Have you ever considered how much energy it takes every time you interact with an AI chatbot? Hugging Face engineer Julien Delavande did—and decided to find out. Delavande built a new tool specifically designed to estimate the electricity used by chatbot messages in real time.
Artificial intelligence models are notoriously power-hungry. They often run on powerful GPUs and specialized processors, both of which consume substantial amounts of electricity. Pinpointing the exact energy footprint of these models can be challenging, but as AI adoption continues to soar, many analysts predict the power demands will skyrocket in the coming years.
This growing appetite for electricity has already started prompting some companies to resort to environmentally problematic solutions. Delavande’s goal is straightforward: by highlighting AI’s hidden energy costs, he hopes to encourage more conscientious decision-making around AI usage and design.
In a statement accompanying the tool’s launch, the creators emphasized how seemingly minor decisions—like selecting different AI models or adjusting the length of outputs—can accumulate significant environmental impacts over millions of queries.
The new energy-tracking tool integrates directly into Chat UI, an open-source front-end commonly used with large language models such as Meta’s Llama 3.3 70B and Google’s Gemma 3. While users interact with the chatbot, energy consumption is measured and displayed live, reported in Watt-hours or Joules. The interface also contextualizes this usage by comparing it with familiar household devices like microwaves or LED lights.
For example, according to this tool, even a simple task such as asking Llama 3.3 70B to compose a standard email consumes around 0.1841 Watt-hours—equivalent to operating a microwave for roughly 0.12 seconds or using a toaster for about 0.02 seconds.
Delavande acknowledges these figures are rough approximations rather than precise measurements. However, the message remains clear: even seemingly innocuous digital interactions carry tangible environmental costs.
In their announcement, Delavande and his collaborators emphasized their broader advocacy for transparency around AI resources. They envision a future where clear disclosures of energy consumption become standard practice, similar to nutritional labels now universally present on food packaging.
