Recently, an article titled “How the AI ‘bubble’ compares to history” was published in the Financial Times; its author, Jonathan Vincent, reminds us of economic crises in previous years and the symptoms that preceded them. The discussion is primarily about overheated expectations and the subsequent collision with reality. What was said by Jonathan strongly resonates with the doubts expressed to us by futurologist and visiting researcher at the Global Catastrophic Risk Institute (GCRI) Siarhei Besarab in a recent interview. We felt it was important to revisit the dangers of flying on the crest of the AI hype wave.
Thus, the brief essence of the article in the Financial Times lies in the fact that the valuation of American stocks is currently higher than before the Wall Street crash in 1929. It is higher than before the mortgage crisis of 2008 and on the same level as the moment before the “dot-com crash.” From the point of view of the price-to-earnings ratio, as well as the ratio of market capitalization to GDP, it turns out that we are at historical extreme levels. Some investors seriously fear that this will end in big problems for the market. True, there are also those who are convinced—current valuations are justified by the rapid growth of profit and the revolutionary role of AI.
The British Office for Budget Responsibility notes: AI is indeed perceived as a transforming technology with rich potential, but the real effect from it still remains extremely uncertain. Jonathan Vincent recalls nine studies estimating the contribution of AI adoption to productivity growth, the results of which showed extremely different results: from a “slight acceleration” to more than doubling of recent growth rates.
Also, the journalist draws our attention to the fact that “since the launch of ChatGPT in October 2022, the S&P 500 capitalization relative to US GDP has grown from 142% to a record 214%, and the share of the tech sector in the index’s capitalization has more than doubled — from 44% to 101%.” Critics note that this makes the market dependent on the results of a small group of companies.
We asked Siarhei Besarab to disclose in more detail the essence of the accumulated risks.
“Today, stock prices have flown into space relative to the indicators of the real profit of companies,” clarifies Siarhei. “The seven largest companies engaged in AI (Nvidia, Microsoft, Apple, Amazon, Alphabet, Meta, Tesla) make up 35% of the entire S&P 500 index. 75% of the entire growth since October 2022 falls on these seven stocks. A so-called ‘monoculture,’ where an error in the strategy of one corporation or disappointment in one quarterly report will give an avalanche-like domino effect for the entire financial system.
In fact, these companies are trading only based on hopes for ‘explosive growth in the future thanks to AI.’ Only hopes; a part of these companies are already unprofitable now. The entire history of humanity says that in mathematics and economics there is no place for miracles in principle. Miracles — this is about the dreams of the buyer of those very stocks. Imagine for a minute, the entire previously pragmatic and rational market today holds on only by faith (!).“
US GDP growth in the first half of 2025 was entirely provided by investments in IT equipment and data infrastructure. Without taking into account investments in AI and “hardware,” the real economy is in stagnation or depression.
Here, Sergei returns to Jonathan Vincent’s article and notes that the analogy with the “railway mania,” when in the 1840s rails were laid anywhere, including villages and small towns, is very illustrative. But there was simply no one to ride on these roads.
“We are observing the same thing now, when the excess capacities utilized for so-called AI do not have tangible solvent business models,” Sergei is confident. “For the last couple of years, I have constantly heard this remark in comments on Altman’s speeches. In our article about the bubble, I proposed AdsGI (implementation of advertising) as a business model, but relative to the aforementioned sums, I understand that this is an absolute drop in the ocean. The only plus in this story is that the hardware, physical infrastructure, and data centers will remain after any collapse. But the investors will turn into dust.”
Siarhei Besarab develops some ideas described in the Financial Times article.
First, in this entire story, there is a quite tangible limitation—electricity. It does not matter how many trillions Nvidia is worth; if power grids cannot pull the chips created by it, then there is no sense in the capitalization. So the bubble may burst not due to a “collapse of faith,” but due to a banal deficit of TWh.
Sergei describes a situation that so far seems unrealistic: “At the end of last year, I was compiling forecasts for 2026 and in the ‘energy’ item I noted for myself the possibility of ‘power supply prioritization,’ when, roughly speaking, houses in the suburbs will be disconnected to ensure the uninterrupted operation of the training cluster of a new neural network. And there is also the problem with access to HALEU for SMRs (which all these CEOs of the companies listed at the beginning so shamefully hope for), the problem of cooling all these data centers (which means brewing competition for access to glaciers, deep-water Arctic currents, and air corridors for heat discharge), etc.“
Also, commenting on the huge spread of productivity estimates (from 0.01 to 3.5%), Sergei expresses doubt that economists currently have an objective methodology for estimating this very productivity.
“They are trying to measure the economy of the post-industrial information society with a ruler of the 19th century, of the society of the beginning of industrialization,” he points out to us. “As I love to say, ‘there are no inexplicable phenomena, there are simply insufficiently precise measuring instruments.’ The spread of figures in estimates may be because we do not have instruments for estimation. It is like measuring humidity with a thermometer.“
As another important accent, Sergei reminds us that President Trump will be with us for another 4 years, which means one cannot exclude from the context his idea about the arms race with China. That is, these are no longer just games of civilians; this can become a “military-political scenario.” Even if profitability collapses, the state (in theory) can continue to subsidize the dead sector for the sake of “national security.” One can assume that the neural network market has become “too big to fail,” has merged with the state.
So, where is everything heading? In the opinion of Siarhei Besarab, there are in total only two scenarios. Two equally bad scenarios.
The first is if the bubble continues to inflate. The US Federal Reserve will begin to actively lower rates in 2026, and new infusions of liquidity will go. The circular guarantee of mutual financing will continue; the market will again break records.
But one day, “unexpectedly,” in some 2027 or 2028, they will be overtaken by a quite expected result. Someone from big business will be found who is capable of seeing and evaluating reality: productivity is on a plateau, profit is overvalued hundreds of times, they bought not a living product but faith in it. One thing will collapse, and after it, like domino tiles, everything else.
The second — the bubble will burst right now. A detonator can serve as, for example, some geopolitical event, a recession, a monstrous (truthful) financial report of one of the AI giants, etc. This causes a market collapse; in a matter of days, up to 75% of profit evaporates, indices lose up to 50%. Since the seven “AI giants” make up 35% of the market — the entire market will collapse. After it, a banking crisis will come with deregulation and overloaded debts. The recession in this case can quite possibly turn into the first “Great Depression” in the 21st century.
