power. Reporting on the same figures, Tech Times pointed to a roughly $17 billion commitment running through Microsoft's Azure cloud. The model that made the company famous is also the thing that makes it expensive. Every query answered, every model trained, every new feature shipped runs on hardware OpenAI mostly does not own.

The timing is the story

None of this would matter quite so much in private. It matters now because OpenAI filed confidentially for an initial public offering earlier this month, with Goldman Sachs and Morgan Stanley steering the process and a private valuation around $852 billion. A confidential filing lets a company prepare its pitch away from the spotlight. Audited losses of this size, leaked ahead of schedule, are exactly the spotlight it hoped to avoid.

Public investors tend to forgive heavy losses when the path to profit is visible and the spending plainly buys growth. Amazon spent years in the red and rewarded the patient. The harder case is one where costs scale in lockstep with revenue, so that growing larger does not obviously bring the finish line closer. OpenAI's defenders argue the research bill is the whole point, that today's spending buys tomorrow's leading position. The skeptics look at the same page and see a company that has to keep raising staggering sums simply to stand still.

OpenAI is not alone in this. The leak lands in the middle of a broader rush of AI giants toward the public markets, all of them asking ordinary shareholders to underwrite spending on a scale once reserved for nation-states. The 2025 numbers do not settle the argument about whether that spending is foresight or mania. They do give it a price tag, and the figure is hard to look away from.