OpenAI’s confidential financial documents paint a stark picture of a company caught between explosive growth and staggering losses. The documents, leaked as the company pursues a blockbuster IPO, show a business that tripled revenue to $13.07 billion in 2025 while spending $34 billion. A burn rate that is only accelerating.
The numbers come from internal financial statements shared with prospective IPO investors and reviewed by multiple outlets including Ars Technica and The Next Web. They offer the clearest look yet inside the economics of the world’s highest-profile AI company, and they raise a difficult question: can OpenAI grow its way to profitability before the money runs out?
OpenAI’s 2025 revenue of $13.07 billion represented a threefold increase over the previous year, beating the company’s own $10 billion target. The growth was driven primarily by ChatGPT subscriptions, API usage from enterprise customers, and licensing deals with major partners including Microsoft.
But the cost side of the ledger tells a different story. Total spending reached $34 billion, more than two and a half times revenue. Research and development consumed $19.18 billion, by far the largest line item, reflecting the enormous compute costs of training frontier models. Sales and marketing added $6 billion as the company scaled its enterprise sales operation globally.
The result was a net loss of approximately $39 billion, heavily inflated by a one-time $30 billion restructuring charge. Excluding that charge, OpenAI’s adjusted operating loss was roughly $8 billion, still deeply negative for a company with $13 billion in revenue.
Compute is the monster under the table
The single biggest driver of OpenAI’s spending is compute. The company spent 67 cents on inference and training for every dollar of revenue it earned in 2025, up from 60 cents the year before. As more users adopted the company’s products, inference costs grew faster than subscription and API revenue, compressing gross margins from 40 percent to 33 percent.
OpenAI’s own projections, disclosed as part of the IPO filing, show cumulative compute spending reaching $665 billion by 2030, $111 billion more than earlier forecasts. The company does not expect to reach positive cash flow until 2030 at the earliest.
That timeline puts significant pressure on the IPO. At current burn rates, even OpenAI’s roughly $40 billion cash reserve (held at the end of 2025) will not sustain operations indefinitely without continued access to capital markets.
The IPO math
OpenAI filed its confidential S-1 with the SEC on May 22, with the filing confirmed publicly on June 8. The company’s most recent private valuation, set during a March 2026 fundraising round, was $852 billion. IPO valuation targets are reported to range from $852 billion to $1 trillion.
But the financial disclosures also reveal how much OpenAI is spending to sustain its growth trajectory. In the first quarter of 2026, the company posted an operating margin of negative 122 percent: for every dollar earned, it spent $2.22. That level of negative margin at this scale is unusual even by aggressive-growth startup standards, and it puts the spotlight on whether public market investors will accept the same math that private investors have.
Comparisons to other high-profile pre-IPO companies are imperfect. Uber, for example, lost $4.5 billion in its last full pre-IPO year (2018) on $11.3 billion in revenue, a much narrower loss profile relative to revenue than OpenAI’s current position.
What the market is saying
Investor sentiment appears divided. Some see OpenAI’s revenue trajectory: tripling to $13 billion in a single year, as validation of the thesis that AI model providers can capture massive value. Others point to the $665 billion cumulative compute commitment through 2030 as a sign that the company’s capital requirements are structurally unbounded.
The leaked documents also reveal a one-time $30 billion restructuring charge, the nature of which has not been fully detailed. Some analysts have speculated it relates to employee equity repricing, infrastructure contract renegotiation, or a combination of both.
OpenAI declined to comment on the leaked figures. The company’s S-1 filing, once publicly available through SEC review, will provide the formal financial disclosures that investors will scrutinize ahead of the IPO roadshow.
The big picture
OpenAI’s financial situation is a window into the broader economics of frontier AI. The company has demonstrated that there is a massive market for advanced AI models. But it has also shown that the cost of building and running those models currently exceeds what customers are willing to pay. The bet, shared by investors who have valued the company at nearly $1 trillion, is that margins will improve as hardware becomes more efficient, inference costs fall, and revenue scales faster than compute requirements.
If that bet pays off, the $13 billion in revenue is the base of a much larger business. If it doesn’t, the $34 billion in spending is a hole too deep to dig out of from a public market standing start.
Sources: Ars Technica (June 16, 2026); The Next Web / Financial Times (June 16, 2026); PitchBook (June 16, 2026); Sacra Research (June 2026); TechCrunch (June 16, 2026)