Holistically Evaluating the Environmental Impact of Creating Language Models

Jacob Morrison, Clara Na, Jared Fernandez, Tim Dettmers Emma Strubell Jesse Dodge · 2025

Read Holistically Evaluating the Environmental Impact of Creating Language Models by Jacob Morrison, Clara Na, Jared Fernandez, Tim Dettmers Emma Strubell Jesse Dodge (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-05804.txt.

Description

Headlines about artificial intelligence and the climate usually quote one training run. Jacob Morrison and five co-authors argue the honest bill is much longer. They measured the full cost of creating a family of language models, from 20 million to 13 billion active parameters. Their count covered hardware manufacturing, the whole development process, water use, and the final training runs.

The family released 493 metric tons of carbon in total, about a year of power for 98 American homes, and used 2.77 million liters of water. Development, the experiments and dead ends before the final run, caused roughly half the training footprint, a share developers almost never disclose. Power draw during training also swung wildly, between about 15 and 85 percent of maximum, which complicates planning for power grids.

Catalog record

CREATORJacob Morrison, Clara Na, Jared Fernandez, Tim Dettmers Emma Strubell Jesse Dodge
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-05804

Browse the full library · About the Tower of Babel