Introduction

Artificial intelligence (AI) represents a double-edged sword in the global effort to address climate change. On one hand, it provides powerful tools for reducing greenhouse gas emissions and aiding climate adaptation. Conversely, the training and deployment of AI systems require significant energy resources, potentially exacerbating environmental pressures. This report examines the interaction between AI and climate change across several channels, including its direct environmental footprint, its contribution to mitigation and adaptation, and the rebound effects resulting from increased productivity and wider adoption.

Opportunities: AI as a tool for climate action

AI is an enabler of climate solutions across multiple domains. In the energy sector, AI algorithms facilitate the integration of renewable power and improve load balancing. For example, DeepMind’s forecasting increased the economic value of wind energy by approximately 20% through superior supply and demand predictions. AI-enabled ‘smart grids’ allow for real-time analysis to balance demand and minimise waste. Additionally, AI-driven control systems in buildings can optimise HVAC (heating, ventilation, and air conditioning) operations, reducing energy consumption by up to 73% without compromising comfort.

In transportation, Google’s AI-based eco-routing in Maps suggests fuel-saving routes, which has prevented over 1 million tonnes of CO2 emissions annually. Beyond mitigation, AI supports climate adaptation by enhancing weather forecasting accuracy. Systems like IceNet and FloodHub provide early warnings for sea ice changes and flood risks, enabling proactive disaster management. From an economic perspective, AI can deliver substantial productivity gains. Early evidence suggests task-level improvements ranging from 10% to 65%, with median estimates pointing to an increase in annual total factor productivity (TFP) growth of around 0.48 percentage points over a 10-year horizon.

Challenges: environmental costs, rebound effects and broader risks

The environmental footprint of AI is significant, particularly for large-scale models. Training a single large natural language processing model was found to emit roughly 284 tonnes of CO2. GPT-3 consumed 1,287 megawatt-hours of electricity during training, generating about 502 tonnes of CO2 emissions. Ongoing use, or inference, also accounts for a large share of consumption; Google estimates that 60% of its AI workload energy is associated with inference.

Data centres currently consume between 0.9% and 1.3% of global electricity, a figure projected to double by 2030. These gains in efficiency may be offset by the Jevons paradox or rebound effects, where lower costs and higher productivity stimulate greater demand for AI services and broader economic activity. Furthermore, the AI supply chain is highly concentrated, raising geopolitical concerns. Advanced economies and large tech firms are best positioned to capture benefits, potentially widening the divide with developing nations that face constraints in digital infrastructure and technical capabilities.

Effects of AI on climate change under two AI development scenarios

The report considers two stylised technological trajectories. Under the ‘AI copilot’ scenario, narrow AI systems complement human decision-making. Benefits and environmental costs are likely to build up gradually, with net impacts depending on whether computational efficiency gains outpace the growth in AI use. The second scenario involves the emergence of artificial general intelligence (AGI), where systems match or exceed human cognitive capabilities. While AGI could accelerate scientific breakthroughs in energy and materials science, it could also trigger a massive increase in resource consumption. Forecasts suggest global energy consumption five years after the emergence of AGI could reach 423,000 terawatt-hours, an increase of approximately 160% relative to 2024 levels.

Implications for central banks and policy responses

For central banks, the AI–climate nexus affects potential output, inflation dynamics, and financial stability. Productivity gains could expand supply capacity, while high electricity demand and grid constraints might exert upward pressure on energy prices and increase inflation volatility. Large-scale investments in data centres and energy infrastructure create financial exposures that could become risks if returns do not materialise or if carbon-intensive assets face lock-in during the transition.

Central banks are also using AI to fill climate data gaps. Projects such as Danu, Gaia, and Symbiosis utilise digital twins, natural language processing, and big data to assess physical risks, extract climate indicators from corporate disclosures, and map supply chain emissions. Policy responses are emerging to address these issues, focusing on transparency and standards. Initiatives like the EU Artificial Intelligence Act include provisions for monitoring resource use, while ‘Green AI’ research aims to reduce the computational resources required for model training.

Conclusions

AI is neither inherently a climate solution nor a problem; its impact depends on technological evolution, adoption rates, and the energy systems supporting it. The coming decade is critical for determining whether AI accelerates the transition to a sustainable economy or intensifies environmental pressures. Policy and governance frameworks will be essential in ensuring that the benefits of AI are realised without a commensurate increase in its environmental footprint.