← Back to feed

OpenAI lost $38.53 billion attributable to the company in 2025, according to audited financial documents first reported by Ed Zitron and independently verified by the Financial Times. The figures entered wider circulation via Ars Technica on Tuesday. The timing is precise: OpenAI filed a confidential S-1 with the SEC on June 8 and has been running the largest private AI fundraising story in history toward a public-markets conclusion.

The core numbers — $13.07 billion in revenue, $34 billion in total costs and expenses, $20.92 billion operating loss — are specific enough to end the abstraction that has defined the AI financing cycle. OpenAI can no longer be understood only through fundraising announcements and user growth claims. The other side of the ledger is now visible.

The Operating Loss Is the Number That Matters

The headline $38.53 billion attributable loss includes a $41.55 billion charge tied to changes in the fair value of convertible interests and warrant liability — an accounting item generated by OpenAI’s 2025 corporate restructuring rather than its day-to-day economics. Strip that out and the operating loss of $20.92 billion is the cleaner measure of what the business costs to run.

That figure compares with an $8.78 billion operating loss in 2024, when OpenAI had $3.7 billion in revenue and $12.48 billion in total costs and expenses. Revenue rose roughly 3.5x between 2024 and 2025. Operating losses expanded about 2.4x over the same period. The ratio has improved, but the absolute cost base is still growing faster than any ordinary software investor would accept as a steady state.

The 2025 spending breakdown: $7.5 billion cost of revenue, $19.18 billion in R&D, $5.73 billion in sales and marketing, $1.57 billion in G&A. The R&D figure is where the compute bill lives.

Microsoft Is Both Partner and Cost Center

The most structurally significant detail in the leak is the Microsoft line. OpenAI paid Microsoft $17.2 billion in expenses in 2025: $10.59 billion booked as R&D, $6.05 billion as cost of revenue, $527 million as sales and marketing, $42 million as G&A. OpenAI carried $3.64 billion in liabilities to Microsoft at year-end.

Microsoft is simultaneously an equity holder (roughly 27% of OpenAI Group as of the 2025 recapitalization), commercial partner, cloud supplier, and the largest single line item in OpenAI’s cost structure. The relationship is not incidental to OpenAI’s economics — it is constitutive of them. OpenAI’s model progress and product adoption remain inseparable from a vendor it is contractually dependent on and financially subordinate to.

That dependency will receive close attention on any IPO roadshow. An investor buying OpenAI equity is also buying a structure where a significant portion of operating costs flow to another public company that holds a quarter of the equity.

The IPO Math

OpenAI closed $122 billion in committed capital in March 2026 at an $852 billion post-money valuation, anchored by Amazon, Nvidia, SoftBank, and Microsoft. At the same time, the company reported $2 billion in monthly revenue, 900 million weekly active ChatGPT users, and 50 million paid subscribers.

The leaked financials do not contradict the revenue trajectory. They make the other side of the equation unavoidable. OpenAI’s pitch to public markets is not that the business is profitable — it is that the cost curve will eventually bend as compute efficiency improves, model commoditisation slows, and the consumer and enterprise surfaces lock in enough pricing power to close the gap.

That is a defensible thesis. It is also an enormous ask. The company is attempting to take an infrastructure-heavy research lab with an expanding cost base through public-market scrutiny while still in the phase where spending is the deliberate strategy. Every comparable company that has done this — Amazon, Netflix, Uber — required investors to absorb years of losses against a narrative of structural inevitability. OpenAI is running the same play, but at a scale and with a dependency structure that has no direct precedent.

What the S-1 Will Need to Answer

The confidential filing will eventually become public. When it does, the questions the leaked numbers raise will need answers: the trajectory of the Microsoft cost relationship as Azure contracts are renegotiated, the path from $2 billion monthly revenue to operating breakeven, the treatment of the $41.55 billion restructuring charge across reporting periods, and what happens to valuation assumptions if the compute efficiency curve takes longer than modelled.

The leak has done something useful for the market: it has moved the conversation from whether OpenAI is expensive to how expensive, and on what trajectory. That is the conversation that belongs in an S-1. It has started early.