Month at this pace
0 g
0 Wh, 0 L water
AI Footprint Calculator
Enter the AI work you actually used. Get a live estimate for energy, water, and CO2 with comparisons that are easier to picture.
Similar everyday use
Environmental summary
Year at this pace
0 g
0 Wh, 0 L water
10-year outlook
0 g
0 Wh, 0 L water
Restore resources
What you can do to make it back
Estimates use industry-average data and may vary by provider, region, and model. Learn more in methodology.
Understand the environmental impact of everyday AI use
AI requests are powered by data centers that use electricity, cooling systems, servers, chips, networks, and repeated computation. This calculator turns those hidden resources into a practical estimate you can compare, share, and reduce.
What is an AI footprint?
An AI footprint is the estimated energy, water, and CO2 connected to using AI tools. It changes with the kind of task, the amount of output, the model size, repeated attempts, and the electricity and cooling assumptions behind the request.
How this calculator estimates impact
The tool groups activity into text prompts, AI search, coding or deep reasoning, file analysis, image generation, voice or audio, and video. It applies public research baselines and global defaults so users do not have to guess data-center details.
Ways to reduce AI energy use
Start with the largest driver in your result. Batch small questions, reuse good answers, cap rerolls, avoid unnecessary uploads, test short media clips before full renders, and choose lighter models when the task does not need deep reasoning.
Further reading
Practical guides for using AI with more care
These pages go deeper than the calculator: environmental impact, privacy, security, and better everyday habits for people who use AI at work or in public.
AI environmental impact
Energy, water, CO2, data centers, model size, and the habits that change the footprint.
Privacy risks of AI
What not to paste into prompts, uploads, screenshots, and workplace AI tools.
AI security risks
Prompt injection, connected tools, sensitive files, and the approval steps worth keeping.
Better AI habits
Simple routines that lower waste, reduce risk, and make AI work cleaner.
Is AI always worse than normal search?
Not always. It depends on the model, answer length, number of searches, and whether the AI task replaces or adds extra work. The calculator estimates your entered activity instead of making one universal claim.
Why does media generation use more energy?
Images, audio, and video usually require more compute than short text. Repeated rerolls and upscales can quickly become the main driver, which is why the calculator separates media from ordinary prompts.
Can I share an exact result?
Yes. Use the copy link action to save the current inputs in the URL, or download the result card as a PNG for social posts, reports, and notes.
01
Methodology
Open for more info
The calculator separates your usage into short text, AI search, coding/deep reasoning, file or document analysis, image generation, voice/audio generation, and experimental video. A typical public-AI default sets the baseline; the action counts automatically apply heavier factors where needed.
The main result estimates serving energy, then adds a small default allowance for hidden infrastructure such as training, chips, server replacement, e-waste, office energy, and networking. It is included so those impacts are not treated as zero, but it is still an estimate.
The app uses worldwide defaults for carbon and water because most users cannot know which data center processed a request. The result is a central estimate based on public research, not exact metering.
- IEA, Energy and AI
- IEA, Key Questions on Energy and AI
- Oviedo et al., Energy Use of AI Inference, Efficiency Pathways, and Test-Time Scaling
- Goldman Sachs, data center power demand and AI query estimate
- Elsworth et al., Google-scale AI serving impact
- Luccioni, Jernite, and Strubell, Power Hungry Processing
- Li, Yang, Islam, and Ren, Making AI Less Thirsty
- Lambert and Luccioni, From Cradle to Cloud
- Guidi et al., U.S. hyperscale data-center carbon and energy assessment
- Patterson et al., Carbon Emissions and Large Neural Network Training
- Luccioni, Viguier, and Ligozat, Carbon Footprint of BLOOM
- Vanderbauwhede, emissions from AI-augmented search
- Shumba et al., water efficiency dataset for data centers
- Wang et al., E-waste challenges of generative artificial intelligence
- U.S. EPA greenhouse gas equivalencies references
02
What the result can and cannot say
Why do results change by action?
Different actions use different amounts of compute. A short chatbot answer, search request, coding task, file analysis, image, voice clip, and video do not have the same footprint.
What if I used several AI products?
Count the actions together. The estimate uses typical public-AI defaults and focuses on the kind of work done: search, coding, file analysis, image generation, audio, video, or short text.
Why include infrastructure allowance?
Training, chip manufacturing, server replacement, e-waste, offices, and networking are part of the footprint, but public data is not consistent enough to assign them exactly to one prompt. The calculator adds a small default base instead of pretending those impacts are zero.
Is water direct cooling water only?
No. The water result is a practical global estimate that combines data-center cooling water with water linked to the electricity used to run the AI request.