The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink

By David A. Patterson, Joseph E. Gonzalez, Urs Holzle, Quoc V. Le, Chen Liang, Lluís-Miquel Munguía, Daniel Rothchild, David R. So, Maud Texier, Joseph Dean. 2022

465 citations recorded by the source corpus.

Machine learning (ML) workloads have rapidly grown, raising concerns about their carbon footprint. We show four best practices to reduce ML training energy and carbon dioxide emissions. If the whole ML field adopts best practices, we predict that by 2030, total carbon emissions from training will decline.

Subfields: Electrical and Electronic Engineering, Efficiency & Compression, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Electrical and Electronic Engineering

In Efficiency & Compression

In Safety, Robustness & Interpretability

Citations in the corpus

What Do NLP Researchers Believe? Results of the NLP Community Metasurvey (2022, background): “Following this, Patterson et al. (2022) argue that advances in model efficiency and energy management can soon lead to a plateau in energy use from training machine learning models.”
Recent Advances in Natural Language Processing via Large Pre-trained Language Models: A Survey (2021, result): “[132] argues that the carbon footprint of ML training is a few orders of magnitude less than what was estimated in studies such as Strubell et al.”
A first look into the carbon footprint of federated learning (2020, background): “Recent research (Patterson et al., 2022) indicates the approaches to reduce energy and carbon emissions in centralized training in the data centers.”
Connecting the Dots in Trustworthy Artificial Intelligence: From AI Principles, Ethics, and Key Requirements to Responsible AI Systems and Regulation (2023, methodology): “Although emissions are amortized over the model lifetime, the recent ChatGPT model was estimated to consume 1,287 MWh that translates into a cost of 522 tCO2e [94].”
Power Hungry Processing: Watts Driving the Cost of AI Deployment? (2023, unknown): “…approximately one-third of their internal end-to-end ML carbon footprint to model inference, with the remainder produced by data management, storage, and training [57]; similarly, a 2022 study from Google attributed 60% of its ML energy use to inference, compared to 40% for training [40].”
POET: Training Neural Networks on Tiny Devices with Integrated Rematerialization and Paging (2022, background): “While the focus of this paper is edge deployemnts, the energy objective is increasingly becoming relevant even for cloud deployments (Patterson et al., 2022).”
TPU v4: An Optically Reconfigurable Supercomputer for Machine Learning with Hardware Support for Embeddings (2023, background): “There is considerable concern about the carbon footprint of ML [42, 45].”
The Computational Limits of Deep Learning (2020, result): “These carbon estimates are similar to those provided by [86, 88].”
ModernVBERT: Towards Smaller Visual Document Retrievers (2025, unknown): “This estimate follows methodologies such as Green Algorithms (Lannelongue et al., 2021) and related analyses of the carbon footprint of machine learning (Strubell et al., 2019; Patterson et al., 2021).”