The AI Economy: Why Market Leaders Aren't Even Trying to Turn a Profit

If you thought Uber or WeWork at their worst were masters of burning money, the artificial intelligence industry would like you to hold its bottomless wallet. Today the market's biggest players, from OpenAI to Anthropic, don't even pretend to aim for operating profit in the foreseeable future. We are watching a unique moment in tech history: companies valued in the hundreds of billions of dollars openly admit that their model-training costs will always outpace their subscription revenue. This is not a temporary hardship but a fundamental feature of the current arms race, where the winner will not be whoever sells best, but whoever can longest ignore an emptying bank account.

Context matters more than any figure here. To grasp the scale of what is happening, recall how Silicon Valley worked for the past decade. Typically a startup burns cash to capture a market, then starts monetizing a loyal audience.

With AI, that scheme has broken. The cost of training each new version of GPT or Claude grows exponentially. If training GPT-4 cost around one hundred million dollars, the next iterations require billions.

The lion's share of that money goes not to programmers' salaries but into the pockets of Nvidia for chips and of energy companies for electricity. We have ended up in a situation where the product gets more expensive to make faster than the market can get used to its current price.

So why do investors keep writing checks for billions of dollars while looking at these financial ruins? The answer lies in a blind faith in AGI, artificial general intelligence. Behind the scenes at OpenAI and Google DeepMind, the prevailing belief is that once they build a system able to replace humans in most cognitive tasks, the question of money will resolve itself.

Such a system, supposedly, will figure out how to earn trillions, optimize the economy and clear all debts. It is a kind of technological messianism: we are building a digital god, and a god does not need quarterly reports. The catch is that you have to survive until that moment, and the price of admission to this exclusive club rises every day.

Meanwhile, the business models on offer today look more like an attempt to soothe shareholders' consciences than a real plan. Twenty-dollar-a-month subscriptions do not even cover the depreciation of the servers these models run on. Enterprise deployments are slow because of security concerns and AI hallucinations. The result is a paradox: a technology that is supposed to automate and cheapen everything under the sun is itself the most expensive and least efficient business in the world. Companies must raise funds constantly just to keep the lights on in their server rooms, creating a dependence on venture capital that resembles a pyramid scheme built on very smart algorithms.

The parallel with earlier tech booms is obvious, but there is one significant difference. In the dot-com era, companies spent money on marketing and capturing attention. In the AI era, money goes to physical infrastructure and raw compute. That would make a crash, if it comes, far more painful for the whole economy. If investors decide tomorrow that the road to AGI is too long, we will be left with more than bankrupt websites: mountains of expensive hardware and giant data centers that consume the energy of entire cities. The irony is that market leaders make no attempt to hide this state of affairs, openly stating that they need hundreds of billions in investment to keep going.

What does this mean for us? We use incredibly expensive tools practically for free while venture funds and Microsoft pick up the tab for the banquet. It is a golden age for users but an alarming signal for the industry. The long-term sustainability of AI companies now depends not on code quality but on geopolitics, lithium supply and the patience of the planet's biggest investors. As long as they believe in a miracle, the party goes on. But the moment someone first asks "where's the money?", the rules of the game will change instantly, and many will have to remember what traditional economics is, with its boring notions of profit and loss.

The bottom line: today's AI industry is not a business in the classic sense but a grandiose scientific experiment run on other people's money. Will the models become smart enough to pay for their own existence before investors run out of patience?