Trillion-dollar AI fire sale: investors suddenly sobered up

Remember that almost religious fervor of late 2022? Every presentation with AI in the title added billions to company valuations. We lived in a beautiful world where Nvidia graphics cards were considered the new oil, and language models were the philosopher's stone. But this week the fairy tale ended. The stock and bond market was hit by a wave of sell-offs that wiped a trillion dollars of market value. This is not just a technical pullback or seasonal profit-taking, this is a full hangover after a three-year party where champagne flowed at the expense of future achievements.

Let's be honest: we all expected this, though we hoped the celebration would last a little longer. You can't endlessly inflate a bubble with expectations that aren't backed by quarterly earnings. While Microsoft, Google, and Meta report record capital expenditures on building data centers and buying chips, investors are beginning to look nervously at the clock. The question "Where's the money?" has finally replaced "What else can your neural network do?" The trillion-dollar collapse showed that the market is no longer willing to pay in advance for promises to change the world over an indefinite period of time. Now everyone is only interested in operating profit here and now.

The problem is that AI infrastructure costs astronomical sums. To train a single model at the level of GPT-5 or its equivalents requires hundreds of thousands of chips, gigawatts of energy, and an army of engineers with astronomical salaries. Companies are spending tens of billions today hoping to make trillions tomorrow. But that "tomorrow" keeps getting pushed back. Corporate clients are in no hurry to implement raw and expensive tools into their business processes, while ordinary users are not yet willing to pay for subscriptions enough to cover at least cloud computing costs. This gap between expenses and actual revenue has become too obvious to ignore.

This week was a moment of truth for many technology giants. If previously any company that mentioned neural networks in a press release got an automatic bonus to its stock value, now the rules of the game have changed completely. Investors are punishing even those showing growth if that growth seems to them insufficiently "artificial-intelligent" or too expensive to maintain. We are seeing a classic crisis of confidence that usually precedes the maturation stage of any breakthrough technology. Euphoria gives way to sober calculation, and not all players will pass this stress test.

What does this mean for the industry as a whole? Most likely, the golden age of free and unlimited neural networks is coming to an end. Companies will be forced to tighten their belts, introduce strict limits, and find ways to monetize literally every user request. The era of "burning money for reach" is rapidly giving way to an era of harsh efficiency. Those startups and giants that cannot prove their usefulness in a specific dollar equivalent in the coming quarters risk simply disappearing, despite all their technological advancement and the number of parameters in their models.

It's interesting to watch how the rhetoric of top managers is changing. We used to be told inspiringly about technological singularity and the complete replacement of human labor. Now the CEOs of the largest corporations on investor calls increasingly use down-to-earth words like "optimization of internal processes" and "reduction of operating costs." The romance of the future has given way to boring accounting. And this is perhaps the best scenario for the industry in the long term. Artificial intelligence needs to stop being an expensive toy and become a real tool of the economy. Even if the path to this lies through a bloodbath in the stock market.

The bottom line: The market stopped trusting on faith and giving credit under the AI abbreviation. Now technologies will have to prove their profitability, otherwise trillion-dollar losses will be just the beginning of the end of a grand illusion.