You've probably seen alarming headlines: “AI uses as much water as a swimming pool!” or “ChatGPT consumes massive amounts of energy!” These stories spread quickly, but they often lack crucial context.
Researchers have found that some widely-cited figures are approximately 1000x too large. Many estimates conflate the enormous one-time cost of training AI models with the much smaller cost of inference (actually using them). Others use outdated figures from less efficient hardware generations, or extrapolate worst-case scenarios.
As Undark reports, experts agree that “people often have a muddled understanding of how data centers use water, and that their overall consumption, in many places, is less of a risk than the public may think.”
This isn't to say AI has zero environmental impact - it does. But understanding the actual scale helps us make informed decisions rather than fear-based ones.
We take a thoughtful approach to AI usage:
- Efficient models: We use optimized inference through OpenRouter, which routes to efficient model providers and hardware.
- Inference only: We never train AI models on your stories. We only run inference (asking pre-trained models questions). Training is where the vast majority of AI energy costs occur; inference is orders of magnitude lighter.