With a single command, we turn a few words into a photorealistic image. We have complex conversations with a chatbot that writes code for us in seconds or analyses voluminous documents. This digital ‘magic’, which only a few years ago seemed the domain of science fiction, is now becoming part of our everyday lives. However, it has a very real, physical price that cannot be seen on the screen – gigantic electricity consumption.
The technology industry has been building its image as a driver of sustainability for years, investing billions in green initiatives. Today, it faces a fundamental paradox. Artificial intelligence’s insatiable appetite for computing power is putting these aspirations to the test, clashing the image with the inexorable laws of physics. This raises a key question: in the search for a stable and emission-free power source for its ‘AI factories’, will the industry be forced to turn to one of the most controversial technologies in history – nuclear power?
AI’s insatiable appetite: why is the cloud so hungry?
To understand the scale of the problem, it is necessary to look under the bonnet of artificial intelligence. Its huge energy requirements are driven by two main processes. The first is model training. Learning a large neural network, such as those driving ChatGPT, is a process that requires thousands of GPUs running continuously for weeks or months to process unimaginable amounts of data. The energy consumption of one such cycle can be compared to the annual demand of a small city.
The second, often underestimated power devourer is inference, the daily use of an already trained model. Every query we make to a chatbot, every request to generate an image, sets off a complex chain of calculations in the data centre. These individual small power consumptions, multiplied by millions of users worldwide, add up to astronomical values. As the latest market research confirms, AI is the main driver of the explosive growth in computing power demand, pushing existing infrastructure to its limits.
Green energy under pressure – dreams versus reality
It is fair to say that the data centre industry is not ignoring the problem. For years, technology giants have been outdoing themselves in investing in renewable energy sources (RES), building huge wind and solar farms to power their server rooms. In line with industry sentiment, as many as 91% of professionals believe that the future of energy for data centres lies in wind, solar and water.
However, this is where the fundamental challenge arises – stability. Data centres are critical infrastructure that must operate with a guaranteed availability of 99.999%. What happens when the sun goes down and the wind stops blowing? While for a household a temporary drop in power from photovoltaic panels is not a catastrophe, for a server room serving global financial or medical systems it is an unacceptable scenario. Renewable energy sources are great at covering part of the demand, but they have the problem of providing the so-called base load – a constant, reliable supply of energy 24 hours a day.
Back to the atom? A new opening in the form of the SMR
It is in this gap that a concept is emerging that until recently lay on the margins of the technological debate: small modular reactors (SMRs). This is a new generation of nuclear power that has little in common with the giant plants of the Cold War era. SMRs, as the name suggests, are much smaller and their key feature is modularity. They are to be mass-produced in factories, potentially lowering costs and dramatically reducing construction time. Most importantly, they use modern passive safety systems to minimise the risk of failure.
For the data centre industry, such a solution seems almost ideal. Imagine a server campus powered by its own small reactor, delivering stable, emission-free energy directly to the site, without dependence on an external grid. It’s a vision that is no longer a fantasy – as many as 75% of industry experts are now open to considering nuclear power as part of the energy mix.
Cold shower – barriers on the road to the atom
However, before declaring the renaissance of the atom, it is necessary to come down to earth. The road to implementing SMRs is long and bumpy, as the experts themselves acknowledge.
Firstly, time. This is not a technology that can be implemented next year. 70% of experts believe that SMRs will not be ready for commercial use until a decade from now at the earliest.
Secondly, the public. Nuclear power still arouses strong emotions and fears. As many as 60% of professionals expect significant public opposition, which could effectively block or delay projects.
Thirdly, regulation and costs. Any nuclear project, even on a ‘small’ scale, is subject to extremely complex and lengthy licensing processes, and the upfront costs remain enormous.
Energy crossroads
The technology industry is at an energy crossroads. On one side stands the unbridled appetite of artificial intelligence, on the other the ideal of a green, sustainable future. In between is the pragmatic, if distant and controversial, promise of the atom.
There is no simple answer to this dilemma today. One thing is certain: the discussion about AI-driven nuclear reactors has returned from academic corridors to the boardrooms of the world’s largest technology companies. The future and pace of development of AI may depend on whether we can find a suitably powerful and clean power source for it. The atom has not yet been ‘resuscitated’, but its pulse, after decades of dormancy, is becoming palpable again.
