AI Tool Reveals Energy Consumption of a Single Chatbot Prompt

AI systems are notorious for their high energy consumption, yet pinpointing the exact energy use, especially for individual chatbot queries, remains elusive. To shed light on this issue, a Hugging Face engineer developed a tool aimed at uncovering these hidden costs.

It's easy to get lost in the ethereal language of AI infrastructure, with its frequent references to "the cloud" and other airy metaphors. However, beneath this veil lies a reality of energy-intensive computers. AI systems, which need to perform complex calculations at breakneck speeds, rely on powerful chips, multiple GPUs, and vast data centers. Every time you pose a question to ChatGPT, these components spring into action, consuming electricity. This is why many chatbots impose usage limits on their free tiers—the cost of electricity makes computing a pricey affair for the companies hosting these services.

Chat UI Energy

To bring some clarity to this murky area, Julien Delavande, an engineer at Hugging Face, created an AI chat interface that displays real-time energy use estimates for your conversations. This tool not only compares the energy consumption of different models, tasks, and requests but also highlights how a prompt requiring reasoning might consume more energy than a straightforward fact-finding query. The interface presents energy usage in relatable metrics like Watt-hours and Joules, as well as more tangible comparisons such as the percentage of a phone charge or driving time, sourced from the Environmental Protection Agency (EPA).

When I asked Chat UI Energy about the weather in New York City, it informed me that my query used about 9.5% of a phone charge. By clicking on this estimate, I could see other equivalent comparisons, such as 45 minutes of LED bulb use, 1.21 seconds of microwave use, and 0.15 seconds of toaster energy. As the conversation continued, the bot displayed the total energy usage and duration at the bottom of the chat window.

Despite the simplicity of my query, it required internet access, which the bot lacked, possibly explaining why it took 90 seconds and more energy than anticipated to respond. Even as an estimate, 45 minutes of LED bulb use seemed surprisingly high, putting into perspective the energy demands of more complex, multi-step prompts.

While only AI companies have precise data on their systems' energy consumption, studies suggest that demand is on the rise. A 2024 International Energy Agency report forecasts a global electricity demand increase of 3.4% by 2026, partly driven by the expansion of data centers. Similarly, a Berkeley Lab report anticipates data center growth rates of "13% to 27% between 2023 and 2028."

The release of this tool underscores the difference between open-source platforms like Hugging Face and more secretive AI companies. "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," the chat's creators stated in their announcement. "One day, energy usage could be as visible as nutrition labels on food!"

How to Try It Yourself

You can experiment with the chatbot here and switch between various open-source models, including Google Gemma 3, Meta's Llama 3.3, and Mistral Nemo Instruct.

Interested in more AI stories? Subscribe to Innovation, our weekly newsletter.