Skip to content

Published on

Within Bounds: Limiting AI's environmental impact

Joint statement

Authors: Green Screen Coalition, Green Web Foundation, Beyond Fossil Fuels, Aspiration, critical infrastructure lab

Joint statement from civil society for the AI Action Summit

Signed by over 100 organizations. If you share our concerns and demands that AI systems be made compatible with our planetary boundaries, and are a civil society organization, can you can sign on here.

INTRODUCTION

We are at a critical threshold in our computational futures. Investment in artificial intelligence (AI) is booming, and its application across society is accelerating at an unprecedented scale. Meanwhile in 2024, the 1.5C global warming limit was surpassed across the entire year, and the boundaries of several life-supporting planetary systems have been exceeded. Devastating heat waves, storms, fires and floods remind us of how human activity impacts all life on this planet.

Scientific consensus is clear: fossil fuels must be phased out to reduce the greenhouse gas emissions heating the planet. Yet there is increasing evidence of AI systems driving up emissions and locking in reliance on fossil fuels, while exhausting critical resources like water, land and raw materials, intensifying environmental harms across the technology supply chain, and accelerating the expansion of resource-intensive computational infrastructure beyond sustainable limits.

However, these facts only occupy a marginal position in governance debates. The tech sector continues to operate as if there were no planetary limits. Tech leaders and governments justify further investment in AI systems by emphasising it as a tool for sustainability. However, AI can never be a “climate solution” if it runs on fossil fuels and is used to extract oil and gas.

To meet the challenge of climate change, environmental degradation, pollution and biodiversity loss, and its attendant injustices, we urge policymakers, industry leaders and all stakeholders to acknowledge the true environmental costs of AI, to phase out fossil fuels throughout the technology supply chain, to reject false solutions, and to dedicate all necessary means to bring AI systems in line with planetary boundaries. Meeting these demands is an essential step to ensure that AI is not driving further planetary degradation and could instead support a sustainable and equitable transition.

We, the signatories, demand that AI systems be made compatible with our planetary boundaries, including:

DEMANDS

I. PHASE OUT FOSSIL FUELS

The data centres that power artificial intelligence demand so much energy that we are struggling to meet those demands without slowing down climate progress. Global data centre electricity consumption could double to over 1,000 TWh by 2026—equivalent to Japan’s annual electricity use—according to the International Energy Agency. In parts of the world, this rising demand is pushing power infrastructure to its limits, prolonging and intensifying our dependency on fossil fuels, the pollution from which is linked to public health issues, including lung disease and premature death.

  1. AI infrastructure including data centres must be fossil-free. Burning more fossil fuels to power these data centres would worsen climate impacts and violate international commitments to limit global warming. Already in the US and parts of Europe, data centres are driving an expansion in fossil gas capacity and in some cases keeping coal plants open. AI technologies must not be powered by fossil fuels, including their main power supply, the use of backup diesel generators, and along the entire supply chain.

  2. Tech companies must invest in bringing new and additional renewable energy to power new data centres. Tech companies must commit to run data centres on 100% renewable energy that is locally produced. Renewable energy for new data centres must also be additional to existing or already planned renewables. If new data centre demand simply captures renewables that are needed by other sectors, this does not replace fossil fuels but simply shifts emissions to others. Large tech companies have invested heavily in renewable energy in recent years, it is important though that they use their vast resources to support the wider energy transition by bringing wholly new renewables to market and investing in renewables-based power infrastructure.

  3. Tech companies must stop relying on offsetting in their energy emissions claims. Amazon and Meta are pushing false solutions like ‘emissionality’ that would allow them to claim the use of renewable energy sources while continuing to burn fossil fuels. Others in the industry are moving towards more credible accounting that reduces emissions by aligning their renewable energy purchases with their real-time energy use.

  4. Tech companies must immediately disclose and end contracts that provide AI to the oil and gas industry especially for the purposes of exploration and drilling. AI tools are sold to the oil and gas industry, enabling millions of tonnes more in carbon emissions at a time when the scientific consensus clearly calls for a phase-out of fossil fuels. AI can never be a tool for sustainability so long as it is used to support the oil and gas industry to pump and extract more reserves of fossil fuels.

II. COMPUTING WITHIN LIMITS

The promise of “green AI growth” and “AI for the environment” are often used to justify the expansion of AI computing infrastructure and continue fossil fuel use, while thwarting decarbonisation and other sustainability goals. Although they are investing in renewables, major tech companies like Google and Microsoft are now questioning if they can hit their own climate and energy targets, while Amazon and Meta engage in efforts to enable them to continue to burn fossil fuels while claiming to be 100% renewable. Meanwhile, nuclear energy is being touted as a solution, but it is largely a dangerous distraction. The long timelines of Small Modular Reactors and safety risks of restarting decommissioned nuclear plants, as exemplified by Microsoft’s controversial plan to reopen the Three Mile Island site, further undermine the credibility of this approach. Without addressing the escalating energy demands of AI and their contribution to the climate crisis, the promise of AI as a “climate solution” is pure fiction.

  1. Governments should place moratoria or caps on the energy demand of data centres. If AI infrastructure cannot grow sustainably, it must be limited. Where demands on energy are extremely high, other sectors and uses can and should take priority over unrestrained AI expansion. Sectors like schools, hospitals and households are critical to society and should receive priority access to energy. Electrifying heat and transport, for example, are also critical steps in reducing global carbon emissions and must be prioritised in cases of energy scarcity.

  2. Ensure any new data centre built does not deplete water and land needed for people. Data centres consume vast amounts of water, millions of litres of water a day, equal to the needs of hundreds of thousands of homes and often in drought-prone areas or stressed watersheds. Larger data centre sites may also require millions of square meters in land, resulting in land clearing and the destruction of biodiversity, not to mention air and noise pollution from the site. Especially as data centres become a lucrative real estate market, they should not be in competition with local populations and essential services.

  3. Prioritise grid-aware computing to run heavy computation when demand is low and renewable energy is available. Not all computation has to happen immediately, or at all. Grid-aware computation considers energy needs and the availability of additional renewables, and computation is scheduled by balancing competing needs and operating within agreed usage boundaries. This demand-side flexibility can alleviate pressure on grids and support the transition to renewable energy. Furthermore, we should consider where the need for computation can be reduced, for example when not supporting essential services or the public interest.

  4. Governments must prohibit planned obsolescence and champion the right to repair. Current hardware business models encourage devices, including servers needed for AI, to break prematurely in a practice called planned obsolescence. Industry and policymakers should incentivise circularity and longevity to extend the life of existing hardware. Furthermore, the right to repair and freely modify technology products should be strengthened.

III. RESPONSIBLE SUPPLY CHAINS

The AI industry encompasses cloud computing platforms, data centres, semiconductor manufacturing, specialised services like AI application development, research labs, and the continual testing and development of new models. Concerningly, less regulatory attention is paid to the global supply chain that is implicated in AI, from the mining of raw materials required for the hardware that hosts and runs AI applications to the environmental and health harms stemming from chip fabrication labs or the exorbitant water use in the manufacturing process. While all entities are accountable for their share of environmental and social consequences of AI, it is the companies with substantial market share and economic and political influence who bear the primary responsibility to ensure a responsible supply chain.

  1. Tech companies must reduce emissions in their supply chains in alignment with the best available science. Carbon offsets and other false solutions that do not genuinely reduce emissions, such as experimental carbon capture technologies, must also be rejected. Tech companies must focus instead on taking concrete steps to decarbonise their entire supply chains. In particular, the manufacturing of semiconductors, which underpin advanced AI hardware, is an extremely resource and energy-intensive process. Tech companies must commit to 100% renewable energy by 2030 across their supply chains and support suppliers like semiconductor manufacturers in a clean energy transition and to advocate for more renewable energy sources in places where they are limited.

  2. Tech companies must ensure that the mining for raw materials in their supply chain does not harm the environment or local communities. Violence at the site of extraction often escalates due to a race-to-the-bottom spurred by tech companies’ demand for raw materials at the cheapest price. The results are egregious human rights violations that further entrench existing inequity. Policy makers have a key role as well: policies that seek to increase extraction and consumption of precious and finite raw materials will not ensure a sustainable future.

IV. EQUITABLE PARTICIPATION

It is crucial to have public interest representation in decisions about what computation is used for and under what conditions. The AI market is dominated by a set of privatised, commercial large-language models developed by the world’s most powerful tech companies with no powerful accountability mechanisms. Data centres, subsea cables, and manufacturing sites are expanding across the world without community consultation and with insufficient disclosures from industry about their impacts on the environment, health and human rights. Meanwhile, civic space is shrinking for public debate and collective action, as climate and environmental activism is increasingly criminalised across the world, including work led by human rights defenders and journalists.

  1. Communities impacted across the supply chain must be included in any decision-making, activities and practices that affect them. People are asked to pay huge social costs for the development of AI systems, losing access to land and water, experiencing negative health effects, and paying higher utility bills. Participatory impact assessments and other public fora are necessary so that the people affected by AI infrastructures can fully weigh its tradeoffs and participate in decision-making. This engagement should be well before the onset of a project and provide ample, meaningful, and involved disclosure of the full expected impacts. All affected actors should be included, not just the community geographically adjacent to the project, and it should be part of a continuous consultation rather than a one-off exercise.

  2. Governments must stop criminalising climate action. In the face of climate inaction, it is both natural and necessary for people to speak out and demonstrate against the mounting threats to our planet. However, governments worldwide are intensifying efforts to silence dissent. Environmental activists, journalists, and human rights defenders particularly in the Global Majority are being targeted, surveilled, and killed for their activism while peaceful protest is increasingly criminalised and even classified as terrorism. People must be safe to assemble and speak out.

V. TRANSPARENCY

Transparency empowers the public, decision-makers, civil society, and impacted communities to track progress, make informed decisions and hold stakeholders accountable. Transparency must be meaningful, and publicly accessible information about the social and environmental implications of proposed AI infrastructure and should be provided to the public before it is built or scaled. Reports on energy or water use alone, without equitable participation of the public, fall short of what is needed. While transparency is not a silver bullet, it is a crucial element that aids the public in deciding whether the infrastructure is truly in their interest.

  1. Environmental impacts including energy and water usage must be tracked and measured across the entire AI lifecycle. Much attention has been given to comparing impacts from different kinds of AI systems to identify potentially less energy intensive models; however, not enough data is in the public domain to fully assess impacts across the entire lifecycle, from the training and application (or ‘inference’) phases as well as the supply chain. Reporting and transparency are also needed on the purposes and types of AI models and their related emissions impacts, including enabled emissions, for example, through the application of AI for oil and gas exploration.

  2. AI infrastructure providers must be transparent about the development, resource consumption, and impacts of data centres before they are built. Data centre builders and owners must publicly disclose the locations of data centre sites and their impacts on local communities and the environment, as well as the energy demand of upcoming data centres so that utilities and grid planners can accurately plan for future energy needs. Data centre operators and companies who run AI applications must be required to disclose per data centre their water use and electricity consumption, including actual fossil fuel mix.

  3. Operators must release data on the environmental impact of their hardware development and transportation in a timely, open, and accessible format. Operators of single tenant data centres and customers of multi-tenant data centres (also known as colocations) should report on the environmental impact of their hardware throughout their life-cycle. This includes reporting on the volume of hardware per location, its embodied costs (water, energy, and pollution), the lifespan of hardware in the data centre, and the disposal of hardware.

CONCLUSION

The demands outlined in this letter offer practical pathways to align AI within planetary boundaries. They represent the bare minimum required to mitigate the ongoing harm to our economies, societies, and shared planet.

It is imperative to recognise that AI and its computing infrastructure are not without cost—they are resource-intensive processes that exert significant pressure on our already finite natural resources. The purported benefits and costs themselves are not equitably distributed. Countries and communities most vulnerable to rapid climate change are first to be impacted by the harms of AI and its computational demands, and they have less say in its development.

A paradigm shift is urgently needed. We must move beyond viewing technological progress as inherently beneficial or limitless, and instead prioritise AI processes that contribute meaningfully to society while minimising environmental and human harm. This requires realigning our expectations, setting new standards, and making deliberate choices that reduce AI’s significant environmental cost and develop the technology within limits.

We urge policymakers, industry leaders and all stakeholders in the AI discourse to acknowledge the true environmental costs of AI, to reject false solutions, and to dedicate all necessary means to bring AI systems in line with planetary boundaries.

Only through such shifts can we harness AI as a tool for sustainability, rather than a driver of further planetary degradation.

With hope and action for our collective futures,

Signatories

Access Now
AI Now Institute
AlgorithmWatch
AlgorithmWatch CH
Amnesty International
ARTICLE 19, Office for Mexico and Central America
Aspiration
Association for Progressive Communications - APC
Athena Coalition
Beyond Fossil Fuels
Bits of Freedom
Campaign on Digital Ethics (CODE)
CIS India
Clean Air Action Group
Climate Action Against Disinformation Climate Action Network (CAN) Europe
Climate Change AI
CNCD-11.11.11
Coding Rights
Computer Says Maybe
Consumer Federation of America
Corporate Europe Observatory (CEO)
Critical Carbon Computing Collective
critical infrastructure lab
Danes je nov dan
Data & Society Research Institute
Data for Good
Data Rights
Defend Democracy
Demand Progress Education Fund
Digital Courage
Digital Defenders Partnership
Digital Freedom Fund
Digital Grassroots
Digital Rights Ireland
Digital Technology for Democracy Lab, UVA
donestech
Electronic Frontier Norway
Electronic Privacy Information Center (EPIC)
Environmental Coalition on Standards (ECOS)
Environmental Investigation Agency UK
Equinox Initiative for Racial Justice
European Digital Rights
European Environmental Bureau
European Sex Workers Rights Alliance
Existential Risk Observatory
Faktograf - the Association for the Informed Public
FAIR SHARE of Women Leaders
Feminist Collective of Romani Gender Experts
Friends of the Earth (England, Wales and Northern Ireland)
Friends of the Earth Ireland
Friends of the Earth US Glitch
Global Voices
Green Coding Solutions
Green Web Foundation
Greenpeace Greece
Heinrich-Böll-Stiftung
Hermes Center
Homo Digitalis
Hubblo
Idec - Instituto de Defesa de Consumidores
Irish Council for Civil Liberties, Enforce
Kairos Action
La Quadrature du Net
Latin American Institute of Terraforming
Mouvement écologique (Luxembourg)
neuland - Büro für Informatik GmbH
Numun Fund
Oeko-Institut
Open Future
Open Knowledge Foundation
Organisation Féministe MARIJÀN (OFMA)
Polska Zielona Sieć
ProtocolLabs
R3D
Racism and Technology Center
Re-set: platform for socio-ecological transformation
Resource Environmental Center (REC) in Albania
Right to Repair Europe
Romnja Feminist Library
SHARE Foundation
Society for the Earth (TNZ)
SOMO
Special interest group “Greening the Digital Society”
Stand.earth
Satewatch
Stowarzyszenie Ekologiczne EKO-UNIA
SUPERRR Lab
Technopolice
TEDIC
The Restart Project
Trans-Atlantic Consumer Dialogue (TACD)
TuNubeSecaMiRio
Usuarios Digitales
Waag Futurelab
Whose Knowledge?
WISE
Women Against Fascism (WAF)
Workshop for All Beings
ZERO
ZVEZA POTROŠNIKOV SLOVENIJE

Bibliography

AI in fossil fuel exploration

Cushing, Simon, James Ingham, and Inna Agamirzian. 2024. Compare AI Software Spending in the Oil and Gas Industry, 2023-2027. Gartner. March 27, 2024

EY. 203AD. Applying AI in Oil and Gas. Survey removed from site; accessed at Internet Archive October 9, 2024,

Hao, Karen. 2024. Microsoft’s Hypocrisy on AI. The Atlantic, September 13, 2024, sec. Technology

Stone, Maddie. 2024. Microsoft Employees Spent Years Fighting the Tech Giant’s Oil Ties. Now, They’re Speaking out. Grist, May 8, 2024

Air pollution

Han, Yuelin, Zhifeng Wu, Pengfei Li, Adam Wierman, and Shaolei Ren. 2024. The Unpaid Toll: Quantifying the Public Health Impact of AI. arXiv

Wittenberg, Ariel. 2023. Coal Power Kills a ‘Staggering’ Number of Americans. Scientific American, August 11, 2023

Energy and emissions

Adams, Chris. 2025. Will New Models like DeepSeek Reduce the Direct Environmental Footprint of AI? Reads, Takes and Links. January 27, 2025

Beyond Fossil Fuels. 2024. 24/7 Renewable Electricity Matching Is a Far More Credible Approach for the GHG Protocol and the SBTi than the Emissions First Partnership Proposal. Beyond Fossil Fuels. October 10, 2024

Çam, Eren, Zoe Hungerford, Niklas Schoch, Francys Pinto Miranda, and Carlos David Yáñez de León. 2024. Electricity 2024: Analysis and Forecast to 2026. International Energy Agency.

Enabled Emissions Campaign

IER. 2024. Big Corporations Are Abandoning Their Climate Commitments. Institute for Energy Research. September 23, 2024.

Kamiya, George, and Paolo Bertoldi. 2024. Energy Consumption in Data Centres and Broadband Communication Networks in the EU. JRC135926. Luxemburg: Publications Office of the European Union

Kerr, Dara, dir. 2024. AI Brings Soaring Emissions for Google and Microsoft, a Major Contributor to Climate Change. NPR.

Milmo, Dan. 2024. Google’s Emissions Climb Nearly 50% in Five Years Due to AI Energy Demand. The Guardian, February 7, 2024, sec. Technology

Mooldijk, Silke, Thomas Day, and Sybrig Smit. 2024. Navigating the Nuances of Corporate Renewable Electricity Procurement: Spotlight on Fashion and Tech. Berlin and Cologne, Germany: NewClimate Institute

O’Brien, Isabel. 2024. Data Center Emissions Probably 662% Higher than Big Tech Claims. Can It Keep up the Ruse? The Guardian, September 15, 2024

Ray, Siladitya. 2024. Amazon Claims It Achieved 100% Renewable Energy Target In 2023—Here’s What It Means. Forbes, October 7, 2024

Rick, August, Katrin Wu, and Tianyi Luo. 2023. Invisible Emissions - A Forecast of Tech Supply Chain Emissions and Electricity Consumption. Greenpeace East Asia.

Stand.Earth. 2024. Microsoft’s Application of AI to Accelerate Oil & Gas Expansion. 2024. Stand.Earth. May 12, 2024

Zhang, Xixi. 2024. Ctrl-Alt-Incomplete: The Gaps in Microsoft’s Climate Leadership. STAND.earth

Greenwashing and false solutions

Jansen, Fieke, and Michelle Thorne. 2024. Predatory Delay and Other Myths of ‘Sustainable AI.’ In Redirecting Europe’s AI Industrial Policy: From Competitiveness to Public Interest, edited by Frederike Kaltheuner, Leevi Saari, Amba Kak, and Sarah Myers West. AI Now

Kazansky, Beck, and Nikita Kekana. 2023. Coming Together to Counter Misleading and False Climate/Tech Solutions. Branch Magazine. September 6, 2023

Lamb, William F., Giulio Mattioli, Sebastian Levi, J. Timmons Roberts, Stuart Capstick, Felix Creutzig, Jan C. Minx, Finn Müller-Hansen, Trevor Culhane, and Julia K. Steinberger. 2020. Discourses of Climate Delay. Global Sustainability 3:e17

Project Drawdown uses different scenarios to assess what determined, global efforts to address climate change might look like. Scenarios shown here are plausible and economically realistic. Drawdown Scenario 1 is roughly in line with a 2˚C temperature rise by 2100, while Drawdown Scenario 2 is roughly in line with a 1.5˚C temperature rise at century’s end. AI’s role in any of these solutions is limited, if needed at all. See “Table of Solutions,”

Steffen, Alex. 2016. Predatory Delay and the Rights of Future Generations. Medium, April 29, 2016

Hardware

Bordage, Frederic, Lorraine de Montenay, Sofia Benqassem, Julie Delmas-Orgelet, Firmin Domon, Damien Prunel, Caroline Vateau, and Etienne Lees Perasso. 2021. Behind the Figures: Understanding the Environmental Impacts of ICT and Taking Action.

Chen, Ann, and Ingrid Burrington. 2023. Manufacturing: The Semiconductor Is an Accretion of Place. Geographies of Digital Wasting. 2023

Smolaks, Max. 2022. Nvidia’s H100 – What It Is, What It Does, and Why It Matters Data Center Knowledge. March 23, 2022

Justice

Agarwal, Anil, and Sunita Narain. 2019. Global Warming in an Unequal World: A Case of Environmental Colonialism.. In India in a Warming World: Integrating Climate Change and Development, edited by Navroz K. Dubash, 0. Delhi: Oxford University Press

Brand, Ulrich, Barbara Muraca, Éric Pineault, Marlyne Sahakian, Anke Schaffartzik, Andreas Novy, Christoph Streissler, et al. 2021. From Planetary to Societal Boundaries: An Argument for Collectively Defined Self-Limitation. Sustainability: Science, Practice and Policy 17 (1): 264–91

Saul, Josh, Naureen S Malik, and Mark Chediak. 2024. AI Boom Is Driving a Surprise Resurgence of US Gas-Fired Power. Bloomberg, September 16, 2024

Velkova, Julia. 2024. Dismantling Public Values, One Data Center at the Time. Reimagining Public Values in Algorithmic Futures, University of Helsinki. February 19, 2024

Market concentration

Ensmenger, Nathan. 2021. The Cloud Is a Factory. In Your Computer Is on Fire, edited by Thomas S. Mullaney, Benjamin Peters, Mar Hicks, and Kavita Philip, 29–50. The MIT Press.

Smith, Hannah, and Chris Adams. 2024. Thinking About Using AI? Here’s What You Can and (Probably) Can’t Change About Its Environmental Impact. Green Web Foundation

Thorne, Michelle. 2024. Critical Dependencies: How Power Consolidation of Digital Infrastructures Threatens Our Democracies—and What We Can Do About It,. Green Web Foundation.

Schulze, Max, Radika Kumar, and Michael Oghia. 2022. Taxonomy Guide: Infrastructure in the Digital Economy, Trade Competitiveness Briefing Paper. Commonwealth Secretariat.

Water

Anson, April, Andrea Ballestero, Dean Chahim, CIEJ, Theodora Dryer, Sage Gerson, Matthew Henry, et al. 2022. Water Justice and Technology: The COVID-19 Crisis, Computational Resource Control, and Water Relief Policy. New York University: AI Now Institute

Li, Pengfei, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren. 2025. Making AI Less ‘Thirsty’: Uncovering and Addressing the Secret Water Footprint of AI Models. arXiv