Artificial intelligence isn’t just an efficiency play—it’s fast emerging as a climate tool with system-level impact. A new peer-reviewed study led by Lord Nicholas Stern estimates that strategic AI deployment across power, transport (light road vehicles), and food (meat & dairy) could abate 3.2–5.4 GtCO₂e per year by 2035, potentially roughly a quarter of those sectors’ emissions. Crucially, the authors find these savings still outweigh AI’s own energy footprint—if we scale the right use cases and support them with policy.

What the Stern study actually says

The analysis models three high-impact sectors and quantifies where AI moves the needle most:

• Power: AI improves grid management and raises solar/wind load factors (up to ~20%), yielding ~1.8 GtCO₂e annual reductions by 2035. 
• Food (Meat & Dairy): AI accelerates alternative proteins via discovery, cost curves, and product-market fit, producing ~0.9–1.6 GtCO₂e savings (up to ~3.0 GtCO₂e in a highly ambitious scenario). 
• Transport (Light Road Vehicles): AI enhances shared mobility efficiency and EV uptake/charging infrastructure planning, for ~0.5–0.6 GtCO₂e per year by 2035. 

All told, the paper’s 3.2–5.4 GtCO₂e range comes from just these three domains—which together represent about half of global emissions—so the total upside could be larger as additional sectors digitalise. The study also notes the 0.4–1.6 GtCO₂e added emissions from data centres/AI power use, which are still outweighed by the abatement potential modelled. 

Real-world signals: where AI is already cutting emissions

• Wind & grid optimisation: Google DeepMind’s forecasting/control tools boosted the value of wind energy and informed a product now piloted by ENGIE—showing how ML can make variable renewables more dispatchable. DeepMind’s control work also cut data-centre cooling energy substantially, a reminder that AI can shrink its own footprint, too. 
• EV charging + grid stability: From GM using AI to site chargers, to smart-charging programmes that shift load off-peak (California’s REDWDS initiative), to analyses showing AI can optimise charging to support the grid, deployment momentum is building. 

The catch: scaling requires policy, markets, and guardrails

The implementation gap is real. Even promising AI climate tools can stall without targeted incentives, procurement reform, open data, and grid-modernisation policy. The FT’s coverage of the Stern study underscores that while AI’s technical potential is large, policy support and financing decide whether it materialises—especially in lower-margin, public-good use cases. 

Where AI abatement will likely concentrate 

• Smarter, flexible grids: Short-term weather/power forecasting, dynamic demand response, and EV smart-charging/orchestration to absorb more wind/solar. 
• Protein transition R&D: Generative/physics-informed models to discover proteins and bioprocess routes that lower costs and improve taste/texture, accelerating consumer adoption. 
• Mobility efficiency: AI-assisted routeing, utilisation, and fleet management that cuts vehicle-kilometres travelled—not just electrifies them. 

Risks to watch

• Energy rebound: Rapid AI expansion increases data-centre demand; system-level energy planning is essential so abatement > AI load remains true. 
• Misaligned incentives: AI is also used in fossil workflows; climate-aligned deployments need standards, disclosures, and incentives to compete for capital. 

Sources

• Stern et al., “Green and intelligent: the role of AI in the climate transition,” npj Climate Action (Nature), 23 June 2025. Nature
• Grantham Research Institute (LSE) press release summarising the study. LSE
• Financial Times analysis on policy, incentives, and real-world scaling. Financial Times
• DeepMind case studies on wind/value optimisation and cooling efficiency. Google DeepMind
• Evidence of AI-enabled EV grid management (GM siting; smart-charging programmes; WEF energy-paradox brief). GM News Utility Dive Reuters World Economic Forum Reports