In 2026, the electricity demand is skyrocketing (expected to exceed Japan’s use in 2 years). Welcome to modernity. Amidst the unusual prospect of believing that generative artificial intelligence will lead us definitively to superhuman capabilities, the multiplication of knowledge, and the potential development of custom-tailored drugs for each of us, we had forgotten that any research on ChatGPT leads to a surge in electricity demand that could cause server strain. The Electricity 2024 report states, “Data centers are significant drivers of electricity demand growth in many regions.
After consuming approximately 460 terawatt-hours globally in 2022, total electricity consumption by data centers could exceed 1,000 TWh in 2026 which is roughly equivalent to Japan’s electricity consumption. Updated regulations and technological improvements, even in terms of efficiency, will be crucial for moderating the increase in energy consumption by data centers.” There is a ninefold difference: with 9 billion daily searches, the additional demand would be almost 10 TWh per year. The Agency also takes into account the paper published by data scientist Alex de Vries, a genuine guru in the field, in Joule at the end of 2023.
A paper that has sparked considerable reflection even among experts. It starts with the assumption that 95% of the AI server market is covered by Nvidia. He calculated that if the company delivered 100,000 servers in 2023, capable of consuming between 5.7 and 8.9 TWh, and by 2027, it will be shipping 1.5 million per year, they would require between 85.4 and 134 TWh at full capacity. Artificial intelligence activities are based on highly sophisticated models that have a radically different development and use paradigm compared to traditional digital tools.
These AI models are created in two phases: an “training” phase for the algorithm offline and a predictive online usage phase, both carried out with the work of powerful data centers, which are among the most energy-consuming infrastructures in existence. The U.S. Department of Energy estimates their energy consumption per square meter to be 10 to 50 times higher than that of any commercial building, concentrating about 2% of the country’s total electricity consumption.