Did OpenAI 'Burn' $34 Billion Last Year? What the Leaked 2025 Financials Actually Show

Leaked 2025 financial documents — described as audited, obtained by independent journalist Ed Zitron and verified by the Financial Times — indicate that OpenAI booked roughly $34 billion in total costs and expenses last year against about $13 billion in revenue, with a headline net loss of around $38.5 to $39 billion. Those figures, first detailed by Zitron on his newsletter Where's Your Ed At and amplified across outlets including The Decoder, Cryptobriefing, TechTimes and Benzinga in mid-June 2026, are startling at a glance. But the popular shorthand — that OpenAI "burned through $34 billion" — packs several distinct accounting concepts into one scary number, and getting them wrong changes the story. This article separates the three things people keep conflating: total spending, net loss, and actual cash burn. All figures below are as reported by the cited sources and have not been independently verified by comparee.ai; where the sources disagree, we say so.

What the $34 billion number actually is

The single most important clarification is also the simplest: the $34 billion is OpenAI's total costs and expenses for the year, not a measure of cash "burned." In Zitron's breakdown of the leaked documents, that $34 billion line is labeled "costs and expenses" and is split into categories — most prominently research and development, plus sales, marketing and other operating costs. Spending money is not the same as setting it on fire. A company that spends $34 billion and earns $13 billion has, at most, a roughly $21 billion operating shortfall on paper, not a $34 billion cash bonfire. The "$34 billion burned" headline, which even The Decoder uses in its title before clarifying in the body that OpenAI "spent" the money, is a vivid but imprecise framing.

Why does this matter? Because "burned" implies cash that has permanently left the building, and spending of this size includes a mix of cash outlays and non-cash items, plus revenue flowing back in. The honest version of the sentence is: OpenAI's expenses far outran its revenue in 2025, and the company is deeply unprofitable on an operating basis. That is genuinely alarming for anyone betting on near-term profitability. It is simply not the same claim as "$34 billion of cash vanished." Keeping those two ideas separate is the whole point of reading financial statements rather than headlines.

The $39 billion loss, and the asterisk that swallows it

The other big number — a net loss reported at roughly $38.5 billion (Zitron's documents cite $38.53 billion) and rounded by some outlets to "about $39 billion" — looks even worse than the spending figure. It would represent an enormous jump from 2024, when OpenAI's net loss was reportedly around $5 billion, an increase of nearly eightfold year over year. Taken at face value, that trajectory looks like a company spiraling. But the face value is exactly what the sources warn against.

The reason the net loss balloons so far past the operating shortfall is a one-time, non-cash accounting charge tied to OpenAI's restructuring from a nonprofit-controlled entity into a for-profit structure. Before the switch, certain investors held convertible interests and warrant-like rights that, under US accounting rules, were treated as liabilities and revalued upward as OpenAI's valuation soared. When the company's worth jumped, the paper value of those obligations jumped too — generating a massive charge that flows through the income statement but involves no money actually leaving OpenAI. Crucially, the sources stress this charge is not expected to recur after the restructuring.

How big is the charge? This is where the reporting genuinely diverges, and readers should be careful. Several outlets, including The Decoder and the Financial Times summaries, describe a roughly $30 billion non-cash charge, after which OpenAI's loss "excluding restructuring charges and other non-cash items" comes out near $8 billion. Zitron's own line-item reading of the documents, however, cites a larger "$41.55 billion loss from changes in fair value of convertible interests and warrant liability," which is part of why some derivations of the underlying loss differ. The takeaway is consistent even when the exact figure is not: a very large slice of the reported loss is a non-cash valuation adjustment, and stripping it out leaves a far smaller — though still substantial — underlying loss.

Cash burn vs. net loss vs. spending: three different things

It is worth slowing down on the terminology, because the entire interpretation of OpenAI's year hinges on it. Total spending (costs and expenses) is what the company laid out to operate: about $34 billion. Net loss is an accounting result — revenue minus all expenses, including non-cash items like depreciation and that restructuring charge — reported at roughly $38.5 to $39 billion. Cash burn is yet another concept: the actual net cash leaving the business after accounting for non-cash charges, working capital and financing. None of these three numbers is interchangeable, and the leaked documents as reported speak most directly to the first two, not to a clean cash-burn figure.

So when someone says OpenAI "burned $34 billion," they have arguably picked the wrong number for the word they used. The $34 billion is spending, not burn. The closest thing to a true cash-shortfall signal in the reporting is the gap between spending and revenue — and even that overstates cash burn to the extent that some costs are non-cash and that OpenAI raised vast amounts of capital to fund the gap. The practical lesson is to treat the dramatic single number with suspicion and ask which of the three concepts it describes before drawing conclusions about how close OpenAI is to running out of money.

Where the money went: R&D, sales, and Microsoft

The composition of OpenAI's spending tells you a lot about its strategy. The largest category by far is research and development, reported at roughly $19 billion (Zitron cites $19.18 billion) — well over half of total costs. That is the price of training frontier models, paying elite research talent, and buying or renting the enormous compute those models require. Sales and marketing, plus other operating items, came to nearly $6 billion (Zitron cites $5.73 billion). For a company often caricatured as a pure research lab, the near-$6 billion go-to-market spend signals an aggressive push to convert mindshare into paying customers ahead of a public listing.

One line item drew particular attention: OpenAI's dependence on Microsoft's Azure cloud. TechTimes, summarizing the audited documents, highlighted about $17.2 billion in total expenses flowing to Microsoft across categories — a striking concentration of cost with a single infrastructure partner and investor. If accurate, that figure underscores how much of OpenAI's spending is, in effect, a compute bill, and how tightly its economics are tied to the cost and availability of cloud capacity. It also helps explain why OpenAI has publicly pursued additional compute partners and its own infrastructure ambitions: reducing single-supplier exposure is both a cost lever and a strategic one.

The growth case behind the losses

None of this is to say the losses are trivial or that OpenAI is obviously healthy. But the bull case is also visible in the same documents. Revenue of about $13 billion reportedly beat the company's own internal target of roughly $10 billion and represented growth on the order of 3.5x versus the prior year. That is an extraordinary top-line trajectory for a company at this scale, and it is the reason investors have been willing to fund enormous losses. As reported, OpenAI completed a funding round in early 2026 raising a very large sum and carrying a valuation in the hundreds of billions, with a public offering said to be in the works.

This is the classic high-growth, high-burn pattern that has defined frontier AI: pour money into research and compute now to build a durable lead, on the bet that revenue and eventual margins catch up. The leaked numbers are consistent with that playbook — losses widening sharply even as revenue more than triples. Whether the bet pays off depends on questions the financials cannot answer: how fast model-serving costs fall, whether OpenAI can hold pricing power as rivals like Anthropic, Google and open-weight models compete, and how durable enterprise demand proves to be. The financials describe the spending; they do not guarantee the payoff.

Why the framing matters for an IPO

The timing is not incidental. These numbers surfaced as OpenAI is widely reported to be preparing for an IPO, and pre-IPO financials are scrutinized precisely because the framing shapes investor perception. A "$39 billion loss" headline reads like a company in crisis; a "roughly $8 billion underlying loss on $13 billion of fast-growing revenue, with a one-time non-cash charge from restructuring" reads like an aggressive growth company. Both descriptions can be drawn from the same documents — which is exactly why understanding the accounting is not pedantry but the core of the story.

For prospective public investors, the distinctions are decision-relevant. The non-cash restructuring charge will not recur, so it should not be projected into future years. The operating loss and the R&D-heavy cost structure, by contrast, are recurring and central to the investment thesis. And the cash question — how long OpenAI can sustain its spend, and on what terms it raises capital — depends on real cash flow, not the inflated accounting loss. Anyone evaluating OpenAI on the basis of one dramatic number is likely to misjudge it in one direction or the other. The responsible read is the boring one: large, growing revenue; large operating losses; an enormous but non-recurring accounting charge; and heavy dependence on compute spending.

The numbers, as reported

Below is a compact summary of the key figures and how the sources characterize them. Where sources disagree, the range is shown. None of these have been independently verified by comparee.ai:

The bottom line

The most accurate one-sentence summary of OpenAI's leaked 2025 financials is not "OpenAI burned $34 billion." It is: OpenAI spent about $34 billion to generate about $13 billion in revenue, posted a headline net loss near $39 billion that is dominated by a one-time non-cash restructuring charge, and would show an underlying loss closer to $8 billion once that charge is stripped out. Every word in that sentence is doing work, and dropping any of it produces a distorted picture — either too rosy or too apocalyptic.

That nuance does not let OpenAI off the hook. An operating loss in the low twenties of billions, an R&D bill of $19 billion, and a $17 billion-scale dependence on a single cloud partner are serious facts about a business that is not close to profitability and is leaning hard on outside capital. But they are different serious facts than "a company that incinerated $34 billion in cash." As OpenAI moves toward a public listing, expect both framings to be deployed — by bulls and bears alike — and treat any single dramatic number with the skepticism it deserves. The documents reward careful reading; the headlines, less so.

Disclaimer: based on leaked financial documents reported by Ed Zitron (Where's Your Ed At) and said to be verified by the Financial Times, as summarized by The Decoder, Cryptobriefing, TechTimes and Benzinga, all linked below. Figures are as reported, sometimes differ between sources, and have not been independently verified by comparee.ai. This is reporting and analysis, not financial advice.