AI Energy & Water per Query Calculator

Estimate the electricity, water and CO2e behind your AI prompts, images and video seconds using published figures from Google, OpenAI, Epoch AI, the IEA and peer-reviewed studies.

Switch between low, median and high estimates and set your own grid intensity and water use effectiveness.

Last updatedHow we build & check our tools
Results will appear here once you enter your values.

How the estimate is calculated

Energy first. Each item is multiplied by a published energy figure: prompts per day times Wh per prompt, plus images times Wh per image, plus video seconds times Wh per second. Monthly totals use 365 / 12 days and yearly totals use 365 days.

Carbon and water from energy. Emissions in grams equal kWh times grid intensity in gCO2e per kWh. The default is 349.7 g/kWh, the US average from EPA eGRID2023 (770.884 lb CO2e per MWh). Water in mL equals Wh times water use effectiveness in L/kWh, because the two factors of 1,000 cancel. The default of 0.55 L/kWh is the US average on-site WUE used by Li et al. in Making AI Less Thirsty. Tick the off-site option to add 3.142 L/kWh for water evaporated at power plants.

Worked example. The default of 20 ChatGPT prompts at 0.34 Wh and 2 images at 2.907 Wh is 12.6 Wh per day, or 4.6 kWh per year. That is about 2.5 liters of on-site water, 1.6 kg CO2e on the average US grid and 242 smartphone charges at EPA's 19 Wh per charge.

Where the per-query numbers come from

Text prompts. As of September 2026 the lowest figure is Google's measured median Gemini Apps text prompt from August 2025: 0.24 Wh, 0.26 mL of water and 0.03 gCO2e. OpenAI CEO Sam Altman wrote in June 2025 that the average ChatGPT query uses about 0.34 Wh and 0.000085 gallons (0.32 mL) of water. Epoch AI estimated about 0.3 Wh for a 500-token GPT-4o response in February 2025, rising to about 2.5 Wh with 10,000 input tokens. The high end is the IEA's 2024 figure of 2.9 Wh and Li et al.'s 4 Wh server estimate for GPT-3.

Images and video. Luccioni, Jernite and Strubell measured a mean of 2.907 kWh per 1,000 images across image generation models, with the least efficient at 11.49 kWh. MIT Technology Review measured 2,282 J (0.63 Wh) for a 1024 by 1024 Stable Diffusion 3 Medium image and about 3.4 million J (944 Wh) for a five-second CogVideoX video, roughly 50 phone charges.

Why the uncertainty is so wide

Estimates, not measurements. Google's figure is the only one measured and published by the operator with its method; OpenAI's is a company statement without methodology, and the rest are outside estimates built from assumptions. The published text figures span 0.24 to 4 Wh, a range of more than 16 times.

What changes the number. Long documents, reasoning models, large context windows, older hardware and low batch sizes all raise energy per prompt. Accounting choices matter too. Google's 0.03 g CO2e per 0.24 Wh prompt implies about 125 g/kWh, far below the US grid average, and its 0.26 mL implies about 1.08 L/kWh of water. Use these results to compare scenarios, not as a precise personal footprint.

Putting AI water usage in context

Everyday equivalents. The calculator compares your yearly energy to smartphone charges (19 Wh each, EPA), hours of a 9 W LED bulb, and kettle boils, where heating 1 liter of water from 20 to 100 degrees Celsius takes 4,184 J per kg per degree times 80 degrees, about 93 Wh.

Your own grid. Replace the grid intensity with your region's value from eGRID or your utility to see how location changes emissions. Water use depends on the data center's cooling design and local climate, so set WUE from your provider's sustainability report when it publishes one.

Frequently Asked Questions

Common questions about the AI Energy & Water per Query Calculator

OpenAI CEO Sam Altman stated in June 2025 that the average ChatGPT query uses about 0.000085 gallons, roughly 0.32 mL or one fifteenth of a teaspoon. Google reported 0.26 mL for a median Gemini Apps text prompt in August 2025. Older academic estimates that include water used by power plants are much higher, around 17 mL per GPT-3 response.
From the same team

Stop paying per token — route AI requests to your own GPU

Wide Area AI is a local-first AI gateway: repeated requests hit an edge cache, the rest run free on your own hardware, and the cloud is only a failover. OpenAI-compatible endpoint, free tier.

Start routing — free
VRAM CalculatorCan I Run This AI?AI API Cost CalculatorGPU for Model

Sources & References

Measuring the environmental impact of AI inference

August 21, 2025 median Gemini Apps text prompt: 0.24 Wh, 0.26 mL water, 0.03 gCO2e.
→Source: Google Cloud blog

The Gentle Singularity

June 2025 statement that an average ChatGPT query uses about 0.34 Wh and 0.000085 gallons of water.
→Source: Sam Altman

Making AI Less Thirsty: Uncovering and Addressing the Secret Water Footprint of AI Models

0.004 kWh per GPT-3 request, US average on-site WUE 0.55 L/kWh and off-site 3.142 L/kWh, 16.9 mL per request.
→Source: Li, Yang, Islam, Ren (arXiv 2304.03271)

Power Hungry Processing: Watts Driving the Cost of AI Deployment?

Image generation mean of 2.907 kWh per 1,000 inferences and 11.49 kWh for the least efficient model.
→Source: Luccioni, Jernite, Strubell (arXiv 2311.16863)

eGRID Summary Data (2023 data)

US total output emission rate of 770.884 lb CO2e/MWh, the 349.7 g/kWh default grid intensity.
→Source: US EPA

Explore More Tools

Explore these related tools for your next step.

ai

LLM Token Usage Calculator

Calculate token usage and associated costs for LLM prompts.

Try it now
ai

AI Embedding Cost Calculator

Estimate the total token count and API cost to embed a document corpus using OpenAI, Cohere, or Voyage embedding models.

Try it now
ai

Neural Network Playground

Train a real neural network in your browser — no code, no installs. Pick a sample dataset (Titanic survival, vehicle classifier, handwritten digits) or upload your own CSV, watch the network learn with live visuals, then ask it questions and export the model as Python/Keras code.

Try it now
ai

LLM Token Counter

Count tokens in text for GPT, Claude, Gemini, and open models like Llama, Gemma, and Qwen — or search any model on Hugging Face. Estimate API costs, check context window fit, and optimize prompts

Try it now
ai

LLM VRAM Calculator

Calculate how much VRAM any LLM needs to run locally — pick a model (Llama, Gemma, Qwen, DeepSeek, or search Hugging Face), choose a quantization and context size, and see which GPUs it fits on, including multi-GPU setups.

Try it now
ai

LLM Inference Speed Calculator

Estimate LLM tokens per second from memory bandwidth, model size, quantization, and context window. Compare generation speed across GPUs (RTX 4090, 5090, A100, H100, Apple Silicon) and understand the memory-bandwidth bottleneck.

Try it now