The leaked financial document circulated through Silicon Valley on Tuesday like a piece of evidence. It showed OpenAI’s 2025 results in precise, unflattering detail: revenue of $13.07 billion and research and development spending of $19.18 billion, a gap that explains why the company is heading to the public markets with a story about growth rather than profitability.
The document, which the company did not confirm or deny, also carried a line that made the headline numbers look worse than the underlying business. After stripping out roughly $30 billion in accounting adjustments tied to the company’s conversion from a nonprofit to a for-profit structure, OpenAI’s operating net loss was about $8 billion for the year. Those adjustments are paper charges — revaluations of investor rights and historical cost bases — that do not represent cash leaving the company, but they dominate the reported figures and make any direct comparison with conventional software companies misleading.
The underlying picture is clear enough. OpenAI’s revenue more than tripled in a year, a growth rate that would be extraordinary for any software company and has rarely been seen for one with $13 billion of annual sales. ChatGPT, the product that made the company a household name, had grown past 900 million weekly active users by the end of 2025, according to people familiar with the metrics. The company has become the default AI provider for a meaningful share of the world’s knowledge workers, and its enterprise business has signed some of the largest software contracts in history.
Costs grew faster than revenue. The $19.18 billion in research and development spending reflects the price of staying at the frontier: training runs on clusters of hundreds of thousands of accelerators, data center leases measured in billions, and the salaries of the industry’s most sought-after researchers. OpenAI’s cost structure is tied to usage — every conversation answered, every image generated, every API call consumes compute — which means the company’s expenses scale with its success rather than being amortized once. This is the fundamental difference between AI and traditional software, and it is why the gap between revenue and profit has widened even as revenue has exploded.
The timing of the leak matters. OpenAI confidentially filed a draft registration statement with the Securities and Exchange Commission on June 8, according to people familiar with the matter, and is widely expected to pursue a public listing later this year that could value the company at up to $1 trillion. The company raised a record round in March at an $852 billion valuation, giving it a balance sheet cushion of more than $73 billion. The leaked numbers give prospective investors their first independent look at the operating reality behind those valuations.
The numbers also frame the IPO’s central tension. OpenAI’s revenue scale — a $13 billion business growing roughly threefold annually — is the argument for a trillion-dollar valuation. The $8 billion operating loss, and the trajectory implied by management’s projection of $25 billion in cash burn this year and $57 billion next year, is the argument for caution. Public market investors will demand a path to profitability that private investors, who have committed more than $100 billion to the frontier-model industry collectively, have not yet required.
The accounting adjustments deserve scrutiny. The roughly $30 billion non-cash charge stems from OpenAI’s restructuring, which converted its nonprofit parent’s stake into a different class of ownership and revalued certain investor rights. Such charges can appear and disappear from one reporting period to the next, and analysts who have studied the document say the $8 billion operating net loss, after excluding them, is the figure that best reflects how the business is actually running. By that measure, OpenAI loses roughly 60 cents for every dollar of revenue — a loss rate that has barely improved as the company has scaled.
The comparison with the industry’s other giant is instructive. Anthropic, OpenAI’s closest rival, has raised more than $100 billion in the past year and is valued at $965 billion, with a similar profile of explosive revenue growth and heavy losses. The two companies are burning cash at comparable rates to build comparable infrastructure, and both are betting that the market for frontier AI is large enough to eventually absorb those costs. The difference is that OpenAI, with ChatGPT’s consumer base and its lead in enterprise adoption, has the clearer path to revenue at scale.
For the IPO, the leaked document cuts both ways. It demonstrates that OpenAI is no longer a research project with a chatbot — it is a business with $13 billion of revenue and the fastest growth in enterprise software. It also demonstrates that the business model has not yet solved its core problem: the cost of serving every customer grows with the customer base. The company’s public offering will ask investors to price that trade-off, and the leaked numbers will be the reference point.
