Ever wonder how much electricity you’re using when you prompt, or thank, an AI model? Hugging Face engineer Julien Delavande did, so he built a tool to help arrive at the answer.
AI models consume energy each time they’re run. They’re run on GPUs and specialized chips that need a lot of power to carry out the associated computational workloads at scale. It’s not easy to pin down model power consumption, but it’s widely expected that growing usage of AI technologies will drive electricity needs to new heights in the next couple of years.
The demand for more power to fuel AI has led some companies to pursue environmentally unfriendly strategies. Tools like Delavande’s aim to bring attention to this, and perhaps give some AI users pause.
“Even small energy savings can scale up across millions of queries — model choice or output length can lead to major environmental impact,” Delavande and the tool’s other creators wrote in a statement.
⚡️ Ever wondered how much energy is used every time you send a message to ChatGPT?
We just built a version of Chat UI that shows how much energy your message consumes — in real time. Should all chatbots display this?
Details below 👇👇 pic.twitter.com/tBL0Pw51PW
— Delavande Julien (@juliendelavande) April 22, 2025
Delavande’s tool is designed to work with Chat UI, an open-source front-end for models like Meta’s Llama 3.3 70B and Google’s Gemma 3. The tool estimates the energy consumption of messages sent to and from a model in real time, reporting consumption in Watt-hours or Joules. It also compares model energy usage to that of common household appliances, like microwaves and LEDs.
According to the tool, asking Llama 3.3 70B to write a typical email uses approximately 0.1841 Watt-hours — equivalent to running a microwave for 0.12 seconds or using a toaster for 0.02 seconds.
It’s worth remembering that the tool’s estimates are only that — estimates. Delavande makes no claim that they’re incredibly precise. Still, they serve as a reminder that everything — chatbots included — has a cost.
“With projects like the AI energy score and broader research on AI’s energy footprint, we’re pushing for transparency in the open source community. One day, energy usage could be as visible as nutrition labels on food!” Delavande and his co-creators wrote.