OpenAI plans to burn 278 billion dollars before it turns a profit
Key takeaways
- OpenAI's own July projections show negative free cash flow of 278 billion dollars across 2026 to 2030.
- Compute and infrastructure account for roughly 856 billion dollars of spending through 2030, the single largest line in the plan.
- The 122 billion raised in March is modelled to run out as early as 2028, which is why a round at 1.2 trillion or more is already being discussed.
- Consumer and API pricing is the first place a funding stumble would show up.
278 billion dollars. That is the negative free cash flow OpenAI projects for itself between 2026 and 2030, according to a private presentation from July that the Financial Times reported on 18 September.
The number is not an outside estimate or a short seller's model. It is the company's own arithmetic, shown to prospective partners.
What the OpenAI cash burn to 2030 actually covers
Revenue in the same deck grows nearly tenfold, from about 36 billion dollars this year to 350 billion in 2030. That is a rate of growth almost no company has sustained at that scale, and it still does not close the gap.
The reason is one line item. OpenAI forecasts roughly 856 billion dollars on compute and infrastructure through 2030. Training runs, inference capacity, data centre leases and the power to run them. Everything else in the business is a rounding error next to it.
Set the two figures side by side and the shape becomes clear. The company expects to sell an enormous amount of software and spend more than twice as much buying the machines to serve it.
The 2028 problem
OpenAI raised 122 billion dollars in March. On the projections in this deck, that money is exhausted as early as 2028.
So the funding conversation has already started. Investors are reportedly discussing a fresh round at a valuation of at least 1.2 trillion dollars, which would be the largest private raise ever attempted by a wide margin. Nvidia's own 279 billion dollars in purchase commitments through 2027 sits on the other side of the same trade, since much of OpenAI's compute spending ends up as somebody else's booked revenue.
What is unusual here is not that a fast-growing company loses money. Amazon did it for years. It is that OpenAI is showing partners a plan in which it loses money deliberately for five consecutive years and asking them to underwrite the whole stretch in advance.
Why this reaches your bill
Every price you currently pay for a frontier model is set inside this structure. API pricing, consumer subscriptions, the free tiers that competitors have to match: all of it is subsidised by capital that has not yet had to justify itself against profit.
That subsidy is what keeps four frontier models within touching distance of each other on price. It is also why per-token pricing has kept falling while capability has climbed, which is not how most industries behave.
If the funding rhythm slips, even briefly, pricing is the first lever available. Not model quality, not availability. Price.
The part worth watching is the 2028 line rather than the 278 billion headline. A company can burn almost any amount of money as long as the next round arrives on schedule. The deck is really a statement about how many rounds have to arrive, and how big each one has to be.
