Artificial intelligence is regarded as a leap forward in software development. Yet its rapid growth is shifting the debate away from the model itself and towards the infrastructure beneath it.
For a long time, the cloud seemed weightless because its technical foundations had largely disappeared from view. To users, the digital world appeared primarily as an interface, while the machinery behind it remained invisible. Files no longer sat on a personal computer, and programs became services accessed through a browser. With artificial intelligence, this convenient abstraction comes to an end. The larger the models become, the clearer it is that digital intelligence does not emerge from data alone. It needs electricity and an infrastructure whose operation has itself become part of the debate.
The term “cloud” was always an aesthetic simplification as well. It made technical infrastructure linguistically invisible and replaced it with an image of lightness. In reality, it rests on data centres whose operation consumes vast amounts of electricity and requires complex cooling systems. Artificial intelligence does not invent this foundation anew; it changes its significance. In the past, the focus was mainly on storage and transmission. Today, systems are emerging whose demand for computing power has grown massively within a very short time.
Data centres are becoming the factories of the present
A modern data centre is not a neutral technical space. It is a factory, even if no visible goods are produced there. Its product is computation; its waste is heat. It is precisely this sobriety that makes the process political. If the electricity demand of European data centres rises sharply by 2030, as forecasts suggest, AI will become a question of public infrastructure. References to more efficient algorithms or future renewable capacity will no longer be enough.
The major technology companies expose this contradiction themselves. On the one hand, they promise climate-neutral infrastructures; on the other, they report rising emissions. This is not merely a communications problem, but the expression of a structural tension. The digital industry sells efficiency while its growth creates new burdens. Advances in chips and models can be cancelled out by more intensive use. The individual query may become more economical, while the overall system continues to expand.
This mechanism becomes particularly clear with generative applications that are not only trained but used continuously. For a long time, the debate focused on the enormous effort required to train such systems. By now, however, energy demand is shifting into everyday use. Texts and images are generated within seconds, but not without material cost. The smooth user interface separates the visible action from the computing power performed elsewhere.
The hunger for data becomes a hunger for energy
AI’s appetite for energy does not begin at the electricity meter. It begins with the ambition to turn vast quantities of data into powerful models. Data alone is not intelligence. Only processing makes it usable, and that processing consumes energy. As long as the hunger for data is described merely as collection or training, it remains abstract. It becomes visible where networks come under strain and new capacity is required.
This sets AI apart from many earlier digital promises. For a long time, the platform economy could claim that it was replacing material processes with smart mediation. AI now exposes the limits of that narrative. Digital efficiency does not automatically mean ecological relief. It can improve processes while also generating new demand. Precisely because AI is being integrated into more and more work and production processes, its consumption is not growing only through individual large-scale projects.
The debate becomes imprecise when AI is described only as a climate risk or only as a climate solution. Both readings are too simple. AI can help manage energy flows and accelerate research. At the same time, it has resource requirements of its own, which cannot be justified in general terms by future savings. What matters is not the promise, but the balance sheet. And that cannot be inferred from corporate targets.
Progress once again depends on the power connection
The strength of artificial intelligence is therefore not revealed only in its models. It also depends on where data centres can be built and whether sufficient energy is available locally. Those with access to chips but no stable power supply do not possess a resilient AI infrastructure. Anyone wishing to build data centres needs more than capital. The digital economy is thus returning to an industrial logic: progress is not infinitely scalable when its prerequisites are limited.
This return is culturally revealing. Modernity often understood the digital as a release from material conditions. What once had to exist locally seemed to move into remote services. Artificial intelligence continues this narrative and refutes it at the same time. It can accelerate processes and make knowledge available. Yet the more powerful it becomes, the less it can be understood as software alone.
AI’s hunger for energy is therefore more than a technical problem. It shows that digital culture can no longer ignore its material foundations. Artificial intelligence will continue to grow because its usefulness is real and its economic momentum remains strong. In the process, however, it loses part of its aura. It no longer appears only as a promise of intelligent automation, but also as costly infrastructure. The future of AI will not depend solely on how powerful machines become, but also on the resources that this power demands.
