Can AI Help or Harm the Climate? New Evidence on Energy Use and Policy

Short summary

This explainer from The Conversation assesses whether AI will worsen or improve the climate by examining AI energy use, the efficiency of AI tasks, and the potential for AI to aid climate policy and disaster planning. It highlights rising data-center power demand, Jevons’s paradox, and emerging governance gaps between climate treaties and AI regulation. The piece also discusses a public AI forecasting system used for risk assessment and climate adaptation, and it closes with a view that governance decisions will largely determine AI’s ultimate impact on energy use and climate outcomes.

• AI energy demand is growing rapidly, driven by data centers and model training.
• Per-task energy is becoming more efficient, but overall demand can still rise due to Jevons’s paradox.
• AI can improve climate forecasting and risk assessment, potentially using less energy than traditional methods in some systems.
• Governance and incentives around AI energy efficiency will largely determine climate outcomes.

Introduction

The article, published by The Conversation, examines a central question in climate policy: will AI be a villain or a help in addressing climate change? It starts with AI’s energy footprint, noting that large AI models are trained and run in data centres that consume substantial electricity. A typical AI-focused data centre can power as many as 100,000 households, underscoring why energy use sits at the heart of debates about AI and climate change.

Data centre use rises and energy intensity

According to the International Energy Agency (IEA), energy used to train large language models (LLMs) and to run them has risen roughly tenfold since ChatGPT’s 2022 launch. Future AI data centres could consume far more electricity, perhaps up to 20 times today’s usage. The piece also notes that AI use has grown rapidly, with Google’s leadership noting a sevenfold increase in tokens processed since last year. At the same time, energy use per AI task is falling as efficiency improves, illustrating the classic energy-economy dynamic: efficiency gains can be offset by rising demand, a manifestation of Jevons’s paradox.

IEA estimates place AI electricity consumption at around 155 terawatt-hours per year, accounting for about one-third of total data-centre electricity use. Even so, this represents roughly 0.5% of the world’s electricity and about 0.1% of total energy use when considering broader energy inputs. The article also mentions that data centres tend to cluster and strain local grids in certain regions, highlighting local infrastructure challenges that accompany AI growth.

Can AI help with climate change?

The piece shifts from energy consumption to AI’s potential climate benefits. In 2025, the European Centre for Medium-Range Weather Forecasts deployed an AI-enhanced model called AIFS that improves probabilistic forecasts of extreme weather with greater geographical detail. This improves insurers’ risk assessment and government planning for disasters, while reportedly using less energy than traditional physics-based forecasts. The broader aim is to use AI as a decision-support tool for climate adaptation and mitigation, extending beyond forecasting to guidance on agricultural choices and energy transition strategies.

Beyond forecasting, the article discusses IPCC’s mission to inform policy for governments and the potential for a public AI model to reduce biases if used appropriately. A well-governed AI system could augment climate policy by describing risks and outlining actionable steps for adaptation and mitigation, thus turning scientific knowledge into practical decisions at multiple governance levels.

Governance gaps and incentives

The contrast between climate governance and AI governance is stark. The COP framework provides a formal, albeit imperfect, governance mechanism for climate change, while AI governance has relied largely on voluntary codes and sporadic summits. The author argues that incentives are needed to push AI toward energy efficiency and better governance outcomes, including more robust institutions that can coordinate global policy with AI deployment across sectors.

Could AI’s energy use be less than feared?

A Vision thinktank paper is highlighted as a means to balance the ledger. The paper emphasizes that AI’s energy appetite could be smaller than feared and that its potential for positive climate impact could be larger, but cautions that neither outcome is guaranteed without effective governance. The article concludes that the ultimate climate impact of AI may hinge more on institutions than on the chips themselves, reinforcing the central message: governance will determine whether AI accelerates or mitigates climate change.

Takeaway

In short, AI’s climate impact is not predetermined. While energy demand is rising, technological efficiencies are improving; AI can improve climate forecasting and adaptation; and strong governance will be essential to realize AI’s potential for positive climate outcomes.