In 1845, 1,394 companies were created in the United Kingdom whose purpose was the construction of railways. The share price of the sector had doubled in the previous twelve months. The government had authorized the construction that year of some 6,500 kilometers of railway lines, approximately the same length as had been built in the previous twenty years. That ended badly for most investors. To begin with, less than 10% of the companies that were created managed to be awarded a construction project, so the vast majority disappeared in a few months, and very often, their shareholders' money went with them.
It was not an exceptional phenomenon. A current of evolutionary psychology links the success of human beings as a species to their unbridled optimism. In finance, it manifests itself with the investment euphoria that is unleashed around new technologies. It happened with the railroad in the United Kingdom and with the construction of canals in the United States in the mid-19th century. It also happened in the 90s, when the arrival of the Internet triggered the creation of companies with no real activity beyond a website, and caused telecommunications operators to install much more fiber optics than was required in the medium term. As with the railroad, most of these investment rushes ended up hurting many investors.
Does it make sense to put investment in artificial intelligence in this group? According to The Economist, yes, at least when it comes to growth. In its first three years of existence in its current format, companies have multiplied their investment in this technology by five. Only the investment in shipping canals in the 1850s comes close. Investment in AI doubles the growth of the railways and triples that of the internet bubble.
And the data is dizzying. The $450 billion that a handful of companies led by Amazon, Google and Microsoft spent on AI last year will double in 2026, and will increase by 50% next year to reach $1.5 trillion, which is the size of the Spanish economy. To add to the tension, half of the spending this year is going to be financed with borrowed money, almost half a trillion dollars.
Investors' concern grows when they hear these figures. Will future income justify these expenses? Will they be able to make this investment profitable? The answer is complex because there is no solid data on how much money companies earn directly generated by AI, although the range of estimates for 2026 is from $150 billion to $220 billion. That does not cover even a third of the investment that is being made. More worrying is that exponential growth in income is not expected: the University of Texas found that the percentage of workers who use AI on a daily basis fell from a maximum of 45% in 2025 to 33% in 2026.
This cooling comes with a relatively low level of adoption among companies: from 10% in Europe to 30% in the United Kingdom, with the United States at an intermediate level. These numbers are not comparable due to methodological differences, but they illustrate that the slowdown in use surprises us far from universal or intensive adoption. With end consumers reluctant to pay for subscriptions, for the numbers to add up, the industry will need companies to spend generously.
However, it's unlikely that big tech companies are paying much attention to those calculations. They believe, probably rightly, that being left out of the revolution would mean its disappearance or, at best, its irrelevance. A financier who comes to the management committee with an analysis suggesting, for example, that he stop investing because he expects a return of 8%, lower than the cost of capital of 10%, will receive, at best, indifference. It's about growing or dying.
In this scenario, the investor's success depends on successful selection. History has repeatedly shown us that it is perfectly compatible for these three elements to occur simultaneously: a new technology with a permanent impact on the world economy (the railway or the internet); investors in that technology who lose money and investors who become millionaires by investing – by luck or skill – in the right companies. Investing in this environment would be equivalent to searching the Stock Market in 2000 for Google, Paypal or Amazon, avoiding Yahoo, AOL, Netscape, Webva or Kozmo, which few remember today, but were among the most promising companies of the time. The task is difficult, but not impossible.
However, a long-term investor would do well to consider other possible futures. Many analysts insist on classifying artificial intelligence as a bubble without considering that it might not be. There have been periods in which investment in a given sector has multiplied in a few years without a subsequent collapse. We call a bubble that does not burst because it generates sustained demand from supply “industrialization.” We have experienced it in Korea, Japan and Western Europe after the Second World War. In South Korea, investment went from 11% of GDP in 1960 to 27% (of a much larger GDP) in 1970, and remained at that level for decades, transforming a country forever.
If artificial intelligence really caused a widespread increase in productivity in the global economy, there would be no need to select anything: an indexation – investing in everything at once – would be enough to benefit from the wealth it would generate.
It is true that the evidence in this first phase may indicate that we are not close to that economic nirvana, but it is worth thinking carefully before ruling out that AI has the potential to profoundly transform the economy in the coming years.
