Science Vs AI: Is It Ruining the Environment?
80 snips
Nov 13, 2025 Join James O'Donnell, a Senior AI reporter at MIT Technology Review, as he dives into the environmental impact of AI. He reveals how much energy and water AI actually consumes, comparing it to everyday appliances. O'Donnell discusses the myth of AI's environmental devastation and the reality behind data center demands. He also examines the challenges of using renewable energy and the nuanced effects on local water supplies. Spoiler: using AI isn’t as dire as it sounds, but thoughtful use is encouraged!
AI Snips
Chapters
Transcript
Episode notes
Model Size Drives Energy Use
- AI models vary hugely in size and energy use because each model has a different number of parameters.
- Bigger models with more parameters generally consume more energy per query than smaller ones.
Per-Query Energy Is Surprisingly Small
- Measured small LLaMA queries used about 114 joules, roughly a tenth of a microwave second per prompt.
- Large models used up to ~8 microwave seconds per text query, still modest per interaction.
Images vs Text vs Video Energy
- Image generation can use similar or less energy than very large text models depending on the model sizes.
- Video generation consumes far more energy, equating to over an hour in the microwave for short clips.