Artificial intelligence is simultaneously one of the most energy-intensive technologies to emerge in recent decades and one of the most promising tools for addressing climate change — but whether its net effect on the planet is positive or negative may ultimately come down to governance rather than hardware, according to new research from the Italian think tank Vision.
AI data centres are already significant energy consumers. A typical AI-focused facility uses as much electricity as 100,000 households, and the power consumed by training and running large language models has risen roughly tenfold since ChatGPT launched in November 2022, according to the International Energy Agency. AI now accounts for around 155 terawatt-hours of electricity annually — about a third of all data centre energy use worldwide — though that still represents only approximately 0.5% of global electricity consumption. The IEA projects that future AI data centres could consume as much as 20 times more power than today’s. Concentrated clusters of facilities in places like Ireland and northern Virginia are already straining local electricity grids.
Efficiency gains are real but may be self-defeating. Power consumption per AI task is falling, yet overall demand is growing faster than those gains can offset — an effect economists call Jevons’s paradox, where improved efficiency stimulates greater use and erases the energy savings. Google’s CEO Sundar Pichai recently noted that monthly usage of Google’s AI products has grown sevenfold in a year, reaching 3.2 quadrillion tokens processed per month.
On the other side of the ledger, AI is already demonstrating tangible climate benefits. In 2025, the European Centre for Medium-Range Weather Forecasts deployed an AI model called AIFS that outperforms traditional physics-based forecasting systems at predicting extreme weather events, with greater geographical detail, more lead time, and significantly lower energy consumption than conventional models. Researchers and policymakers see broader potential: AI systems could help assess disaster risk for insurers and governments, suggest adaptation strategies alongside mitigation options, and bring scientific guidance down to the level of individual farmers or city councils deciding whether to plant a different crop or insulate a building.
Vision’s new analysis concludes that AI’s energy appetite may prove smaller than the most alarming projections suggest, and its climate potential larger — but warns that neither outcome is guaranteed without deliberate policy action. The contrast in governance frameworks is stark: climate change has the UN’s Conference of the Parties, meeting almost annually since 1995, while AI governance remains largely a patchwork of voluntary codes, occasional summits, and a UN scientific panel established only last year. The think tank’s findings are set to inform a conference on global climate governance in Venice in October 2025. The core message is that whether AI ultimately warms or cools the planet will depend less on the technology itself than on the institutions built to manage it.