Artificial intelligence appears weightless. A question is typed on a screen and a response appears seconds later. The interface hides nearly everything that makes it possible: chips, servers, cooling systems, power lines, transformers, water, buildings and land. The cloud is not a cloud. It is an increasingly large and concentrated industrial infrastructure.
According to the International Energy Agency, global data-center electricity consumption could more than double by 2030 to roughly 945 terawatt-hours. AI is the largest driver of this rise alongside other digital services. The global share would remain below three percent, but that average hides local effects: one large AI center can concentrate demand comparable to a city or an energy-intensive factory.
Concentration changes energy politics. A national grid may have sufficient generation while a region lacks the lines required to connect a new facility. Building power generation takes time; permitting transmission and substations can take even longer. Data centers are planned quickly and request very large connections. The race for AI therefore encounters the physical timescale of the grid.
Cooling is also part of the problem. Servers turn electricity into heat and must remain within precise operating conditions. Systems use air, liquid loops and in some cases evaporated water. Consumption varies by climate and technology, but communities notice the competition immediately when a new facility arrives in a drought-prone area or where infrastructure is limited.
The Australian debate on August 25 showed how digital infrastructure is becoming a matter of government. Concerns include energy, sustainability, grid costs and the relationship with communities. The data center is no longer discussed only as a technology investment, but as a territorial project that must explain which resources it uses and which benefits it returns.
Efficiency matters: smaller models, specialized chips, optimized software, heat recovery and intelligent load distribution can reduce the energy required for the same performance. Yet efficiency can be absorbed by growth. If each operation costs less, millions of new operations become affordable. Lower consumption per request does not automatically mean lower total consumption.
ENERGY FOR AI brings the technology back to earth. Every model inhabits a place, uses a grid and produces heat. Once this materiality becomes visible, the question changes: not only how intelligent the system is, but how much energy we are willing to dedicate to it, who will build the infrastructure and who will pay the cost. AI’s future will also be decided far from the screen, in power plants, grids and the communities that host them.