Estimate your organisation’s AI footprint.
CO₂e, electricity and water from the AI tools your teams use, estimated with SOMA AI’s published method. Answer a few questions and your results build here, live.
4 or 5 quick questions, about 2 minutes. No sign-up, nothing to upload.
Your footprint
CO₂e, electricity and water per month, with a range.
By team and tool
Which teams and tools drive it.
Where your AI runs
The electricity grid behind each data centre.
Your action plan
Changes you can actually make, each recalculated.
Method: SOMA AI inference factors v3 · DOI 10.5281/zenodo.23371447 · peer review in progress
How the free AI footprint calculator works
What it estimates
The CO₂e, the electricity (including data-centre cooling) and the water behind the AI tools your organisation uses: chat assistants such as ChatGPT, Microsoft 365 Copilot, Claude and Gemini, coding tools such as GitHub Copilot and Cursor, and AI APIs on OpenAI, Anthropic, Azure, AWS Bedrock and Google Vertex AI. Per team and per tool, with a range.
How it works
Your answers become usage: people, messages or tokens per tool. SOMA AI’s inference factors turn that usage into energy, carbon and water by model class and by the region where the provider runs the model. The factors are an open dataset (DOI 10.5281/zenodo.23371447) and the method is in peer review.
From rough to sharp
A first estimate needs only tools, people and how often they use AI (Tier 2b). Message counts from an admin export and token numbers from an invoice (Tier 2a) replace the defaults, and the data completeness score shows how much of the input is your own data.
Read the methodology, browse the AI emission factors, or see the AI footprint data hub.
Frequently asked questions
What does the free AI footprint calculator estimate?
It estimates the greenhouse-gas emissions (CO₂e), the electricity (including data-centre cooling) and the water behind the AI tools your organisation uses, such as ChatGPT, Microsoft 365 Copilot, Claude, Gemini, GitHub Copilot and the AI APIs. You get a figure per month, quarter or year, split by team and by tool, with a range. It is an estimate from published defaults, not a measurement.
How is the estimate calculated?
With SOMA AI's published inference factors (DOI 10.5281/zenodo.23371447). Each tool is mapped to a model class (small, mid-size or frontier), with energy per token from the ML.ENERGY Leaderboard v3 on H100 GPUs, a data-centre overhead (PUE) of 1.2 and the grid carbon intensity of the region where the provider runs the model (Ember 2023, US EPA eGRID 2023). Hardware manufacture, data-centre construction and final training runs are added, spread over their life. The method is in peer review.
What information do I need?
For a first estimate: which AI tools your teams use, how many people use each one and how often. Each further answer replaces a default with your own data: the models people pick, where your providers run them, message counts from an admin export, and token numbers from an invoice or API dashboard.
What do the range and the data completeness score mean?
The range is the spread of measured models within each class and, for estimates from seats or exports, of message sizes. It is not a confidence interval, and the true value is more likely above the central figure than below it. Data completeness shows how much of the input is your own data rather than defaults. It is a disclosed scoring rule, not a probability or an accuracy figure.
What are the data tiers?
Estimates from seats, usage bands or admin message counts are Tier 2b. Estimates from token data or token logs are Tier 2a. Tier 3 is reserved for a provider's own audited carbon figure for your usage, which no major AI provider offers yet. The free calculator covers Tier 2b and token data; token logs need the full SOMA AI calculator.
Does the figure include training the AI models?
It includes the final training run of each model, spread over the tokens the model serves in its service life. That is a lower bound. All model development (experiments, failed and final runs) is reported separately as a development line, about 1.3 times the inference footprint (0.6 to 2.2 times), and is never added to the total.
How do I report the result?
In a greenhouse gas inventory, as Scope 3 Category 1 (purchased goods and services, the purchased cloud and data-centre services sub-category), location-based, grid year 2023. Remove the same supplier spend from any spend-based Category 1 line so it is not counted twice. The same figure supports a CSRD (ESRS E1) disclosure.