Maybe there is no AI bubble

Rationality. We recently published a series of articles about data centers in Ireland, Norway and elsewhere. I’m always struck by the sheer volume of capital flowing into infrastructure for a product, generative AI, that has yet to prove its commercial value. Some analysts even calculate that OpenAI could face bankruptcy as early as next year. And yet forests are razed, aquifers depleted and energy squandered on thousands of computer farms.

I asked historians of technology for precedents. “History does not repeat, but it does instruct,” as the US historian Timothy Snyder once wrote. Martin Schmitt, a research associate at Paderborn University, pointed out that the data center bubble is not really unique. The dotcom bubble, he said, was comparable. Back then, data centers already consumed roughly 1% of the world’s electricity, and investors ploughed $100bn into the “new economy” in the United States in 2000 alone. In 2025, AI companies in the US received a comparable sum (though others reach different conclusions using alternative metrics). From this, one might infer that the AI boom will end as the dotcom bubble did: with massive losses.

Path dependence. Frank Uekötter, a professor at the Ruhr University Bochum, told me that when comparing industrial developments, the essential question is not merely to identify surface similarities but to analyze path dependence: Past events or decisions can limit or influence future choices and outcomes. Now that thousands of data centers have been build, what choices do they leave us with for the future? Could they become the foundation of a whole new type of infrastructure? Or will they just become vestige for urban explorers like the hundreds of decaying Olympic stadiums across the globe?

If AI is merely a bubble, the future remains wide open. After all, the ruins of the dotcom era provided fertile ground for all kind of innovations, some emancipatory, such as Wikipedia or Sci-Hub, others more troubling or even oppressive, including surveillance tech. Yet I suspect the current AI investment craze is different.

Political alignment. As a consumer product, AI is unlikely to recoup its vast capital costs, even if a minority of workers find it useful. As an instrument of government, however, it is far better suited. The current strain of authoritarianism, in the United States, Argentina, China, Russia and elsewhere, blends anti-intellectualism with an appetite for extensive surveillance. Generative AI offers both. It negates the value of the work of scientists and artists, as well as critical thinking in general, by asserting that large language models can replace all of it, and it can sift through virtually any form of communications, written and spoken, forcing entire populations towards continuous self-policing, even within the home, thanks to ever-listening smart devices.

On this reading, data centers would be here to stay. Even if investors eventually conclude that consumer appetite for generative AI is thin, governments may well step in to bail the industry out. Not to save jobs, but to safeguard their own powers.

A precedent. Generative AI would not be the first industry to entrench itself despite lukewarm public sentiment. Until the 1930s, automobiles in central Europe were regarded as dangerous and as symbols of the contempt with which affluent urbanites oppressed rural poor communities and factory workers (there is scant research on anti-motorism, but Uwe Fraunholz’s Motorphobia is a useful starting point). However, political and industrial elites steadily reshaped everyone’s environment, constructing highways, introducing fiscal incentives and dismantling alternatives – until living car-free became not merely inconvenient but almost impossible.

A similar dynamic is now unfolding with generative AI. Large tech companies are embedding it deep within their products. Once cell phones, which are de facto prerequisites for civic life, incorporate generative AI at the operating- system level, opting-out will become impossible. Meanwhile, governments are promoting its adoption across public administration, education and beyond. Even if the AI bubble were to burst, data centers will continue to hum, less as speculative ventures than as fixtures of the digital surveillance state.

This is an excerpt from the Automated Society newsletter, a bi-weekly round up of news in automated decision-making in Europe. Subscribe here.