OpenAI, the company behind ChatGPT, earns billions and loses bigger billions at the same time. Both are true. Here is the maths, in plain words.
Openai-money-summary and losses
Money in
2023: about 1 to 2 billion dollars. 2024: 3.7 billion. 2025: 13.1 billion. 2026: about 30 billion expected. 2027: about 62 billion planned.
Where it comes from: over 50 million people pay 20 dollars a month for ChatGPT Plus. More than 9 million office workers use paid business plans. Companies also rent OpenAI’s models for their own apps (the API). New income from ads and gadgets is just starting.
Money out
From the account books. 2024: costs about 9 billion, loss about 5. 2025, audited: costs 34 billion, loss about 21. 2026: costs projected around 56 billion. 2027: around 122 billion, when giant computing bills land.
Note: the 13.1 billion is OpenAI’s full income, not a share of something bigger. Microsoft takes about a fifth of it, counted as a cost.
Three 2025 numbers, all true
34 billion: costs written in the books. 22 billion: cash that actually left. Smaller, because paying staff in shares costs no cash, and some bills were unpaid at year end, including 3.6 billion owed to Microsoft. 9 billion: cash burn, meaning 22 out minus 13 in. When you read a headline, check which number it uses.
How the “38.5 billion loss” was calculated
Step 1. Earned: 13.1 billion.
Step 2. Spent: about 34 billion. Business loss: about 21 billion. Interest on its cash pile added back about 2.
Step 3. One-time paper charge: 41.5 billion, from converting nonprofit to for-profit. No cash left the company. Like the tax office revaluing your house: the paper changes, your wallet does not.
Step 4. Total accounting loss: 60.35 billion.
Step 5. Partners in joint projects carry 21.82 billion of it. 60.35 minus 21.82 = 38.53 billion. That is the headline.
Reality: real cash lost in 2025 was about 9 billion. Huge, but not 38.5.
Two words that decode every headline
Cash burn: real cash lost in one year.
Cumulative: years added together. One famous headline gets these wrong: the “665 billion through 2030” figure is total spending on training and running AI models, a costs number, not burn.
The burn figure is separate: about 218 billion of cash shortfall from 2026 to 2029 (25 + 57 + roughly 136 in 2028-29). OpenAI expects 2030 to be its first cash-positive year, at plus 39 billion. The next four years decide if that bet pays.
