Environmental Implications of AI

Energy, water, hardware lifecycle, and the rebound problem

Contents

Reading · Papers to read

9 papers covering AI energy and water consumption, embodied carbon, the Jevons rebound, sustainable AI practice, and the token cost of agentic coding. The OECD policy report and several agentic-energy references are link-only.

Each entry links to the canonical version of the paper, on arXiv, the journal, or the publisher. Where a paper is paywalled, the DOI is given for UCT-library access.

3.1 · What Does AI Actually Consume?

Making AI Less ‘Thirsty’: Uncovering and Addressing the Secret Water Footprint of AI Models

Li, P., Yang, J., Islam, M. A., & Ren, S. (2023)

View source · arXiv:2304.03271

Power Hungry Processing: Watts Driving the Cost of AI Deployment?

Luccioni, S., Jernite, Y., & Strubell, E. (2024) — FAccT ’24

View source · arXiv:2311.16863

How Do AI Agents Spend Your Money? Analyzing and Predicting Token Consumption in Agentic Coding Tasks

Stanford Digital Economy Lab & Microsoft Research (2026) — supports the agentic-multiplier estimate in 3.1

View source · arXiv:2604.22750

3.2 · Infrastructure, Scale and the Rebound Problem

Chasing Carbon: The Elusive Environmental Footprint of Computing

Gupta, U., et al. (2021) — IEEE HPCA 2021

View source · arXiv:2011.02839

Efficiency is Not Enough: A Critical Perspective of Environmentally Sustainable AI

Wright, D., Igel, C., Samuel, G., & Selvan, R. (2023)

View source · arXiv:2309.02065

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

Patterson, D., et al. (2022)

View source · arXiv:2204.05149

3.3 · Critical Minerals and AI

Sub-lesson uses news reporting and policy documents (CHIPS Act, EU CRMA, US Geological Survey) which aren’t redistributable PDFs.

3.4 · Sustainable AI: What Can Be Done?

Green AI

Schwartz, R., Dodge, J., Smith, N. A., & Etzioni, O. (2019)

View source · arXiv:1907.10597

Tackling Climate Change with Machine Learning

Rolnick, D., et al. (2019)

View source · arXiv:1906.05433

Measuring the Carbon Intensity of AI in Cloud Instances

Dodge, J., et al. (2022) — FAccT ’22

View source · arXiv:2206.05229

The agentic-energy estimate in Sub-Lesson 3.1 also draws on these non-paper sources (reports and documentation rather than academic papers):

Stanford Digital Economy Lab (2026). How are AI agents spending your tokens? — analysis of the token-consumption study above.
Anthropic (2026). Effort documentation (Claude effort levels: low / medium / high / xhigh / max) — platform.claude.com
US EPA. Greenhouse Gas Emissions from a Typical Passenger Vehicle (~0.25 kg CO₂/km) — epa.gov
Republic of South Africa (2024). 2022 Grid Emission Factors Report (~0.9 kg CO₂/kWh) — gov.za PDF

Linked but not redistributed

OECD (2022). Measuring the environmental impacts of artificial intelligence compute and applications. DOI:10.1787/7babf571-en 3.4

Open-access on the OECD library. It is included as a link rather than a downloaded copy because the OECD URL changes occasionally.

← PreviousWeek 2 Next →Week 4

Drafted with Claude (Anthropic) and reviewed by Jonathan Shock before publication. AI-generated errors are possible — if you spot one, please email jonathan.shock@uct.ac.za. Full detail: AI Content Disclaimer.

Reading
2 min read