OpenAI's leaked financials don't show an AI company quietly printing money. They show a company spending at a scale that makes Microsoft, Nvidia and the rest of the infrastructure layer look like the safer side of the trade.
The documents weren't supposed to be public yet, and that's why they matter. Financial figures first shared with the Financial Times by independent journalist Ed Zitron show OpenAI spent about $34 billion in 2025 while booking about $13 billion in revenue. That is a huge business by almost any normal standard. It is also a business still spending far more than it brings in.
You should be careful with the loss number, because this is where the original version of the story went too far. The FT reported that OpenAI's net loss attributable to the company rose to about $39 billion in 2025, but most of that came from a roughly $30 billion non-cash accounting charge tied to its old investor structure before the company became a public benefit corporation. Strip out that charge and other non-cash expenses, including stock-based compensation and Microsoft computing credits, and the underlying loss was closer to $8 billion.
That is still a serious number. It just isn't the same story as a clean $20.92 billion operating loss.
OpenAI confidentially filed its S-1 with the SEC on June 8, according to the company's own announcement, giving it the option of a public listing. The Guardian reported that the listing could value OpenAI at more than $850 billion, while the FT said investors are already talking about a possible valuation above $1 trillion. A trillion-dollar IPO candidate with an $8 billion underlying loss is not a normal software company. Frankly, it shouldn't be treated like one.
The useful question isn't whether OpenAI is growing. It plainly is. The FT reported revenue rose from about $1 billion a quarter in late 2024 to $2 billion a month by the end of 2025. Very few companies have ever moved that quickly. The harder question is who gets paid while OpenAI tries to make the math work.
The money is moving upstream
Start with Microsoft. OpenAI runs heavily on cloud computing, and Microsoft has spent years building its AI pitch around that relationship. Windows Central reported earlier this year that Microsoft had a $625 billion cloud backlog, with about 45 percent tied to commitments linked to OpenAI. You don't need to believe every AI demo in a keynote to understand what that means. When OpenAI trains models, serves ChatGPT users or sells enterprise access, a large part of the bill runs through infrastructure.
Microsoft's position is not risk-free. Investors have already questioned whether the company is spending too aggressively on data centers and AI capacity. A customer that burns billions can become a burden as well as a prize. But the basic shape of the deal is obvious: OpenAI is trying to build the most valuable model company in the world, and Microsoft gets paid to provide part of the machinery.
Nvidia's role is cleaner. The chipmaker sells the scarce hardware that model labs and cloud providers need. In September 2025, Nvidia and OpenAI announced plans for a massive AI infrastructure partnership that was described as reaching up to $100 billion and at least 10 gigawatts of Nvidia systems. Business Insider later reported that Jensen Huang said the full $100 billion investment was probably not in the cards, with Nvidia instead finalizing a $30 billion investment in OpenAI. Even at the lower figure, the point holds. Nvidia is not waiting for OpenAI to prove that frontier models can produce software margins. It sells the hardware required to keep the race going.
That is why the leaked financials are more useful as a map of the AI economy than as a scandal about OpenAI alone. The company spent about $19 billion on research and development in 2025 and nearly $6 billion on sales and marketing, according to the FT. Some of that bought researchers, product work and customer growth. A large share also bought compute, and compute has suppliers.
If you're an investor, founder or operator trying to read the AI market, this distinction matters. The model lab gets the headlines because ChatGPT is the product people touch. The infrastructure companies get the revenue because every query, training run and enterprise rollout consumes capacity somewhere. That doesn't make OpenAI weak. It makes its suppliers powerful.
OpenAI can still win. Revenue at a $24 billion annualized pace by the end of 2025 is not a toy business, and a public market listing would give it another way to fund the next stage. But the leak makes one thing harder to ignore: frontier AI is not behaving like a classic software business yet. The gross margin story may arrive later. The capex bill is here now.
That is the real lesson from the documents. OpenAI may become the company public investors fight to own, but the companies selling cloud capacity, chips and data center equipment are already collecting. In an arms race, the weapons maker doesn't have to win the war. It has to keep the orders coming.
Also read: Researchers have finally worked out why AI models keep inventing the same fake names, Japan's chip equipment giants are losing China and betting on AI to fill the gap, and OpenRouter's Fusion API bets the future of AI belongs to panels, not single frontier models
