Generative artificial intelligence consumes enormous amounts of electricity. Each request made to an advanced language model requires much more energy than a traditional web search. As these tools spread across every production sector, the energy hunger of data centers is growing at a pace that traditional power grids struggle to sustain.
This digital gold rush is reshaping not only the IT industry but the entire global energy planning, bringing nuclear energy back to the forefront as the only source capable of providing stability and low emissions continuously.
Dossier Summary
- Energy Impact: A single AI request consumes up to 10 times more electricity than a traditional Google search.
- The Nuclear Race: Microsoft, Google, and Amazon are making direct agreements with power plants and small modular reactors (SMR).
- Water Crisis: Data centers require millions of liters of fresh water daily to cool supercomputers.
- Hidden Costs: The enhancement of public power grids risks burdening citizens' bills in the form of system charges.
The Insatiable Appetite of Digital Infrastructures
The servers that train and run AI models work tirelessly. They need two things to function: a constant and enormous power supply and water cooling systems to prevent processor overheating. Intermittent renewable sources, such as wind and solar, alone are not enough to ensure the constant operation of machines on windless or sunless days.
Energy Consumption Comparison
Web Search vs Generative Query
A traditional search consumes an average of 0.3 Wh of energy. A single query to a complex generative LLM (with active reasoning steps) consumes between 3 and 9 Wh (up to 30 times more), motivating the accelerated transition to sources with very high constant energy density (nuclear).
As a result, tech giants are forming direct partnerships with atomic energy producers. Recent agreements in the United States see web giants purchasing shares or signing long-term supply contracts with existing nuclear plants, or investing in the development of small modular reactors (SMR) to be placed next to new data centers.
The Water and Cooling Problem
The power supply is only one side of the coin. Cooling the circuits requires millions of liters of fresh water daily. Many large facilities rely on evaporation systems that draw water reserves from surrounding areas.
The transition to closed-loop and liquid cooling technologies represents a necessary step forward to reduce the local ecological impact, but it also requires significant investments and further energy consumption to keep the fluids circulating.
Who Pays the Final Bill?
The expansion of data centers requires the enhancement of pylons, transformer stations, and national distribution networks. Although Big Tech finances the construction of their own facilities, the costs for upgrading public electrical infrastructures risk falling, in the form of network charges and regulated tariffs, on the bills of all citizens and small businesses.
The challenge for the coming years will be to find a regulatory and fiscal balance that allows technological innovation to develop without burdening the community with the costs of this epochal energy transition.