Developer Introduces Tool to Assess AI Energy Consumption

Hugging Face developer Julien Delavande has introduced the AI Energy Score tool, designed to estimate the electricity consumed by artificial intelligence queries.

The service displays energy consumption by graphics processors in kilowatts or joules for each message sent. For convenience, it provides equivalents in other measurements, such as the percentage of a phone battery charge, seconds of microwave operation, or the duration of LED lamp usage.

? Short demo in the image/video below pic.twitter.com/Lzrioz6Alx

— Delavande Julien (@juliendelavande) April 22, 2025

“Even small energy savings can scale to millions of requests — the choice of model or output length can significantly impact the environment,” states the tool’s page.

The service is compatible with Qwen 2.5 7B, Llama, Gemma, and Mistral.

The tool evaluates the energy consumption of messages in real-time. For instance, asking Llama 3.3 70B to write a standard letter would consume approximately 0.1841 Wh — equivalent to running a microwave for 0.12 seconds or a toaster for 0.02 seconds.

Delavande emphasizes that while the results are not perfectly accurate, they serve as a reminder that everything, including chatbots, comes at a cost.

“Thanks to projects like AI Energy Score and broader research into AI’s energy footprint, we are achieving transparency within the open-source developer community. One day, energy consumption will be as clear as food labels,” the developer noted.

In April, OpenAI CEO Sam Altman revealed that the company spent tens of millions of dollars responding to users who wrote “please” and “thank you.”

In August 2024, a study by the Bitcoin Policy Institute indicated that AI’s energy consumption likely already exceeds that of Bitcoin mining.

In December, Google and Microsoft planned to “green” the energy procurement process and cooling systems of data centers.