Calculation Explanation
Select a model and consumption type to see how the calculations were done.Creating an AI model requires training it on an extremely large number of parameters. This causes the company creating the model to use an extremely large amount of water in the training phase. If you would like to see the number of parameters some of the most popular AI models are trained on, visit Our World in Data1for information on over 900 different models.
Below are the self-reported water use trends (in cubic meters) of some large companies that have created their own AI models, such as Google2, Microsoft3, and Meta4, over the past few years.
Water Usage Data by Company (in cubic meters)
| Company | 2021 | 2022 | 2023 |
|---|---|---|---|
| 17,268,291 | 21,064,680 | 24,044,935 | |
| Microsoft | 4,772,890 | 6,399,415 | 7,843,744 |
| Meta | 2,569,000 | 2,638,000 | 3,078,000 |