AI could cut global emissions by billions of tonnes, study finds

Artificial intelligence has the potential to significantly reduce global greenhouse gas emissions over the coming decade, with savings that could outweigh its own carbon footprint, according to a study published in Nature. The research, Green and intelligent: the role of AI in the climate transition, examines the role AI could play in accelerating decarbonisation, focusing on three sectors that together produce nearly half of all emissions: power, food and mobility.

The paper estimates that by 2035, AI applications in these sectors could cut annual emissions by between 3.2 and 5.4 gigatonnes of CO₂ equivalent. This figure includes projected efficiency gains and technological breakthroughs, and takes account of emissions from AI’s own energy use, including the electricity demands of data centres.
The authors state that this represents not just incremental progress but the possibility of ‘systemic transformation’ with far reaching implications for economies and societies. They identify five ways in which AI can advance climate action: transforming complex systems such as electricity grids and transport networks, speeding up the discovery of new low carbon technologies, changing consumer behaviour, improving climate modelling and increasing resilience to climate impacts.

How could AI impact power, food and transport?

In the power sector, AI’s ability to manage vast amounts of real time data could optimise the operation of energy grids. This would help integrate variable renewable sources such as wind and solar more effectively, balancing supply and demand and reducing reliance on fossil fuel plants. The study estimates annual savings of 1.8 gigatonnes of CO₂ equivalent from this sector alone.

In food production, AI could accelerate the development and adoption of alternative proteins by improving taste, reducing costs and boosting consumer acceptance. These products could replace emissions heavy meat and dairy. The potential reductions here are the largest of the three focus areas — up to 3.0 gigatonnes annually in the most ambitious scenario.

In transport, the study points to the benefits of AI enabled shared mobility services, such as optimised routing and ride sharing, as well as advances in the design and management of electric vehicle batteries. These measures could avoid up to 0.6 gigatonnes of emissions a year.

The importance of deliberate deployment and governance

The authors note that while AI’s direct energy use will grow, the emissions from this expansion are small compared to its potential climate benefits if applied in targeted ways. They caution, however, that these outcomes are not guaranteed. Without deliberate policy action, market forces alone may fail to deliver the scale or direction of change needed.

For you

Be part of something bigger, join BCS, The Chartered Institute for IT.

They argue for active government involvement to ensure AI deployment is both equitable and sustainable. ‘Policymakers must create enabling conditions for AI deployment, provide financial incentives for research and development, and ensure that AI applications are directed toward public goods and high impact areas’, the paper urges.

The analysis highlights the importance of governance in avoiding risks such as exacerbating social inequalities or causing unintended environmental harm. If left unchecked, the expansion of AI infrastructure could lock in high emissions energy sources or lead to applications that favour private gain over public benefit.

How AI can be successfully embedded

One of the study’s underlying messages is that AI’s climate value will come from embedding it within wider systems of change, not from isolated technological fixes. For example, optimising a delivery fleet with AI will have limited effect unless combined with a shift to electric vehicles and renewable energy charging infrastructure. Similarly, AI enabled climate models will only influence outcomes if they are integrated into planning, investment and disaster response systems.
The authors also point out that AI can help adapt to unavoidable climate impacts. For instance, predictive analytics could improve early warning systems for extreme weather, and machine learning could assist in managing water resources in drought prone regions. Such applications would not directly reduce emissions but would strengthen resilience and reduce vulnerability, especially in lower income countries.

The study is careful to frame its projections as potential rather than guaranteed outcomes. Its modelling assumes widespread adoption of best practice AI applications and cooperation between sectors and governments. Delays in deployment, inadequate infrastructure, or public resistance could all reduce the scale of benefits.

Nevertheless, the authors see a clear opportunity. If AI is deployed with purpose and guided by strong governance, it could be a powerful enabler in meeting the targets of the Paris Agreement and limiting global temperature rise.

The full open access article, as linked above, is available on Nature’s website.