AI and Climate Change: Friend, Foe, or the Future of Adaptation?
Andy Paterson • December 1st, 2025.
When we talk about AI and climate change, most likely, people think of the energy and emissions generated by large language models like ChatGPT and Claude. But is AI’s contribution to climate change overblown, or is it accurate? Can the potential application of machine learning to mitigate and adapt to climate change offset any harm?
AI will clearly increase global energy consumption, but will its ability to optimize energy use and help companies adapt make it a net benefit to climate action? That’s the questions we will answer here.
Key Takeaways:
AI does have a substantial environmental footprint, but today it’s still smaller than most digital activities like streaming or gaming.
AI’s climate benefits, from energy savings to extreme-weather forecasting, will likely far outweigh its emissions impact.
Purpose-built, efficient AI (like ClimateAi’s) can deliver climate gains without the massive compute demand of LLMs.
Measuring AI’s “climate ROI” is essential: avoided losses, improved resilience, and reduced emissions will likely matter more than the energy use they displace, but it’s essential that companies measure that.
Want a deeper dive on AI and climate change? In this podcast, our Chief Operating Officer, Will Kletter, unpacks AI’s climate trade-offs and why adaptation AI is fast becoming a major ROI driver..
The climate costs of AI are measured in the data centres that run computations for each prompt and train AI models. Data centres are a massive user of resources:
Energy use: Around 1.5% of global energy use went to data centres in 2024, and that number is set to double to 3% by 2030, primarily due to the expansion of AI systems.
Emissions: While some of the increased energy for AI data centres will come from low-carbon sources, much of it will come from fossil fuels. Currently, only around 0.5% of global emissions come from AI. However, that could expand to 1.4% by 2030 under certain growth scenarios.
Water use: Cooling requirements for GPUs can consume millions of liters of freshwater daily. As water needs grow and climate change worsens drought, water security could become an issue, especially in areas that need it most.
The chips for AI also create other environmental and social problems from mining and manufacturing the rare earth materials required to make the chips, to the e-waste at the end of the chips’ life.
AI and Energy Use
While it is true that AI computations use more energy than non-AI activities, for example, a prompt on ChatGPT can be 10x more energy-intensive than a search result on Google. However, more efficient models, like DeepSeek, can be 90% less energy-intensive than others.
And when put into perspective with other global computing power, like gaming or streaming, AI currently does not come close. A prompt to a chatbot uses ~0.2 watt-hours (Wh), roughly the equivalent of charging an iPhone to 16%. Compare that to:
One hour of video streaming on YouTube or Netflix would be ~80Wh
One 30-minute Zoom call between four people would be ~36 Wh
All of this is not to say that AI energy use and emissions are negligible. They should be taken seriously, and this is the lowest they will ever be. Some estimates say AI will take up half of data centre computational power by 2028. However, it is good to put AI’s energy use today in perspective with other online activities that require data centre computations.
How AI Can Help Climate Change
Despite the impacts of AI, its potential to mitigate and adapt to climate change is already taking shape. In the near future, it will enable more efficient energy use to reduce emissions, and assess and mitigate climate risks, which will likely offset any contributions.
Energy Efficiency: AI’s ability to automate grid decisions, control building temperatures, forecast renewable energy, and optimize transport routes could save substantial energy across all sectors. Building energy optimizations, for example, could reduce energy consumption in the built environment by up to 30%.
Emissions: With all of these efficiencies and other decarbonization opportunities. One recent study claimed that AI could reduce CO₂ emissions by 5.4 billion tonnes by 2035, around 13% of 2024 emissions.
Adaptation:AI for climate adaptation enables better forecasting, with implications for modelling how changing weather affects crops, supply chains, and investments. AI-driven early warning systems for extreme weather events, for example, can help avoid damage by up to 30% with just an additional 24 hours of warning.
These estimates could change drastically depending on the speed of adoption, the use of renewables at data centres, and the development of more efficient chips and data centres. Second-order impacts could, for example, increase energy use through more goods being transported after AI makes it more efficient.
The Net Impact of AI on the Climate
AI’s potential net impact on climate change is measured by more than the emissions it reduces. There are also the assets it can save from climate risks, the number of days a store stays operational during a weather disaster, crop yield increases, and ROI.
De-risked Supply Chains: Global supply chains are highly exposed to climate risks. AI can give companies a better understanding of when extreme weather events will occur. For example, a global roofing company adapted its supply chain logistics when ClimateAi helped them predict a hurricane well before it hit. This helped them secure vital supplies, increasing their market share by $15 million.
Faster Research and Development Cycles: AI helps companies accelerate R&D cycles, enabling them to bring better, more efficient, and resilient products to market more quickly. Using ClimateAi’s models, an agricultural company was able to research new regions for growing crops in just hours, rather than years, at 10% of the cost.
Researching Consumer Behavior Under Climate Change: ClimateAi recently used our AI models to understand how climate trends and extreme weather events affect consumer spending. We found that companies with better adaptation strategies outperformed competitors by more than 40% during an extreme weather event, as they stayed open longer and built a brand reputation around resilience.
How ClimateAi Maximizes the Positive Side
We ensure ClimateAi’s models are a friend of climate action rather than a foe through two main ways. First, we make sure our models are appropriately sized for the problem. Secondly, through the avoided emissions, our models provide.
Purpose-built models instead of oversized LLMs
We start with the problem and design domain-specific models (weather, crop, supply chain, consumer behavior) rather than running giant general-purpose LLMs.
Smaller models mean lower compute
Lower compute means lower energy use and emissions
We also validate our models using hindcasting, which involves evaluating them without massively energy-intensive training.
Operational efficiencies and avoided losses
At ClimateAi, our forecasts and adaptation strategies enable clients to reduce their high-emission fertilizer inputs, optimize supply chains, and prevent crop losses.
Crop loss prevention: Protecting yields means less replanting, less land disturbance, and fewer inputs.
Input optimization: Fewer and more effective sprays of fertilizers, and optimized irrigation means lower emissions
Supply chain and logistics optimization: Knowing exactly when crops will be ready and how much enables better logistics timing, saving on trucking fuel. And avoiding climate risks in the supply chain helps prevent the need for emergency supplies and product loss.
How To Make AI Climate-Positive At Your Company
Companies are adopting AI at breakneck speed. To ensure AI’s climate benefits outweigh its footprint, they must:
Determine whether the problem needs AI: Businesses today want to apply AI to everything, thinking it will solve all climate problems. However, for some climate-related issues, AI is not necessary and would have a net negative climate impact.
Prioritize domain-specific, efficient models over large general-purpose ones: General-purpose LLMs will require many more parameters and therefore be much more energy and emissions-intensive than smaller bespoke models. Companies should look for models that solve their specific pain points.
Develop climate ROI metrics to quantify avoided losses and emissions: Develop a system to measure the avoided emissions from all efficiency gains and the costs of climate-related risks, and offset those against the emissions the AI requires to understand the cost-benefit.
AI’s power use and emissions have made the headlines, but, as we have shown here, its potential to mitigate and adapt to climate change will likely make it a friend of the climate.
Its ability to parse vast amounts of data and automate decisions makes it an ideal ally in creating more efficient systems and adapting to reduce emissions and adapt to climate change.
If you want to learn more about AI solutions for climate change and how our product can help businesses and governments adapt, schedule a demo.
AI and Climate Change FAQ
Yes, but modestly, and far less than most public narratives imply. AI accounts for about 0.5% of global emissions today, mostly from data center power consumption. With responsible design and renewable energy, this footprint can be kept low even as adoption grows.
AI models require:
Energy-intensive training cycles for large LLMs
GPUs/TPUs that consume significant electricity
Cooling systems for data centres that use large amounts of water
But purpose-built, smaller AI models (like ClimateAi’s) use far less energy and require no massive retraining cycles, making them much more sustainable.
Yes. One major study found AI could reduce 5.4 billion tonnes of CO₂ by 2035 through:
Smarter grids
Efficient buildings
Optimized transport
Industrial automation
Better energy forecasting
That’s roughly 13% of current global emissions.
AI improves the precision and speed of climate risk decisions. Examples:
Predicting drought, floods, and extreme heat
Modeling crop yield and pest risk
Rerouting supply chains before a disaster hits
Identifying which assets are most exposed
Adaptation AI can reduce climate-related losses by tens of billions annually
At ClimateAi we keep our AI-related emissions low by:
Providing clients with insights that avoid emissions and bring operational efficiencies
Using domain-specific, compact models (not huge LLMs)
Hindcasting instead of repeated high-energy training
Andy Paterson is a content creator and strategist at ClimateAi. Before joining the team, he was a content leader at various climate and sustainability start-ups and enterprises.
Andy has held writing, content strategy, and editing roles at BCG, Persefoni, and Good.Lab. He has helped build one of the industry’s most popular newsletters and regularly publishes environmental science articles with Research Publishing.
Ready to find out what risk-intelligence can do for your bottom line?
