Infrastructure Bubble

A bubble, in stock market terms, is a state in which prices rise far above the actual value of something. One speaks of an infrastructure bubble when this pattern does not affect stocks but large physical assets: railway lines, fiber-optic cables, power plants, or data centers. Many companies then build simultaneously because everyone expects enormous future demand. The money for this often comes from loans or from investors expecting a quick return. If the expected demand arrives later or not at all, the facilities stand half-empty. The debts remain nonetheless, and many of the builders run into trouble.

Why empty data centers are more expensive than empty warehouses

Infrastructure differs from other investments in its size and its inertia. A data center quickly costs several billion euros and takes years to complete. Whoever builds today is betting on the demand of the day after tomorrow. This bet can no longer be reversed once the concrete has been poured.

Added to this is a contagion effect. The investments are usually financed through bonds, that is, through money borrowed from banks and funds. If the bubble bursts, not only the builders lose out, but also their lenders. In the past, this has dragged entire industries down with it.

An important difference from pure speculative bubbles: the built facilities do not disappear. After the fiber-optic boom around the year 2000, unused cables lay in the ground for years. They were later sold cheaply and contributed to the rise of streaming and cloud services. The investors of that time, however, saw none of that benefit.

How expectation becomes detached from reality

At the beginning there is usually a genuine development. The railway in the 19th century and the internet around 2000 were not figments of the imagination. The problem is the speed of expectation. Investors simply extrapolate short-term growth rates into the future.

Then a race begins. Whoever builds first secures the best locations, power connections, and customers. So every provider would rather build too much than too little. Because everyone thinks this way at the same time, massive overcapacity results in total, even though each individual decision seems reasonable on its own.

The problem only becomes visible late. As long as new money keeps flowing in, even unprofitable projects appear healthy. Once sentiment turns, lenders demand higher interest rates or pull out entirely. Often a single disappointing quarterly report is enough to set the chain in motion. A typical misconception, by the way, is the assumption that a bubble can be recognized by nonsensical ideas. Usually the idea is right — it’s just the price and the pace that are not.

The debate over AI data centers

The term has appeared in business news almost daily since around 2024. Large technology companies are investing hundreds of billions of dollars in data centers for artificial intelligence. These house specialized chips on which AI models are trained and run. Critics ask whether enough customers are willing to pay for this computing power over the long term.

Circular deals are particularly striking here: a chip manufacturer invests in an AI company, which in turn uses the money to buy that manufacturer’s chips. Such arrangements make revenue appear larger than actual outside demand really is. Many observers consider them a warning sign.

Whether this is really a bubble remains open and can only be reliably assessed in hindsight. Anyone reading reports about new data centers, however, can pay attention to three points: Where does the money come from, are there firm purchase agreements, and is the available electricity even sufficient for the planned facilities?