Understanding AI Energy Consumption

This website was created independently to improve public understanding of AI power usage, and to reduce people’s concern about a very important technology for our future.

AI is a broad term referring to software running on computer chips called GPUs. Almost all computers contain GPUs, including the device you are reading this webpage on.

AI systems are very large, running on hundreds of thousands of powerful GPUs in commercial data centers, and supporting billions of users.

Viral news stories have created fear that AI is an environmental disaster. But the energy (and water) consumption for each individual user is very low, making up a negligible part of your daily consumption.

After reading these comparisons, it is our hope this has calmed concerns about the environmental impact of AI. Normal chatting with AI, especially for education or productive purposes, is very low-impact on a per use basis, and should not be avoided. The one outliter is AI video generation - which does use considerably more energy than any other popular AI uses. Detailed data on the energy consumption of AI video generation is limited, and we will update this site as more trustworthy figures become available.

This website is based on the latest data, which reflects significant improvements in efficiency. Every 12 months, advances in AI software, GPU technology, and datacenter design lead to significant increases in power efficiency.

This website was created in August 2025 for Andrej Karpathy's challenge to improve humanity's understanding of AI.

If you work for a AI lab or datacenter and would like to contribute data to this project, email us