🔗https://www.iea.org/reports/energy-and-ai
The International Energy Agency (IEA), which provides guidance on energy policy to 44 countries, has released a groundbreaking report analyzing the energy footprint of Artificial Intelligence (AI). This includes not just the energy used in data centers and chip production, but also in the extraction of raw materials essential to the AI ecosystem. The report presents both troubling forecasts and promising opportunities, highlighting the dual nature of AI’s relationship with energy: as a voracious consumer and as a powerful tool for optimization.
Dark Clouds
The IEA outlines four scenarios for AI’s energy trajectory: a base case where trends continue, a take-off with accelerated adoption, a high efficiency path with reduced consumption, and a headwinds scenario with slowdowns. Even in the base case, electricity demand from data centers is expected to more than double by 2030—jumping from 415 TWh to 945 TWh, approximately 2.5% of today’s global energy usage. The numbers could soar to 1700 TWh by 2035.
AI accelerator chips are particularly power-hungry, and their energy use is projected to quadruple by 2030. Most of this growth will come from the U.S. and China, which, along with Europe, dominate global data center distribution. These regions may face significant pressure on local power grids.
Silver Linings
Despite its heavy energy demands, AI holds immense potential for energy efficiency. Current AI systems already help forecast energy needs and manage the flow from renewable sources like solar and wind, reducing the use of fossil fuels. If AI-driven efficiency programs expand by just 1%, global COâ‚‚ emissions could drop by 120 megatons by 2035.
Even more promising, broader application of AI in sectors such as industry, buildings, and transportation could slash emissions by 1.4 gigatons—nearly five times the emissions produced by data centers. Enhancing HVAC systems with AI could save 300 TWh of energy. Meanwhile, major cloud providers are locking in long-term renewable energy contracts, primarily solar, encouraging growth of green power by 450 TWh.
On top of that, advances in AI model design, energy-efficient hardware, and smarter usage methods are helping lower the energy costs of AI operations like training and inference.
Conclusion
This report marks the first comprehensive analysis of AI’s growing energy demands and its capacity to solve the very problems it exacerbates. AI will undoubtedly drive a surge in energy use, but if managed wisely, it can also become a key player in achieving global energy efficiency and climate goals. The technology that challenges our energy systems today may very well be the catalyst for their transformation tomorrow.