AI Could Cut Global Emissions by 5 Billion Tonnes: What Environmental Nonprofits Should Understand
A landmark study from the London School of Economics projects AI could reduce global emissions by up to 5.4 billion tonnes per year by 2035. The research is compelling and the caveats are substantial. Here is what every environmental nonprofit needs to know about both.
A major research report published in the journal npj Climate Action has landed in the center of the climate AI conversation with a striking headline finding: artificial intelligence applied to food systems, electricity grids, and transportation could reduce global greenhouse gas emissions by between 3.2 and 5.4 billion tonnes of CO2-equivalent per year by 2035. That range represents roughly the annual emissions of the entire European Union. The study, led by researchers from the London School of Economics and Systemiq, frames this as an "unprecedented opportunity" for AI to serve the climate transition.
For environmental nonprofits, this finding warrants both serious attention and serious scrutiny. Organizations working on climate, conservation, clean energy, and environmental justice are already grappling with whether and how to incorporate AI into their work. A study projecting gigaton-scale emissions reductions from AI deployment raises the stakes of that question considerably. It also raises questions that the headline figure alone does not answer: which applications actually deliver reductions, under what conditions, with what risks, and what role environmental nonprofits can and should play in realizing or governing these outcomes.
At the same time, a contrasting body of evidence is emerging. A 2026 report examining AI greenwashing found that major tech companies' AI climate promises have largely not delivered verifiable emissions reductions, and that no current example of generative AI has produced material, measurable environmental benefits at scale. These two sets of findings are not necessarily contradictory, but they suggest that the path from AI's theoretical emissions potential to realized climate impact is longer and more complicated than the most optimistic framing implies.
This article unpacks the evidence, examines where AI is already producing environmental results (as opposed to projecting them), considers the important counterarguments including AI's own energy footprint, and identifies the concrete ways environmental nonprofits can engage with AI both as a tool for their own work and as a policy and advocacy issue in the climate space.
What the Research Actually Shows
The LSE/Systemiq report "Green and intelligent: the role of AI in the climate transition" focuses on three sectors that collectively represent a substantial share of global emissions and where AI applications are already demonstrably reducing environmental impact. Understanding the structure of the argument, not just the headline number, is essential for evaluating its relevance to environmental nonprofit strategy.
Food Systems
0.9-1.6 billion tonnes CO2e/year potential
AI applications in agriculture include precision crop management that reduces fertilizer and pesticide use, demand forecasting that cuts food waste, and supply chain optimization that reduces emissions from cold storage and distribution. These applications are already operating at scale in commercial agriculture.
- Precision fertilizer application reduces nitrous oxide
- Food waste reduction through inventory AI
- Livestock emissions monitoring and management
Electricity Grids
1.5-2.3 billion tonnes CO2e/year potential
The electricity sector holds the largest identified potential. AI can optimize grid management to reduce curtailment of renewable energy, improve demand forecasting to reduce reliance on peaking plants, and accelerate integration of distributed energy resources like rooftop solar and battery storage.
- Smart grid optimization reduces transmission losses
- Renewable curtailment reduction through better scheduling
- Building energy management and demand response
Mobility and Transport
0.8-1.5 billion tonnes CO2e/year potential
Transportation AI ranges from route optimization for logistics fleets to traffic flow management that reduces idling, EV charging optimization, and autonomous vehicles that may drive more efficiently than human operators over time.
- Freight route optimization reducing empty miles
- Traffic management systems reducing urban idling
- Modal shift promotion through mobility AI platforms
A critical nuance in the research: these projections represent the upper bound of what is technically achievable if AI is widely deployed and effectively managed in these sectors by 2035. They are not predictions of what will happen under current trajectories. The difference matters enormously. The gap between potential and realized impact is governed by deployment rates, policy environment, infrastructure availability, and how AI tools are actually used in practice, all of which are uncertain.
The researchers also explicitly acknowledge that AI's own energy consumption must be netted out of these estimates. Data centers running AI systems consume significant electricity and produce substantial emissions. The study concludes that "the estimated emissions reductions would outweigh increases from global power consumption of data centres and AI," but this conclusion depends on assumptions about grid decarbonization rates and AI efficiency improvements that may or may not materialize on the projected timeline.
The Greenwashing Problem: Where AI Climate Claims Fall Short
Environmental nonprofits should engage with the emissions reduction potential of AI with clear eyes about the current state of evidence, particularly regarding generative AI tools. The same year the LSE study was published, research examining AI company climate claims found a stark gap between promises and verifiable results.
What the Greenwashing Research Found
Key findings from 2026 analysis of Big Tech AI climate claims
- No current example found where generative AI tools like ChatGPT, Gemini, or Copilot produced "material, verifiable and substantial" emissions reductions
- Major technology companies' AI climate commitments largely lack measurable milestones or third-party verification
- The aggregate electricity demand of AI data centers is rising faster than clean energy deployment in most regions
- AI climate projections are frequently cited by tech companies in ways that obscure the distinction between potential and current impact
This tension between potential and present reality matters for how environmental nonprofits position themselves in relation to AI. Organizations that accept AI climate claims at face value risk inadvertently lending credibility to corporate sustainability narratives that are not yet backed by measurable outcomes. Organizations that dismiss AI's climate potential entirely risk missing genuine tools that are already delivering environmental results in specific contexts, particularly in areas like emissions monitoring, biodiversity protection, and precision conservation.
The distinction that matters most is between purpose-built AI applications in specific environmental domains (where evidence of impact is accumulating) and general-purpose generative AI tools whose primary function is content generation and where environmental claims rest almost entirely on projected rather than realized impact. Environmental nonprofits are well-positioned to make this distinction clearly, both in their own technology choices and in their advocacy work.
AI's Own Environmental Footprint
Any honest accounting of AI's role in climate must reckon with the energy and resource demands of AI systems themselves. This is not a reason to avoid AI tools, but it is a dimension that environmental nonprofits should understand, both for their own organizational decisions and for their advocacy work on technology policy.
The Energy Demand Reality
Training large AI models consumes substantial electricity, comparable to the lifetime energy use of several cars. Inference, running AI models in response to user queries, also consumes significant power at scale. A single complex query to a large model uses measurably more energy than a standard web search, and AI query volumes are growing rapidly.
- Data center electricity demand growing faster than renewable additions in many regions
- Water consumption for data center cooling is significant and often in water-stressed areas
- Hardware manufacturing involves extraction of rare materials with environmental costs
Mitigating Factors and Trends
The energy picture is not static. Smaller, more efficient models are delivering comparable performance to larger models at a fraction of the energy cost. The trend toward edge AI, running models on local devices rather than data centers, reduces transmission losses and can run on locally generated renewable power. AI-optimized chips are becoming significantly more energy-efficient with each generation.
- Model efficiency improving faster than energy demand is rising
- Major providers making meaningful renewable energy commitments
- Small, task-specific models increasingly viable for many nonprofit uses
For environmental nonprofits choosing which AI tools to adopt, the energy dimension argues for preferring purpose-built, efficient tools over large general-purpose models where both will accomplish the task. It argues for considering which AI providers have credible renewable energy commitments. And it argues against using AI simply because it is available, reserving AI capabilities for the tasks where they genuinely add value relative to simpler alternatives.
This is also an emerging advocacy dimension for climate organizations. As AI infrastructure becomes a significant and growing contributor to electricity demand, questions about where data centers are sited, how they are powered, and what standards govern their resource use are legitimate policy issues for the environmental sector. Several state-level policy conversations about data center regulation are already underway in 2026, and environmental nonprofits with technology policy capacity are well-positioned to engage.
Where AI Is Actually Delivering Environmental Results Now
Setting aside projected potential and generative AI greenwashing, a meaningful body of evidence documents AI applications that are producing measurable environmental outcomes today. These applications deserve particular attention from environmental nonprofits both as models for their own programs and as evidence for advocacy about AI's genuine climate role.
Emissions Monitoring and Attribution
Climate TRACE, co-founded by WattTime, uses AI algorithms and satellite data to monitor emissions from over 352 million emitting facilities worldwide. This capability represents a genuine breakthrough: for the first time, it is possible to attribute emissions to specific facilities, track changes over time, and identify the highest-emitting sources with independent verification. Environmental nonprofits working on industrial accountability, regulatory compliance, or climate finance can use Climate TRACE data to ground advocacy in facility-level evidence that was previously unavailable.
The availability of this data also changes what is possible in climate litigation, green procurement, and supply chain transparency work. When emissions data can be tied to specific facilities and companies, accountability campaigns can be more specific and harder to rebut with self-reported figures.
Illegal Deforestation and Habitat Monitoring
Rainforest Connection (RFCx) deploys solar-powered "Guardian" devices in forest canopies that listen for chainsaw sounds and vehicle noise associated with illegal logging and poaching. AI models process the audio in real time and alert rangers within minutes. The organization has deployed devices in forests across multiple continents, and the system has demonstrably reduced illegal logging in covered areas. For conservation organizations working in forest protection, this represents a template for applying AI to enforcement and monitoring challenges that previously required constant human patrol.
Google's SpeciesNet, released as open source, enables similar AI-powered biodiversity monitoring from camera trap images at a scale that would be impossible through manual image review. Organizations already using camera traps for wildlife monitoring can adopt SpeciesNet to dramatically accelerate analysis and extend the reach of their monitoring programs.
Waste Reduction and Circular Economy
AI-powered waste identification is enabling more effective recycling and waste diversion. Greyparrot's AI systems have scanned billions of waste items, generating data that helps facilities improve sorting accuracy and material recovery. When more recyclable material is diverted from landfills and incinerators, the avoided emissions from producing virgin materials are real and measurable. For environmental nonprofits working on waste policy, circular economy advocacy, or zero waste programs, AI-powered waste characterization data provides evidentiary support for program design and policy advocacy.
Grid Integration and Clean Energy Optimization
DeepMind's AI work on data center cooling optimization reduced energy consumption for Google's data centers by a significant percentage, and the company has since applied similar techniques to grid optimization. While this represents a self-serving application (reducing the energy footprint of AI infrastructure), the approach has legitimate broader application. Grid operators in several countries are using AI to reduce renewable energy curtailment, the practice of dumping excess solar and wind power because the grid cannot absorb it, which directly increases the effective output of existing renewable capacity without building new generation.
What This Means for Environmental Nonprofits in 2026
Environmental nonprofits occupy a unique and important position in the AI and climate conversation. They are simultaneously potential users of AI tools for conservation and environmental protection, important voices in governance and accountability conversations about AI's own environmental footprint, and credible analysts who can help translate AI climate claims into actionable policy and program insights. Each of these roles deserves intentional attention.
Using AI Tools in Your Conservation and Climate Work
Environmental nonprofits should focus AI adoption on the applications with the strongest evidence base for mission impact. Biodiversity monitoring through camera trap AI, emissions tracking through satellite data analysis, deforestation detection through acoustic or imagery AI, and climate data analysis for program design all represent areas where AI delivers verifiable results for conservation missions.
For general organizational operations, environmental nonprofits can apply AI to the same tasks it helps any nonprofit with: grant writing and prospecting, donor communications, volunteer coordination, content creation, and meeting summaries. These applications have a smaller footprint than purpose-built environmental AI but deliver genuine efficiency gains. Consider selecting providers with credible renewable energy commitments and preferring lightweight models where they are sufficient for the task. This connects to broader AI strategy considerations for nonprofit leaders.
Engaging on AI Governance and Environmental Standards
The questions being decided now about how AI data centers are regulated, what energy and water standards apply, how AI climate claims are disclosed and verified, and who bears the environmental costs of AI infrastructure are questions that environmental nonprofits have standing to engage. Several states are considering data center siting and resource standards. Federal policy conversations about AI energy disclosure are underway. International climate bodies are beginning to incorporate AI energy demand into long-range projections.
Environmental organizations that develop AI energy and emissions literacy in their advocacy staff can contribute meaningfully to these conversations. The alternative, leaving the policy landscape to be shaped entirely by technology companies and utilities, risks outcomes that prioritize growth over environmental protection. This is an emerging advocacy opportunity that aligns directly with existing organizational missions. The kind of internal AI expertise your team develops will support both your operational AI use and your external engagement on AI governance.
Distinguishing Real Impact from AI Climate Washing
Environmental nonprofits are uniquely positioned to help donors, policymakers, and the public distinguish between AI applications that produce measurable environmental results and those that primarily generate favorable narratives for technology companies. This means developing and applying frameworks for evaluating AI climate claims that ask: Is the impact measurable and verified? Does it represent net emissions reduction after accounting for AI energy use? Is the scale of impact proportionate to the claims being made? Is the application purpose-built for the environmental goal, or is environmental impact a secondary claim attached to a general-purpose commercial product?
Organizations with research and communications capacity can publish this kind of analysis, contributing to a more accurate public understanding of where AI and climate intersect. This is meaningful educational work that serves the sector regardless of whether your organization directly deploys AI environmental tools.
Funding and Resources for Environmental AI Work
Environmental nonprofits interested in deploying AI tools for conservation or climate work are not starting from scratch in terms of available support. Several initiatives specifically fund and provide technical assistance for this work.
Climate Change AI
Climate Change AI is a nonprofit initiative that runs innovation grants for researchers and practitioners working at the intersection of machine learning and climate change. It maintains one of the most comprehensive resources on AI climate applications and convenes a global community of researchers and practitioners. Their annual report on "Tackling Climate Change with Machine Learning" is an essential reference for any environmental organization developing an AI strategy.
- Innovation grants for AI climate projects
- Comprehensive resource library on AI climate applications
Google AI for Nature Accelerator
Google's AI for Nature Accelerator provides technical mentorship, cloud computing credits, and access to Google's AI tools and expertise for environmental conservation organizations. Accepted organizations work directly with Google teams to develop AI applications for their specific conservation challenges. Applications typically focus on biodiversity monitoring, habitat mapping, and protected area management.
- Direct technical mentorship from Google AI teams
- Cloud computing credits and tool access
Bezos Earth Fund
The Bezos Earth Fund has specifically identified AI as a priority area within its broader climate and nature funding strategy. It has supported AI-powered conservation monitoring, climate modeling, and clean energy optimization projects, and its grantmaking in this area is expected to grow through the remainder of the decade as part of its commitment to deploy $10 billion for climate and nature.
- Large-scale funding for AI climate applications
- Supports both technology development and deployment
Tech to the Rescue
Tech to the Rescue connects environmental and social nonprofits with volunteer technology teams that build pro-bono AI and software solutions. For environmental organizations that have identified a specific AI application but lack the technical capacity to build it, this matching model can provide a path to implementation without large upfront cost. Multiple environmental nonprofits have used this pathway to deploy AI monitoring and analysis tools.
- Pro-bono technical capacity for nonprofits
- Specializes in organizations with limited tech capacity
A Measured but Genuine Opportunity
The LSE/Systemiq study's projection of 3.2 to 5.4 billion tonnes of annual emissions reductions from AI by 2035 represents a genuine opportunity assessment, not a guaranteed outcome. It documents what AI could contribute to climate action if deployed at scale in the right sectors with appropriate policy support. That is a meaningful finding, and it provides a legitimate basis for environmental organizations to take AI tools seriously as part of the climate response toolkit.
At the same time, the greenwashing research serves as a necessary corrective. The distance between potential and current impact is substantial, and it is not closing as fast as the most optimistic projections imply. For environmental nonprofits, this argues for engaging with AI selectively and evidentially, prioritizing purpose-built applications with demonstrated environmental results over general-purpose tools with speculative climate benefits.
The most important takeaway for the sector may be this: environmental nonprofits should not position themselves either as uncritical AI adopters or as reflexive skeptics. The technology is genuinely capable of contributing to climate solutions, and it is also genuinely capable of increasing emissions and resource consumption if deployed carelessly or governed poorly. Organizations that bring rigorous, evidence-based judgment to both dimensions, as users and as advocates, are better positioned to shape outcomes than those who default to either enthusiasm or dismissal.
As AI capabilities continue to develop and more deployment data accumulates, the picture will become clearer. Environmental nonprofits that engage with the issue now, building internal expertise on AI tools and AI governance, will be better positioned to evaluate new evidence and adapt their strategies as the technology and its climate implications evolve. That kind of ongoing strategic engagement with AI is not optional for organizations serious about climate impact over the next decade.
Ready to Build Your Nonprofit's AI Strategy?
One Hundred Nights helps nonprofits, including environmental organizations, develop AI strategies that align with mission and deliver measurable results. From tool selection to staff training, we provide the guidance your team needs.