Artificial Intelligence (AI) is often portrayed as a looming disruptor, raising fears about energy consumption, ethics and surveillance. But a new peer-reviewed study published in npj Climate Action suggests AI might just be one of the most powerful tools available to help avert climate breakdown.
According to the report, Green and Intelligent: The Role of AI in the Climate Transition, authored by researchers from the Grantham Research Institute at the London School of Economics and Systemiq, the targeted application of AI in just three sectors — power, transport and food — could cut global greenhouse gas emissions by 3.2 to 5.4 billion tonnes of CO₂ equivalent per year by 2035.
That’s roughly 7 to 10 percent of current global emissions. And crucially, the authors say these savings would outweigh the carbon cost of AI’s energy use, including from data centres powering generative models.
“What if we were able to use AI to accelerate the tipping points of key low-carbon technologies?” said Dr Mattia Romani, one of the authors, who spoke exclusively to Climate Solutions News. “What AI does is optimise complex systems, it helps you discover new materials, it nudges behaviours, it improves modelling — and these are all things which are at the core of the low-carbon transition.”
Five Key Ways AI Can Drive Climate Impact
The research identifies five key “impact areas” where AI can support mitigation, adaptation and resilience:
- Transforming complex systems such as electricity grids or shared mobility networks
- Accelerating technology discovery and resource efficiency
- Nudging consumer behaviour toward lower-carbon choices
- Improving modelling for climate systems and policy interventions
- Managing adaptation and resilience, including disaster forecasting
These aren’t abstract categories. They describe how AI is already beginning to change the rules, whether by helping Google’s FloodHub forecast disasters or DeepMind’s AlphaFold speeding up the design of alternative proteins.
“We just thought: let’s do the maths and figure out how big that contribution could be,” said Romani.
Prefer to listen to a podcast audio version of the interview with Dr Romani? Either listen here on Acast, or head to Spotify, Apple Podcasts or Amazon Music.
Sector Deep Dive: Power, Meat & Dairy and Light Vehicles
Instead of using top-down economic models, the researchers took a bottom-up approach. They analysed how AI could impact the adoption and efficiency of low-carbon solutions in three sectors, each selected for their emissions footprint and the quality of available data.
Power: smarter grids, better forecasting
AI could cut emissions in the power sector by up to 1.8 billion tonnes per year, mainly by optimising how renewable energy is integrated into grids. This includes forecasting demand more accurately and improving the load factor of solar and wind by up to 20 percent.
“AI is great at prediction,” said Romani. “It does the kind of forecasting that power systems need far better than traditional models.”
Food: alternative proteins at scale
In the food sector, AI can accelerate the adoption of alternative proteins, helping them taste better and become more affordable.
Romani explained that many meat products, like those used in fast food or pet food, could be easily substituted by protein alternatives enhanced by AI-designed textures and flavours.
“We’re not talking about replacing your Sunday roast,” he said. “It’s the meat in pizza toppings, ready meals and pet food where substitution is easier.”
The report models three scenarios. In the most ambitious, AI could lift alternative protein consumption to as much as 50 percent by 2035, reducing emissions by up to 3 billion tonnes per year.
Mobility: shared and electric
In transport, AI can reduce emissions by up to 0.6 billion tonnes annually through optimised shared mobility systems and increased electric vehicle (EV) uptake. For instance, AI can help discover better battery materials or optimise EV charging infrastructure location.
In all, the cumulative annual savings by 2035 could reach 5.4 billion tonnes — all from just three sectors. And, as Romani points out, this excludes potential spillover effects into other areas such as industry or buildings.
“We deliberately focused only on direct effects in three sectors,” he said. “But the reality is that improvements in one sector often cascade into others.”
AI’s Energy Use: A Justifiable Trade-off?
Of course, AI doesn’t come free in terms of emissions. Data centres use significant amounts of electricity, and AI workloads are growing. But the report estimates that total emissions from all AI-related activity, not just those used for decarbonisation, would be around 1 gigatonne per year by 2035.
Even under conservative assumptions, the carbon savings dwarf the costs.
“We took a cautious approach,” Romani explained. “We assumed no improvement in data centre efficiency and that all electricity would come from average grid power. But even then, the emissions savings far outweigh the increases.”
The International Energy Agency’s own estimate, released after the report’s modelling was complete, aligns with this figure, suggesting the study’s assumptions hold up.
What Needs to Happen Now?
Despite its potential, the report is clear that AI won’t deliver climate gains without direction.
“At the moment, we’re simply not there,” said Romani. “Tech companies, governments and energy providers need to come together to drive more intentional applications of AI for climate.”
While the private sector has a role, public leadership is essential.
“Governments must create enabling conditions,” Romani said. “They need to incentivise research, direct investment and ensure AI applications serve the public good.”
This includes addressing digital inequalities between the Global North and South. “Public investment in AI infrastructure and education in developing countries will be essential,” the report states.
Romani echoed this point, warning of deepening divides if access to AI remains unequal.
“Public intervention is particularly important in addressing the potential risks… such as the exacerbation of inequalities between the Global North and South.”
Not a Silver Bullet, But a Powerful Lever
The researchers are cautious about overclaiming. The analysis doesn’t model rebound effects, such as increased consumption from more efficient systems, nor does it capture dynamic economic changes.
But even with conservative assumptions, the report shows AI could be a net-positive force for climate, provided it’s used with care and intent.
“This is not a silver bullet,” Romani said. “But it’s a powerful lever. And one we are not yet using to its full potential.”
