Generative Artificial Intelligence

There are several ways generative AI impacts the environment:

• Training AI models
  • Training AI models generates carbon dioxide - a lot (Sim, 2024). It's estimated that training Chat GPT-3 used up the same amount of electricity as that used to power 120 US homes for a year (Zewe, 2025).
  • This is also equivalent to 500 tons of carbon dioxide (Coleman, 2023) - or what it takes to drive a gasoline car over 800,000 kilometers.
• Data centre operation
  • These are essential for AI model training and consume large amounts of energy (Sim, 2024).
  • Currently data centres consume ~ 1.5% of global electricity (Zewe, 2025) but this is forecast to rise to 6-12% of US energy consumption by 2028 (Kneese & Young, 2024).
  • Energy demand is outpacing renewable energy infrastructure so is relying on fossil fuels (which means more carbon) (Kneese & Young, 2024).
  • While current energy use is great, some estimate that AI will drive down emissions in the future as processing power improves and as machine learning is used to explore how to reduce emissions across industries such as airline travel and food production (Azhar & Frey, 2024).
• Hardware such as Graphical Processing Units (GPUs)
  • GPUs are essential to the operation of AI and require water to keep cool. One estimate is that the demand for water for this purpose could be equal to half of the current demand of the entire country of the United Kingdom by 2027 (Crawford, 2024).
  • Already, "water used for cooling is contributing to stressed watersheds" (Foy, 2023).
• Using generative AI for research
  • Some estimates put a single search in genAI at fivefold the carbon footprint of a single Google search (Kneese & Young, 2024).
  • Note that there is uncertainty around estimates as AI companies are not revealing definitive numbers. Some research estimates that the carbon cost of using a typical LLM-powered chatbot is "fairly low by the standards of other ordinary uses of electricity" (You, 2025).