OpenAI's Math: How to Earn Billions and Stay Deep in the Red

Sam Altman is looking for money again, and that is hardly news anymore; it is more like background noise in the industry. OpenAI's finances recall the old joke about a businessman who buys eggs for one ruble, boils them and sells them for one ruble, insisting that at least he is always busy. In reality, the picture is both more complicated and far more interesting.

Look closely at the company's cash flows and something surprising emerges: operationally, OpenAI is perfectly viable. Every model it has released now brings in enough subscription and API revenue to fully cover its compute costs and even its own research. The catch is that in generative AI you cannot simply ship a successful product and quietly collect the profit.

The industry lives in a mode of permanent, extremely expensive upkeep. As soon as GPT-4 started producing steady cash, all of that money moved straight into the GPT-5 training budget. And when it predictably fell short, the company had to go cap in hand to Microsoft and the big venture funds again.

This is the classic scaling trap taken to the absolute: you have to run flat out just to stay in place, and keep increasing your speed while you do it. Silicon Valley is used to unprofitable startups that burn capital for years to capture a market. Amazon did it for decades, and Uber taught us that profit is something out of history textbooks.

But OpenAI has a wrinkle: its main asset is not a loyal user base that is easy to monetize, but computing power, which gets more expensive with every new generation of chips.

Investors are now making a colossal bet that a plateau will eventually arrive. That blessed moment when the next iteration of a model becomes so refined that improving it no longer requires exponential growth in spending on hardware and electricity. For now, though, the scaling law dictates the opposite logic: to get a modest gain in quality, you need to invest ten times more resources than last time.

That turns the business model into a kind of technological pyramid, where today's well-being rests entirely on the expectation of a miracle from the next version. If GPT-5 does not deliver a qualitative leap that justifies tens of billions of dollars in costs, even Altman's most devoted supporters will start asking questions.

Meanwhile, the company's revenue is growing at a pace any software giant would envy. ChatGPT Plus subscriptions and enterprise contracts generate huge sums. But those figures look almost comical next to the bills from Nvidia and the plans to build data centers whose power consumption is already measured in gigawatts.

Altman is building a future in which AI becomes the new oil, but for now he looks more like the owner of an oil rig who spends all the oil he extracts on drilling the well even deeper. The dependence on Microsoft plays a key role here: the Redmond giant effectively hands OpenAI "vouchers" for its cloud capacity in exchange for a stake in the company, creating a closed ecosystem in which real cash is just one element of the equation.

The risk is not that investors will run out of money; there are still plenty of people eager to touch the "creators of the future." The main danger lies in a possible intellectual dead end. If it turns out that simply adding GPUs and internet text no longer produces magic, OpenAI's entire financial structure will collapse in on itself.

We are watching the most expensive experiment in human history, where the grand prize is a full-fledged artificial intelligence and the price of losing is the bankruptcy of the most talked-about company of the decade. For now, Sam Altman is successfully selling hope, but math is stubborn, and sooner or later the models will have to learn to earn money faster than they become obsolete.

Bottom line: OpenAI's business model is a race against its own shadow. As long as each new development requires more resources than the previous one brought in, the company remains hostage to the market's faith in an inevitable technological miracle.