The financials leaked on June 16, passed to the Financial Times and to the writer Ed Zitron, and within an hour the internet had already split into its two usual camps. One side reading “$38.5 billion loss” as proof the whole thing is about to fold. The other waving it away as the cost of building the future. Both are missing what the document actually says, which is more interesting than either.
Here is the short version, then the part worth sitting with.
What the leak actually shows
In 2025, OpenAI booked $13.07 billion in revenue. The year before, it was $3.7 billion. So the top line more than tripled in twelve months, which is one of the fastest revenue ramps any company has ever posted. That is not a detail to skip past. Whatever else is true, a lot of people are now paying OpenAI real money, and far more of them than last year.
Against that revenue, total costs came in around $34 billion, for an operating loss of roughly $20.9 billion. The headline everyone screenshotted, the $38.5 billion net loss, is a bigger number than the operating loss because it sweeps in a pile of non-cash items on top.
The single line that tells you the most: research and development cost $19.18 billion. More than the entire company made. OpenAI spent more teaching the next models than it earned from all the current ones combined. Add $5.7 billion on sales and marketing and you have a company buying growth and capability with both hands.
The $38.5 billion number is not the cash burn
This is the bit the panic headlines bury, and it is the bit that actually matters.
That $38.5 billion net loss includes things that never left the building as cash. Stock-based compensation, the value of computing credits from Microsoft, other accounting charges that show up as “expenses” without a wire transfer attached. Strip those out and the real out-of-pocket cash burn for the year lands somewhere around $8 billion.
Eight billion is still an enormous amount of money to set on fire in a year. Nobody should pretend otherwise. But $8 billion in actual cash burned is a very different animal from $38.5 billion, and if your read of OpenAI’s health is based on the bigger number, your read is wrong. The scary figure is doing a lot of work that the cash-flow statement does not back up.
There is a useful habit hiding in here for anyone evaluating any AI company this year. When a number is designed to be screenshotted, find out what it includes before you repeat it. The gap between “net loss” and “cash burned” is where most of the bad takes live.
The trend line is the actual story
One number moved in the right direction, and it is the one I would watch.
In 2024, OpenAI spent $2.37 to generate every $1 of revenue. In 2025, that fell to $1.60 for every $1. Still upside down, still losing money on every dollar, but losing meaningfully less of it. The unit economics are improving, not collapsing.
That matters because it changes the question. The question is no longer “is this a company or a bonfire.” The question is whether that ratio keeps falling toward and past the line where a dollar in costs less than a dollar to earn. OpenAI is forecasting $25 to $30 billion in revenue for 2026, roughly double again. If the efficiency trend holds while revenue doubles, the loss starts to look like a phase rather than a permanent feature.
If it does not hold, the conversation changes very quickly, and so does the price of everything built on top.
What this means for the tools you actually pay for
You did not come here for a valuation debate, so here is the part that touches your wallet.
The AI tools you use, whether you pay OpenAI directly or pay a smaller company that quietly runs on OpenAI underneath, are being sold to you below cost. That $1.60-to-earn-$1 ratio is partly your subscription being subsidised by investors and by Microsoft. It is a good deal for you right now, in the same way an introductory mortgage rate is a good deal right up until it resets.
Two practical consequences worth holding in mind.
First, prices on AI tools have more room to rise than to fall, at least at the frontier. The pressure of these losses points one way. The counterweight is cheap competition from labs willing to undercut, which is exactly why the MiniMax price war matters more than it looked at first. Cheap challengers are the only thing keeping the subsidised incumbents honest on price.
Second, OpenAI is not going anywhere in the near term. A tripling revenue line and Microsoft standing behind the compute bill is not the profile of a company about to vanish and take your workflow with it. If you have built something on its models, you do not need to panic-migrate on the back of this leak. You do need to make sure you could move if you had to, which was good practice before the financials leaked and is good practice now.
If you are weighing which model to actually build on, costs included, our LLM comparison for business is the less dramatic place to start.
The honest verdict
OpenAI lost a staggering amount of money in 2025. It also tripled its revenue and got noticeably more efficient at turning spend into sales. Both things are true at once, and any take that needs you to forget one of them to believe the other is selling you something.
The $38.5 billion is real on paper. The $8 billion is what actually left the bank. The $1.60 is the number that decides how this ends. Read in that order, the leak is neither a death notice nor a victory lap. It is a company spending the GDP of a small country to find out whether it has a business, and getting closer to the answer than it was a year ago.
Frequently asked questions
How much did OpenAI lose in 2025?
OpenAI recorded a net loss of about $38.5 billion in 2025, on revenue of $13.07 billion, according to financials reported by the Financial Times and the writer Ed Zitron. The operating loss was roughly $20.9 billion. After stripping out non-cash items, the true out-of-pocket cash burn was closer to $8 billion.
Why is the $38.5 billion loss misleading?
It is not fake, but it includes large non-cash expenses such as stock-based compensation and the value of Microsoft computing credits, which inflate the accounting loss well beyond the actual cash spent. The real cash burn for the year was around $8 billion.
How much revenue does OpenAI make?
OpenAI generated $13.07 billion in revenue in 2025, more than triple the $3.7 billion it made in 2024. The company is forecasting $25 to $30 billion in revenue for 2026.
Is OpenAI going bankrupt?
Nothing in the leaked financials suggests imminent failure. Revenue is tripling year on year, unit economics are improving, and Microsoft continues to back the company’s compute.
Will AI tool prices go up because of this?
Frontier AI pricing has more room to rise than fall, because current pricing is partly subsidised by investors and partners. The main thing holding prices down is aggressive competition from cheaper labs.
One short, honest email a week: what’s real, what’s hype, and what’s worth your money. Subscribe and get my full tool-stack guide, free.
Get the free guide