OpenAI’s 2025 audited financial documents have been leaked to journalist Ed Zitron and independently verified by the Financial Times, giving the public its first detailed look at the economics of building frontier AI ahead of the company’s anticipated IPO.

The numbers are striking. Revenue reached $13.07 billion in 2025, more than tripling from $3.7 billion in 2024, clearing OpenAI’s own internal target of $10 billion for the year. But the spending side tells a different story.

What the Leaked Numbers Say

Total costs and expenses reached $34 billion in 2025. Research and development alone consumed $19.18 billion, up from $7.81 billion the prior year. Sales and marketing hit $5.73 billion, compared to $1.11 billion in 2024.

The operating loss for 2025 was $20.92 billion. The net loss widened further to approximately $38.5 billion, largely because of charges tied to OpenAI’s conversion from a nonprofit to a for-profit entity. That conversion produced a $41.55 billion non-cash loss tied to changes in the fair value of convertible interests and warrant liabilities. Strip those out, and the underlying operating loss was around $8 billion.

One figure stands out in the spending breakdown: OpenAI paid Microsoft $17.2 billion across the year, split between $10.59 billion for research and development, $6.05 billion in cost-of-revenue charges, and further amounts for sales and administrative costs. That is more than Microsoft paid Anthropic in its entire 2024 investment round.

At year end, OpenAI held just over $50 billion in assets, with roughly half in cash.

The IPO Context

These numbers are landing weeks after OpenAI filed a confidential S-1 with the Securities and Exchange Commission, with Goldman Sachs and Morgan Stanley leading the process. The company had raised a $65 billion Series H round earlier this year that pushed its valuation to $965 billion.

Investors considering an OpenAI IPO will need to weigh a business that is growing fast but spending faster. Revenue grew roughly 3.5x in one year. But spend grew at a similar clip, and the company’s single largest vendor is also one of its most important strategic partners and a competitor in enterprise AI.

The efficiency ratio is improving. In 2024, OpenAI spent roughly $2.37 for every dollar of revenue. In 2025, that ratio fell to about $1.60. The trend is moving in the right direction, but at $34 billion in annual spend, it is still an enormous engine to fund.

What This Means for Business

The cost of frontier AI is real, and it is scaling. OpenAI is not a cautionary tale here so much as a datapoint. The infrastructure required to build and run models at the frontier is staggering. That is why cloud providers are spending hundreds of billions on data centers. That is why inference costs have fallen so dramatically. And it is why the pricing pressure on AI API services is likely to continue.

For business leaders evaluating AI tools and vendors, a few things are worth keeping in mind:

Vendor concentration risk is real. OpenAI’s $17.2 billion relationship with Microsoft is a reminder that AI platform economics are deeply entangled. The company that provides your AI may be structurally dependent on a third party in ways that affect pricing, availability, and strategic direction over time.

Revenue growth does not equal profitability. OpenAI grew 3.5x in revenue and still posted a massive operating loss. As you evaluate AI vendors and platforms, it is worth asking how sustainable their pricing is. Companies burning $34 billion a year to generate $13 billion in revenue will eventually need to change that equation, either through price increases, feature tiering, or model quality trade-offs.

The open-source and on-premise alternatives are gaining ground for a reason. When a vendor’s cost structure is this capital-intensive, enterprises that need cost predictability and sovereignty start looking at alternatives. That is one reason why open-weight models from Alibaba’s Qwen and Meta’s Llama have gained serious traction in 2026.

For any business currently building on top of OpenAI’s APIs, this is not necessarily a reason to panic. The company’s revenue trajectory is genuinely impressive. But it is a reason to build in optionality, avoid deep proprietary lock-in, and watch how the company navigates the transition to profitability as it prepares to go public.

The AI companies that survive and thrive long-term will be those that find a path to sustainable unit economics. OpenAI is working on that path. The leaked financials show how far it still has to travel.