0800 464 7677
Ian Wright AI Electricity & Water Use Cost Calculator
The calculator below provides an extremely rough estimate of how much water and electricity various AI applications use. It also estimates the cost of providing that electricity to power each use.
You can also get an idea of how much energy AI uses by looking at our earlier piece: ChatGPT Energy Consumption Visualized.
Business Energy UK · Resource calculator
How much does an AI prompt actually cost?
Estimate the electricity, water and money behind text, image and video prompts from the leading AI tools, using figures published by Google, OpenAI and independent researchers. Energy adds up fast when a whole business is prompting all day.
Default 0.24 (about the UK business average of 24p). Change to your own tariff.
Paying too much to power your business?
AI is just one line on the bill. Compare business energy prices in under 60 seconds and see how much your business could save.
Where these numbers come from ▾
Reporting on AI energy and water use is inconsistent: figures vary enormously by model, prompt length, hardware and what each study chooses to count. These are mid-range estimates drawn from the most credible public sources, so treat them as order-of-magnitude guides, not precise meter readings.
Text prompts
Google reported a median Gemini text prompt at 0.24 Wh and 0.26 mL of water. OpenAI's Sam Altman cited roughly 0.34 Wh and 0.32 mL per average ChatGPT query. Longer answers and reasoning models run far higher, with some reports putting GPT-5 style responses at 18 Wh and up to 40 Wh.
Image generation
A Hugging Face / Carnegie Mellon study measured an average of about 2.9 Wh per image, up to 11.5 Wh for the least efficient model. MIT Technology Review measured Stable Diffusion 3 Medium at around 0.32 Wh for a 1024px image.
Video generation
Far higher and highly uncertain. Researchers measured a five-second CogVideoX clip at about 3.4 megajoules (around 944 Wh), and estimates for Sora-class models land near 1 kWh for five seconds. Energy roughly quadruples each time clip length doubles.
Water
Where a provider hasn't published a water figure, water is estimated from energy using a configurable intensity (default 1.1 mL per Wh, in line with Google's and OpenAI's own text figures). This covers data-centre cooling and excludes indirect water used in electricity generation, so it is conservative.
Sources: Google Cloud (2025), OpenAI / Sam Altman (2025), Luccioni et al. / Hugging Face & Carnegie Mellon (2023), MIT Technology Review (2025).
Estimates only. Real consumption depends on the specific model, prompt length, data-centre and grid. Figures last reviewed June 2026.