OpenAI expects to burn through $278 billion in cash between 2026 and the end of 2030 while lifting annual revenue from roughly $36 billion this year to $350 billion, according to internal projections first reported by the Financial Times and corroborated Friday by Bloomberg and Reuters. The figures come from a company presentation prepared in July as part of a computing agreement, per a person familiar with the matter.

A nearly tenfold revenue climb inside four years isn’t a forecast. It’s a commitment device.

The math is legible enough to sketch on a napkin. RuntimeWire, working from the same underlying reporting, pegs cumulative revenue through 2030 at about $840 billion against $856 billion in projected compute and infrastructure spending. Revenue and inference cost, in other words, are running neck-and-neck for the rest of the decade, which means every non-compute dollar has to be extracted from somewhere the company doesn’t yet fully monetize.

That’s the frame in which the rest of OpenAI’s 2026 makes sense. Conversational advertising inside ChatGPT and the shopping layer built with Amazon aren’t experiments. They’re the monetization vehicles the presentation implicitly requires.

The July presentation itself already reflects a mild upward revision. May’s projection put cash burn at $305 billion; July’s model releases lifted annualized revenue about 20% and trimmed the forecast by $27 billion. The same month, OpenAI cut the price of its GPT-5.6 Luna model by roughly 80%, a move that looks contradictory only if you don’t read it as a share-grab priced against a future in which ads, commerce fees, and enterprise seats close the gap.

The capital stack tells the same story. A $122 billion round anchored in March by Amazon, Nvidia and SoftBank, with continued participation from Microsoft, sealed OpenAI’s $852 billion post-money valuation and is expected to be exhausted by 2028. Early talks with investors now target a valuation above $1.2 trillion, roughly 41% higher than March. Sam Altman told Fortune an IPO remains planned but won’t happen in 2026, citing A.I. safety concerns, which is one of the more elegant framings of “we still need private capital that behaves better than a quarterly earnings call.”

Every prior platform cycle ended the same way. Distribution that felt free eventually got a price. The 2030 number sets that price’s floor.

Sources