Over-accumulating AI. Rationale and fault lines beyond the US AI investment boom

In the mid-2020s, the world watched as a new wave of technology swept through the economy, promising to reshape how we work, create, and think. This technology, known as artificial intelligence, relies on massive amounts of computing power stored in giant data centers. To build these centers, a handful of giant technology companies began spending money at a pace never seen before in history. They poured billions into buying specialized computer chips and constructing facilities to run them. This surge in spending drove up the stock prices of these companies and became a major engine for economic growth in the United States. However, such rapid and massive investment often raises a question among economists: is this growth built on solid ground, or is it a bubble waiting to burst? The concern is not just about whether the technology works, but whether the money spent on it can ever be earned back through the products and services it creates. If the returns do not match the costs, the entire system could face a sudden and painful correction.

A team of researchers from the University of Geneva, Gaia Valenti and Cédric Durand, set out to examine this situation through the lens of economic history. They looked at the current boom not just as a technological success story, but as a potential crisis of "overaccumulation." In simple terms, this concept describes a situation where companies are forced by competition to invest so much money that there is more capital available than there are profitable opportunities to use it. The researchers argued that while the excitement around artificial intelligence is driven by a single, optimistic path where productivity soars and demand grows forever, the path to a crash is far more complicated. They identified three distinct ways the boom could fail, each interacting with the others to create a fragile situation.

The first path to trouble lies in the financial system itself. The researchers found that the massive spending by tech giants is not being paid for by the profits these companies are currently making. Instead, they are relying heavily on debt and complex financial deals that link them together. For example, some companies are borrowing huge sums to build data centers, while others are signing long-term contracts to buy computing power from those same centers. This creates a cycle where the value of one company depends on the success of another. If the demand for artificial intelligence does not grow as fast as predicted, these companies could find themselves unable to pay their bills. Because their finances are so deeply intertwined, a problem for one could quickly spread to all, potentially causing a broader financial shock.

The second path involves the physical and political hurdles of building the necessary infrastructure. The researchers pointed out that while companies have secured vast amounts of money, actually building the data centers is becoming increasingly difficult. These facilities require enormous amounts of electricity and water, and they need to be built on land that is often subject to local opposition. In the United States, many states have begun to restrict or delay the construction of data centers due to concerns about energy use and environmental impact. Furthermore, the supply of the specialized chips needed to run these systems is vulnerable to global conflicts and shipping disruptions. This means that even if companies have the money to spend, they may not be able to turn that money into working technology fast enough to meet the expectations of investors.

The third and perhaps most fundamental path to crisis concerns the actual work that artificial intelligence is supposed to do. The entire boom is justified by the promise that these machines will make human workers much more productive, allowing companies to earn back their huge investments. However, the researchers found evidence that this promise might not hold up. While artificial intelligence can speed up specific tasks, it often disrupts the way teams work together. It can create a situation where junior workers lose the chance to learn essential skills because they are not doing the work themselves, and senior workers become overwhelmed trying to check the machine's output. This can lead to a decline in the overall quality of work and a loss of motivation. If the technology fails to deliver the expected improvements in how people work, the demand for it will stall, and the massive investments made to build it will not pay off.

The researchers concluded that the outcome of this boom is uncertain, but they outlined two possible futures. One possibility is an "AI winter," where the bubble bursts, investment collapses, and the technology retreats to a smaller, less influential role in the economy. The other, perhaps more likely scenario, is what they call "AI gloom." In this future, the technology does not disappear, but it also does not deliver the massive economic gains that were promised. The companies that built the infrastructure remain powerful, but the economy as a whole does not see the prosperity that was expected. Instead, the costs of maintaining this expensive system are passed on to society, and the power of a few large technology firms becomes even more entrenched. The study suggests that while the technology is real, the economic logic driving its current expansion is fraught with danger, and the path forward is far less certain than the hype suggests.