ByteThirst
Estimate water, energy, and carbon usage of every AI prompt
I built ByteThirst because I wanted to answer a simple question: what does it actually cost the planet when I send a prompt to an LLM?
As you use tools in your browser or IDE, ByteThirst estimates electricity, water, and CO₂ consumption. The estimates use published per-token energy figures, regional grid intensity, and typical datacenter PUE. All the assumptions and sources are visible on the methodology page.
Being honest about what this isn't: it's an estimator, not a measurement. Real costs depend on which datacenter, time of day, and a lot of things only the providers can see. But an order-of- magnitude feel beats zero visibility.
I'd love critique on the methodology — especially from anyone who's worked on LLM serving infrastructure or datacenter efficiency. Where am I likely off? Also launched on Product Hunt today: https://www.producthunt.com/products/bytethirst-know-your-queryweight
1 Comment
It’s crazy how we all use these models every day without ever really seeing the physical footprint they leave behind in terms of real-world resources. Most people don't realize that even a simple query triggers a chain reaction of energy and water cooling that's almost entirely invisible to the end user.
Since you're using regional grid intensity for the estimates how do you factor in the specific provider's carbon offset claims versus the actual live energy mix of the datacenter's location?