The artificial intelligence bubble and the repetition of an old market mistake

Every major technological transformation carries a paradox, where while it is inevitable, it is also overestimated in the short term. Artificial intelligence seems to have reached exactly that point, not because it is fragile or fleeting, but because it has been elevated too soon to the condition of unavoidable destination.

The question, therefore, is not whether AI is relevant, this is already resolved. The most honest question is whether the market is managing to separate infrastructure from euphoria, real narrative value, and concrete result of well-packaged promises.

History offers a parallel to this scenario, where at the end of the 19th century, the railroads symbolized the future and investing in rails meant betting on progress. The problem is that at a given moment, it stopped importing where the rails took, it was enough for them to exist. Lines were built without demand, companies emerged without a sustainable business model and wrong metrics began to define success, such as installed and non-passenger kilometers.

Today, the speech is different, but the pattern is repeated with larger models, more parameters and more processed tokens. Sophisticated technical metrics, however, often disconnected from the operational impact. As in the past, progress was measured by the extension of the rail network, innovation is now measured by the model scale, not by the result delivered.

In 2024 alone, global investments in AI startups reached about US$ 110 billion, according to an analysis by Dealroom, data platform and intelligence. These investments were concentrated mostly in initiatives that were still precarious, with unclear return cycles. At the same time, we saw that a portion of companies that started large-scale AI projects were unable to move from pilot to production consistently. This bottleneck is rarely technological, economic, organizational and operational.

This mismatch does not invalidate the technology, on the contrary, just as the railroad bubble burst, investors lost money, companies disappeared and, even so, the tracks remained and became critical infrastructure for the industrial growth of the following decades. The same tends to happen with artificial intelligence.

The biggest risk is not in the eventual market correction, but in the psychological that accompanies the height of any bubble, which is the fear of being left behind. When the discourse becomes “if you do not adopt now, you will become irrelevant”, rationality gives way to haste and strategic decisions are taken based on anxiety, not analysis.

At this point, some questions should precede any major AI initiative, such as: Is there real demand for this application or are we forcing a problem to justify the solution? Is the return on investment measurable or just projected on presentations? Does computational, energy and operational cost talk to the expected benefit? Is there enough governance to deal with risks such as systemic error, model hallucinations and regulatory impacts? Ignoring these issues is putting tracks where there is no route.

It is in this environment of pressure that the difference between those who uses it is formed as a strategic prop and who incorporates it as a structural advantage. Organizations that cross bubbles with maturity are those that treat technology as a means, not an end, connecting it to clear processes, objective indicators and concrete business decisions. Understanding that smart automation is not about replacing everything, but about orchestrating better what already exists.

Artificial intelligence will, indeed, redefine operations, productivity and decision models, but not in the magical way that many narratives suggest. Just as the trails that really thrive were those connected to cities, industries and people, the AI that will survive will be connected to real problems, clear metrics and sustainable results.