Savior and disruptor: AI’s double-edged impact on clean energy

In 2022, AI systems were responsible for approximately 2% of global electricity use, with data centers alone consuming over 460 terawatt-hours. That number is expected to rise sharply, reaching up to 1,000 TWh by 2026. In the United States, AI and data centers accounted for 4% of electricity use in 2022, with projections indicating a rise to 6% by 2026 and possibly 12% by 2028 if high-demand scenarios persist. The training of advanced models such as GPT-3 or ChatGPT has already required up to 1,200 megawatt-hours of electricity per instance, equivalent to the annual usage of hundreds of homes.

Artificial intelligence may be accelerating the clean energy transition, but its rapidly growing electricity consumption threatens to undermine global climate goals, according to a new study published in Sustainability. Researchers at Gazi University found that while AI systems significantly contribute to renewable energy optimization, their expanding infrastructure and computing demands could soon outpace sustainability benefits unless urgent mitigation strategies are adopted.

The peer-reviewed study, titled "The Energy Hunger Paradox of Artificial Intelligence: End of Clean Energy or Magic Wand for Sustainability?" and authored by Hafize Nurgul Durmus Senyapar and Ramazan Bayindir, projects that AI could consume as much energy as Japan by 2026 and potentially triple that figure by 2030. The researchers described the current trajectory as an “energy hunger paradox,” where AI is simultaneously a solution and a burden for clean energy systems. The findings raise concerns about the environmental costs of training large-scale models and maintaining AI-intensive infrastructure like data centers, cloud services, and connected hardware.

In 2022, AI systems were responsible for approximately 2% of global electricity use, with data centers alone consuming over 460 terawatt-hours. That number is expected to rise sharply, reaching up to 1,000 TWh by 2026. In the United States, AI and data centers accounted for 4% of electricity use in 2022, with projections indicating a rise to 6% by 2026 and possibly 12% by 2028 if high-demand scenarios persist. The training of advanced models such as GPT-3 or ChatGPT has already required up to 1,200 megawatt-hours of electricity per instance, equivalent to the annual usage of hundreds of homes.

Beyond electricity consumption, the study highlights other environmental impacts, including water usage for cooling and increasing e-waste. The daily operation of AI systems like ChatGPT is estimated to require up to 500,000 kilowatt-hours and as much as two liters of clean water per 10–50 prompts for cooling purposes. Meanwhile, the short lifecycle of AI hardware and reliance on rare earth minerals exacerbate ecological strain, particularly in developing economies with limited access to sustainable disposal infrastructure.

Despite these risks, the study emphasizes that artificial intelligence remains a critical enabler of decarbonization efforts. AI technologies are currently improving grid resilience, enabling real-time energy balancing, and enhancing forecasting for wind and solar generation. According to the study, machine learning models have improved wind energy predictability by as much as 30%, reducing the need for fossil fuel-based backup systems. Predictive maintenance tools have also extended the life of renewable infrastructure by years and cut operational costs by up to 30%.

The economic benefits of green AI are also substantial. The report estimates that by 2030, AI integration into the energy, agriculture, and transport sectors could deliver up to $5.2 trillion in economic gains globally. However, the researchers caution that without parallel investment in energy-efficient AI systems and renewable-powered data centers, these gains could be offset by the sector’s carbon footprint.

The study draws attention to the misalignment between AI development and clean energy investment. Currently, only 20% of AI data centers operate on renewable energy. Most remain powered by grids reliant on fossil fuels, raising questions about the long-term compatibility of AI growth with the United Nations Sustainable Development Goals, particularly SDG 7 (Affordable and Clean Energy) and SDG 13 (Climate Action). The authors noted that over 60% of the world’s population resides in developing economies, yet those regions receive only 15% of global clean energy investments, further highlighting the equity gap.

AI’s dual role in energy systems introduces complex risks for the clean energy transition. On one hand, technologies like smart sensors, AI-assisted battery storage systems, and digital twins are streamlining operations, reducing emissions, and enhancing reliability. On the other, AI’s constant need for computing power intensifies pressure on existing grids, which are also tasked with supporting the growth of electric vehicles, smart homes, and industrial automation.

The authors propose a series of recommendations to reduce AI’s growing energy burden while preserving its sustainability contributions. These include deploying energy-efficient AI algorithms that use less computational power, such as zero-shot and one-shot learning models. Custom hardware solutions, such as Tensor Processing Units (TPUs) and neuromorphic chips, are also identified as key enablers of low-energy, high-performance AI processing. The study emphasizes aligning AI deployments with on-site renewable sources, such as wind, solar, or hydrogen, to minimize reliance on carbon-intensive grids.

Regulatory intervention is also cited as necessary to establish sustainable AI standards. The European Union’s AI Act, which includes environmental transparency requirements for high-risk AI systems, is identified as a potential model. Public-private partnerships and clean technology investments are seen as critical to expanding access to green AI infrastructure in underserved regions.

The study calls for interdisciplinary coordination among stakeholders - engineers, regulators, businesses, and civil society - to ensure AI development remains aligned with global climate targets. Without such collaboration, they warn, artificial intelligence could exacerbate energy inequality and environmental degradation, especially in regions already vulnerable to climate change.