Sources and methodology

Our methodology relies on recent scientific and institutional sources.

Calculation methodology

How Calculia estimates your environmental footprint

Calculia uses a multi-criteria approach to estimate the environmental impact of your generative AI usage. The global score (0-100) combines several dimensions: the LOWER the score, the LOWER your footprint.

Main sources

Explained: Generative AI's environmental impact

MIT News2025-01-17

Data centers: 460 TWh in 2022, expected 1050 TWh by 2026. GPT-3 training: 1287 MWh, 552 tons CO2. Cooling: 2L water per kWh.

energyco2waterdata-centers

AI boom has caused same CO2 emissions in 2025 as New York

The Guardian2025-12-18

AI carbon footprint 2025: up to 80 million tons CO2. Water consumption: up to 765 billion liters.

co2waterglobal-impact

How much energy does ChatGPT use?

Epoch AI2025-02-07

ChatGPT GPT-4o: approximately 0.3 Wh per text query, 10x less than older estimates.

energyinferencechatgpt

Measuring the environmental impact of AI inference

Google Cloud2025-08-21

Gemini LLM: approximately 0.24 Wh per text query, equivalent to 1 second of microwave use.

energyinferencegemini

Energy and AI – Analysis

International Energy Agency (IEA)2025-04-10

Comprehensive report on AI energy consumption with global and regional modeling.

energyglobal-impactpolicy

Additional sources

Tracking the carbon footprint of global generative AI

Cell / The Innovation2025

Artificial Intelligence, Data, Computing: Final Report

The Shift Project2025-10-01

US data centers' energy use amid the AI boom

Pew Research Center2025-10-24

New milestone in generative AI environmental transparency

Carbone 42025-07-23

We did the math on AI's energy footprint

MIT Technology Review2025-05-20

Sources and methodology

1. Global score calculation

2. Energy estimates (kWh)

3. CO₂ emissions (kgCO₂e)

4. Water consumption (liters)

5. Data risk score

6. Educational equivalents

7. Limitations and caveats