OpenAI is spending heavily on a bet that it can build AI that does most work as well as people, known as artificial general intelligence, or AGI. That spending sets up a race between rising costs and growing capabilities.
The business depends on turning use of its models into revenue faster than its computing costs rise. Its models read and write content in small chunks called tokens. OpenAI largely charges by the token for each conversation, piece of code, image or result. Tokens are the meter and computing power is the engine, with a bill that grows with demand. When the product a company sells also drives its biggest cost, growth can punish the company as quickly as it rewards it.
Internal financial projections reported by the New York Times put numbers on that race. The company expected about $3.7 billion in revenue in 2024 and roughly $5 billion in losses. Both figures come from leaked forecasts rather than audited accounts.
The costs follow from how AI models get better. They have historically improved by getting bigger and training on more data with more computing power, a pattern known as scaling laws. The gains are real, but they are not free.
So OpenAI needs more than token sales. Ads and shopping features inside ChatGPT, its AI assistant, are logical responses, though some of those plans were delayed so the company could improve ChatGPT first. It has also explored earning a commission, known as affiliate revenue, when links in a conversation lead to a purchase. Each would raise revenue per user.
Then come the infrastructure commitments. In an interview, Sam Altman, OpenAI’s chief executive, said the company has committed to about $1.4 trillion of infrastructure spending. That money is tied to roughly 30 gigawatts of data center capacity. A commitment that large means constantly raising money or constantly finding new revenue, usually both, so runway and capacity come before neat profits.
Paying for all of it on the promise of AGI is like backing a friend who keeps borrowing money and says the payback will come with the breakthrough. Even if the models keep improving, the world will not switch from “AI” to “AGI” overnight in a way that fixes the business model.
