BIS Paper Warns AI Climate Benefits Could Be Offset by Rising Data Centre Energy Demand

Artificial intelligence could accelerate climate action through more efficient energy systems, improved weather forecasting and advances in clean technology, but its growing electricity requirements may offset some of these benefits, according to a research paper published by the Bank for International Settlements (BIS) in October 2026. The paper, "Artificial Intelligence and Climate Change: Balancing Innovation and Sustainability," examines how AI's environmental impact will depend on the pace of technology adoption, improvements in computing efficiency and the carbon intensity of electricity used to power digital infrastructure.

Authored by Leonardo Gambacorta of the BIS and Salvatore Polizzi of the University of Palermo, the research highlights the growing tension between AI's contribution to climate solutions and the resources required to operate increasingly powerful computing systems. The authors examine two potential development pathways, one involving AI systems that primarily assist human activities and another involving artificial general intelligence (AGI). They find that both the environmental benefits and risks could increase as AI capabilities advance, with important consequences for energy systems and financial markets.

 

AI Could Improve Energy Efficiency and Climate Adaptation

 

The paper identifies several areas where AI could contribute to emissions reductions, particularly through electricity grid management, industrial energy efficiency and transportation optimisation. AI systems can improve renewable energy forecasting, predict equipment failures and help utilities balance electricity supply and demand. The research cites an earlier study showing that AI-based forecasting increased the economic value of wind energy by approximately 20% through better predictions of generation and market conditions.

AI also offers opportunities for climate adaptation. Advanced weather forecasting systems can support earlier warnings for floods, droughts and extreme storms, while satellite imagery and machine learning can help assess agricultural risks and monitor environmental changes. In the insurance sector, AI can improve assessments of physical climate risks, potentially helping insurers and property owners make more informed decisions about adaptation investments.

 

Read more: AI and Data Centres Could Help Make ASEAN Clean Energy Projects Bankable

 

Data Centre Electricity Demand Could Double by 2030

 

Despite these potential benefits, the computing infrastructure required to develop and deploy AI creates additional electricity demand. The paper cites International Energy Agency projections indicating that global data centre electricity consumption could roughly double by 2030, with AI workloads among the main drivers. It also references estimates suggesting that AI-related computing currently represents less than 0.2% of global electricity consumption, although this share could increase substantially as adoption expands.

The environmental footprint extends beyond training large AI models. Running deployed models, known as inference, can account for a considerable share of ongoing energy consumption as AI becomes integrated into consumer services and business operations. Data centres can also place pressure on electricity networks through higher peak demand and transmission congestion. The paper emphasises that the emissions associated with additional AI workloads depend on the electricity generation sources meeting that demand, which may remain carbon-intensive even in grids with significant renewable energy capacity.

 

Efficiency Gains May Not Deliver Equivalent Emissions Reductions

 

The researchers highlight rebound effects as a major uncertainty in assessing AI's environmental performance. Improvements in computing efficiency can reduce the energy required to complete individual tasks, but lower operating costs may encourage greater use of AI services. Similarly, AI-driven improvements in logistics and manufacturing could reduce emissions per unit of output while stimulating additional economic activity and resource consumption.

This phenomenon is associated with the Jevons paradox, where greater efficiency can contribute to higher overall resource consumption. The paper explains that AI's climate impact will depend on whether improvements in technology and energy efficiency can keep pace with expanding demand. It also identifies water consumption for data centre cooling and the extraction of critical minerals for computing hardware as environmental pressures that need to be considered alongside electricity use.

 

Explore OneStop ESG Marketplace: Climate risk & physical risk analytics

 

AI Development Could Create New Climate and Financial Risks

 

The paper considers two possible trajectories for AI development. Under an AI copilot scenario, specialised systems would increasingly support human decision-making, delivering gradual improvements in productivity, resource efficiency and climate-related services. Under a more transformative AGI scenario, advances in scientific discovery and climate technology could be substantially greater, but the demand for computing resources and associated energy infrastructure could also rise sharply.

These developments have implications for central banks and financial stability. AI-related productivity improvements could increase economic output and reduce certain production costs, while growing electricity demand and constraints on grid capacity could create upward pressure on energy prices. The paper also identifies potential financial risks from large investments in data centres and power infrastructure, particularly if expected returns from AI development fail to materialise. Investments in fossil fuel-based generation to meet AI electricity demand could introduce additional transition risks if climate policies become more restrictive.

 

Greater Transparency Needed on AI's Environmental Footprint

 

The research highlights the absence of a universally accepted framework for measuring and reporting emissions from AI training and inference. Without consistent standards, comparisons between AI models, technology companies and data centre operators remain difficult. The authors identify improved carbon accounting, energy efficiency reporting and independent environmental assessments as areas receiving increasing attention from policymakers and researchers.

The paper also discusses opportunities to align AI expansion with renewable energy development. Shifting flexible computing workloads towards periods when low-carbon electricity is available could reduce associated emissions, while investments in additional renewable generation and energy storage could help accommodate growing demand. However, the researchers caution that moving workloads based solely on electricity prices may not reduce emissions if the electricity supplied during cheaper periods remains carbon-intensive.

For policymakers and financial institutions, the distinction between AI's potential climate benefits and its measurable environmental footprint is becoming increasingly relevant. The BIS research identifies improved reporting, investment in cleaner electricity infrastructure and greater coordination between AI and climate policy as areas that could influence whether future technological advances contribute to emissions reductions or intensify pressure on energy systems.

 

Source: Bank for International Settlements