trains lasted a century

Amazon has promised to invest 100,000 million dollars in AI data centers in 2025. Microsoft, 80,000. Google, 75,000. Goal, 65,000. The figures are absolutely dizzy, and are only part of the global investment that many other companies are carrying out in this area. The situation begins to be comparable to what occurred more than 140 years ago, when another investment fever conquered the United States.

In that case it was the railroad. But there are crucial differences between one and the other.

We are spending as if there were no tomorrow. Investor Paul Kedrosky did That interesting comparison Between spending on AI data centers and the one that was done on railroads more than a century ago. According to him, Capex’s figures (capital vessel, capital expenditure) for the United States in 2025 in the AI scope are absolutely fired and could represent according to their data between 1.2% and 2% of the Gross Domestic Product of the US. What is that? A lot. Lot.

More investment than with the Puntocom. The situation seems to overcome even the one that was lived with the investment in telecommunications companies both during the Puntocom bubble and during the era of the 5G networks, in which there was also a shotx fired. According to Kedrosky, the percentage of the US GDP on that occasion was 1%, but the capex for AI already round data centers (as little) 1.2%. A WSJ chart recently showed the situation:

Only the railroad attracted more investment. There has only been a situation that exceeds the current one. It occurred in the 1870-1880 decade, with the railway fever that made CAPEX shoot up to 6% of the US GDP at that time. The investment in AI data centers is still somewhat far from that figure, but it is still amazing, and above all, it does not seem to go less.

Money everywhere. As this analyst explains, the CAPEX set is not based only on the capex of the large technology companies that we mentioned at the beginning of this article. In addition to these gigantic investments there is a growing debt issuance to support these investments, private capital and new “Special investment vehicles“(SPVS) that are created precisely to support these massive capital flows.

Even Xi Jinping is scared. The Chinese president, Xi Jinping, has warned of the danger of excessive investment in data centers. In your country there are More than 250 data centers Under construction, and last week he warned of the risk of betting without brake on AI and electric vehicles:

“With regard to projects, there are some important aspects: artificial intelligence, computer power and new energy vehicles. Should all provinces in the country develop industries in these directions?”

Do not invest in other things. That extraordinary investment in AI data centers – or talent, tell them to the finish line– It is making any other segment much more difficult to capture money to continue developing. According to Kedrosky, the situation is analogous to which it was lived with fever for telecommunications and investments in other types of infrastructure, something whose effects continue to place.

Bubble danger. Faced with this unbalanced obsession with Big Tech to create more and more data centers there is a reality: AI, although useful for certain scenarios, still does not prove to be really revolutionary. Openai’s commitment, Google, Microsoft or goal is absolute, but it is almost more for the fear of getting too late to the market than by the fact that this market makes sense in itself. All this has made it talk for a long time AI bubblethat It could be comparable (or more worrying) than the bubble of the Puntocom.

Better spend now than regret it later. Mark Zuckerberg, Meta CEO, He already reflected About the situation with a clear message:

“It is very likely that many companies are oversizing (their investments in AI). But, on the other hand, I think that all the companies that are investing are making a rational decision, because the disadvantage of being left behind is that you could stay out with the most important technology of the next 10 to 15 years.”

They are still spending little. While many think that companies are spending too much money on AI, some analysts think they are spending little. Jim Cramer, from CNBC, affirmed That “these companies are not spending more in AI, they are missing.

Maybe they are not investing bad. The expert Noah Smith Analyze in your Newsletter Noahpinion The situation and remember important details:

• The telecos boom in the 90s and that overinversion led to the bubble of the Puntocom
• The railway boom led another gigantic crisis in 1873.
• In both cases, companies created too much infrastructure and bet more. The expectations were exaggerated, and the companies could not meet the loans they had asked.
• But be careful: in both cases something more important happened. Those who invested extraordinarily did not invest badly, but invested too soon. The railroads were disruptive, as were the telecommunications. There are those who defend that in fact Bubbles are good For innovation.

But it is not the Railroad (for the moment). Kedrosky concludes with a comparison that serves as a warning: what is being built with these AI data centers are not railroads:

• The train tracks have lasted more than a century and their cost-benefit ratio It has not been reduced just after the passing of the decades (although there has been investment to renew them)
• Data centers are short -term facilities, and the expensive GPUS on which they are based have a much shorter life cycle. Your cost-benefit relationship is much more debatable

Inflating the economy. This analyst has one more revealing data: without that investment in AI data centers, the US GDP contraction in the first 2025 quarter could be close to -2.1%. Or what is the same: AI is artificially holding the economy before demonstrating that it serves something.

Image | Martín Sánchez

In Xataka | Rumors about the huge goal data centers point to a problem: the Big Tech are playing everything to AI