Unveiling AI's Growing Energy Footprint and Its Climate Impact

Artificial intelligence (AI) has become a transformative force in online life, with hundreds of millions relying on AI-powered tools for tasks ranging from homework help to creative content generation. However, this revolution comes with a significant and often underappreciated energy cost. Recent comprehensive analyses reveal that while the energy consumed by individual AI queries—such as generating text, images, or videos—may seem small, the aggregate demand is rapidly escalating as AI adoption grows exponentially.

Data centers, the backbone of AI operations, have seen their electricity consumption double between 2017 and 2023 due to the energy-intensive hardware required for AI workloads. Currently, data centers consume about 4.4% of all US electricity, with AI-specific tasks accounting for a growing share. Projections indicate that by 2028, AI could use between 165 and 326 terawatt-hours annually—equivalent to powering 22% of US households. This surge is driven by the increasing complexity and personalization of AI models, including reasoning and autonomous agents.

The energy footprint of AI queries varies widely depending on model size, task complexity, and data center location. For example, smaller text models require tens of joules per response, while the largest models can consume thousands of joules. Image generation typically demands less energy than large text models, but video generation is far more energy-intensive, sometimes requiring millions of joules for just a few seconds of footage. These variations underscore the challenge of accurately measuring AI’s environmental impact.

Compounding the energy demand is the carbon intensity of the electricity powering data centers. Many centers rely on fossil fuels, including natural gas and coal, leading to emissions significantly higher than the US average. Although major tech companies are investing in nuclear and renewable energy sources, these transitions are slow, and current energy mixes contribute substantially to AI’s carbon footprint.

Transparency remains a critical issue. AI companies often withhold detailed data on energy use and emissions, citing trade secrets. This opacity hampers accurate assessments and planning for sustainable AI infrastructure. Researchers advocate for anonymized data sharing and greater disclosure to enable better forecasting and accountability.

Looking ahead, AI’s energy demands are poised to grow dramatically as models become more sophisticated and integrated into everyday applications. Autonomous AI agents, voice and video interactions, and personalized AI services will multiply energy consumption far beyond current levels. This trajectory necessitates urgent attention to energy sourcing, efficiency improvements, and policy frameworks to mitigate environmental impacts.

In conclusion, while AI’s individual queries may seem low-impact, the collective energy footprint is reshaping power grids and contributing to climate change. Addressing this challenge requires collaboration among AI developers, energy providers, policymakers, and researchers to ensure AI’s growth is sustainable and transparent. Understanding and managing AI’s energy use is essential for balancing technological innovation with environmental stewardship.