AI energy consumption isn’t something most people think about when chatting with a bot — but maybe it should be. Hugging Face engineer Julien Delavande built a tool that shows just how much electricity is used every time you send a message to an AI model. And the results might surprise you.
Large language models, like Meta’s Llama 3.3 or Google’s Gemma 3, require serious computing power. They run on GPUs and advanced chips that consume plenty of electricity. Every query — even simple ones — contributes to rising AI energy consumption. As usage skyrockets, experts warn the energy demands of AI could reach new highs in the next few years.
Delavande’s tool brings this invisible cost into focus. It works with Chat UI, an open-source front-end for large models, and tracks AI energy consumption in real time. Each message you send or receive is measured in Watt-hours or Joules. To make it more relatable, the tool also compares that energy use to everyday appliances like microwaves and toasters.
Let’s say you ask Llama 3.3 to write an email. The tool estimates that this task uses about 0.1841 Watt-hours. That’s like running a microwave for 0.12 seconds or using a toaster for just 0.02 seconds. Tiny numbers — but at scale, those small bits of energy add up fast.
While the estimates aren’t precise, that’s not the point. This tool helps raise awareness around AI energy consumption. It shows users how much electricity is spent on what might feel like simple digital interactions. The message? Every query has a cost.
Delavande and his collaborators believe transparency is key. Even small energy savings — like picking a smaller model or cutting down on output length — can make a big difference. When you multiply those small choices across millions of users, the impact becomes clear.
“We’re pushing for energy awareness in the open-source AI community,” Delavande shared. “AI energy consumption should be as visible as nutrition labels on food.”
That kind of visibility could shape how future AI tools are built. If developers and users become more mindful of power usage, it could lead to more efficient, sustainable AI. And as AI keeps expanding, tools like this one could make a real difference.
