Lin Hsin Hsin Intelligence Center





AI Model Energy Consumption
by Brand & Query Type








Brand Name AI Model Query Type Energy Consumption by Year Measured Closest Comparison
Google Gemini Text 0.24 watt-hours per median prompt (May 2025) Running a standard microwave for about one second
OpenAI ChatGPT Text 0.34 watt-hours per average query (June 2025) Slightly more than Google Gemini
Stability AI Stable Diffusion 3 Image As much energy as fully charging your smartphone (2023/2025) Generating an image is much more energy-intensive than text
Open-Source Models Meta Llama (text), Stable Diffusion 3 (image), Video Models Text, Image, Video Text: 1,000 queries = 16% of a smartphone charge; Image: 1 query = full smartphone charge; Video: significantly more than image (2025) Video queries require the most energy, followed by image, then text
OpenAI Stargate Initiative (planned) All Each data center: up to 5 gigawatts (2025 announcement) More than the total power demand of the state of New Hampshire
Global Data Centers AI and Data Center Usage All Less than 300 TWh in 2020; projected nearly 1,000 TWh in next five years (2025) More than Japan's total electricity consumption








versus






Lin Hsin Hsin's BeyondAI™
Energy Consumption


3D Equation-based Creations
2D Equation-based Creations










Sources: MIT Technology Review, May 2025 articles on AI energy usage, Google's Gemini energy data, and open-source model studies