Artificial intelligence (AI) promises advances in medicine, transportation, industry, and many other domains.
Yet beneath the breakthroughs, lies a growing tension: the energy consumption of AI systems is soaring, and this trend may stray far from the world’s climate and sustainability objectives.
This article examines how AI’s energy demands are misaligned with global environmental goals, what factors are driving the misalignment, what studies reveal, and what might be done to reconcile AI growth with decarbonization.
AI Huge Energy Input
One of the most striking findings about AI involves data center power demands. In 2022, worldwide electricity usage by data centers was already about 460 terawatt‑hours, which would place data centers among the top global electricity consumers if they were a country.
Generative AI workloads, such as large language models and image/speech generation, are driving a large portion of recent growth in data center energy consumption.
Training a model like GPT‑3, for example, was estimated to consume over 1,200 megawatt‑hours of electricity by itself, generating hundreds of tons of CO₂.
Increases aren’t just from training; each inference (using a model to make predictions or generate output) has its energy cost, and multiplied by millions or billions of uses per day, the aggregate becomes substantial.
It Clashes with Climate Goals
Climate goals generally rest on reducing greenhouse gas emissions, moving energy systems from fossil fuels to clean sources, limiting global warming (for instance, to 1.5 or 2 degrees Celsius), and preserving or restoring environmental quality (including water and air).
When AI’s energy demands surge, they can undermine those goals in multiple ways. In many places, electricity is still generated from coal, natural gas, or other carbon‑intensive sources. When AI causes rising electricity demand, absent clean power, this results in more emissions.
Even in regions pushing renewable energy, increasing demand could strain grids, force reliance on backup fossil fuel generation, or raise electricity costs, thus slowing transitions in other sectors.
Studies suggest that data center power needs could nearly double in some regions, or that AI might come to account for up to half of data center electricity usage in some projections.
Besides energy, AI’s environmental footprint includes water usage (for cooling, manufacturing), hardware production, and eventual e‑waste. These also strain environmental systems and resources.
AI Also Shows Some Promise
Despite the concerning trends, not all studies conclude that AI is wholly incompatible with environmental objectives.
In certain settings, AI can help improve efficiency, reduce waste, optimize routing and logistics, improve energy use in manufacturing, smart buildings, and other systems.
Some projects show reductions in emissions or energy consumption when AI is used for prediction, control, or optimization.
For example, studies in China show that applying AI in non‑heavy‑pollution industries, or state‑owned firms with energy goals, yielded modest but measurable reductions in energy consumption.
But even where AI brings efficiency, the overall increase in demand may still outweigh these savings unless efficiency gains are very large or sources of energy are clean.
What Needs to Change
To reconcile AI growth with global saving‑the‑environment goals, we need multi‑pronged action.
First, there must be stronger transparency: full disclosure of energy use, emissions, water use and other resource impacts, especially across the entire life cycle of AI systems (hardware production, deployment, retirement).
Second, energy sources for AI infrastructure must shift decisively toward renewables; if AI systems are powered largely by fossil fuels, environmental harm is baked in.
Third, AI research could prioritize efficiency: not just pushing for larger models or more data, but for smarter model selection, efficient inference, reducing waste. The “small is sufficient” approach holds promise.
Fourth, regulation or policy incentives may be required: governments could set standards for energy efficiency in data centers, provide incentives for green AI infrastructure, or limit energy usage for AI workloads in regions where grid capacity or environmental cost is especially high.
