Partly. AI has real environmental costs.

There is no single yes or no answer, because the effect depends on scale, energy source and use. AI's environmental costs are real: electricity demand, associated carbon emissions, water use and hardware manufacturing. They are currently a low single-digit percentage of global electricity use, but they are growing fast and are concentrated locally. Potential benefits, such as efficiency gains and climate modelling, are possible but not well quantified. Whether this is 'bad' overall also involves weighing costs against benefits, which is partly a values judgment. Projections are uncertain and the figures above may change as new data comes out.

• The International Energy Agency (IEA) estimated data centres used about 415 TWh in 2024, roughly 1.5% of global electricity. AI is only part of that.
• The IEA projected data-centre demand could roughly double to about 945 TWh by 2030, with AI the main driver.
• Data centres also use water for cooling and for generating their power, and can strain local grids and water supplies.
• Training and running large models and building chips also cause emissions and use materials. Impact per query varies widely by model and by the grid's energy mix.
• AI could also cut emissions by improving efficiency in grids, buildings, logistics and research, but such gains are uncertain and unevenly documented.

The environmental costs are measurable, not merely hypothetical. However, AI is a broad collection of technologies rather than one activity with a single footprint: emissions depend on the electricity supply, while water and hardware impacts depend on infrastructure and location. Evidence supports both actual costs and useful environmental applications, but does not establish one worldwide net balance for all AI. Calling every use harmful—or assuming future benefits cancel existing costs—goes beyond what the evidence shows.

• Electricity consumption by AI-focused data centres increased by about 50% in 2025, according to the International Energy Agency.
• All data centres—not AI alone—used about 485 terawatt-hours of electricity in 2025; the IEA projects about 950 terawatt-hours in 2030, roughly 3% of global electricity demand.
• AI infrastructure uses water for cooling and requires hardware whose manufacture, mineral supply chains and disposal have environmental impacts.
• AI is already used to detect methane emissions and can improve energy efficiency in buildings, transport and industry.
• Potential environmental savings are not guaranteed: wider consumption can offset efficiency improvements, and beneficial applications may not be widely adopted.

1. iea.org
2. iea.org
3. iea.org
4. unep.org
5. iea.org
6. link.springer.com
7. iea.blob.core.windows.net
8. unep.org

Artificial intelligence creates undeniable environmental pressures through its physical infrastructure. Training and deploying advanced machine learning models demands enormous computational power, driving up electricity demand, greenhouse gas emissions from fossil-fueled grids, cooling water consumption, and hardware turnover. Conversely, AI can act as an environmental tool by optimizing power grids, detecting deforestation, improving supply chain efficiency, and accelerating materials science for clean energy. Because AI drives resource-heavy resource consumption while simultaneously offering tools to reduce emissions elsewhere, its overall impact cannot be described simply as good or bad.

• Data centers, including those running AI, consume an estimated 1% to 2% of global electricity, a share that is growing.
• Cooling data centers and generating their power consumes substantial volumes of water, sometimes in water-stressed areas.
• Hardware manufacturing for AI accelerators requires rare earth minerals and generates substantial electronic waste.
• AI tools are actively used to optimize electrical grids, improve renewable energy forecasting, and reduce industrial emissions.
• Whether AI's net environmental impact is positive or negative depends on clean-energy adoption and which applications are prioritized.

The claim is not a simple yes/no. The physical footprint of AI is well documented: electricity for training and, more importantly, for day-to-day inference, plus water for cooling and embodied emissions in chips and servers. That footprint is growing fast, and on today's grids most of that electricity still comes from fossil fuels, so AI adds to emissions and water stress. Against that, AI can reduce energy and emissions in other sectors, and efficiency per query is improving quickly. Because the net sign depends on the grid, the workload and the counterfactual, the honest answer is 'partly' rather than a flat yes or no — with the evidence on the sector-wide totals strong, and the evidence on the net global effect mixed.

• The IEA estimates data centres used about 460 TWh of electricity in 2022 — roughly 1.5–2% of global electricity demand — and projects around 945 TWh by 2030.
• The IEA has put data centres at roughly 0.6% of global greenhouse-gas emissions on a direct-energy basis; broader estimates that add hardware manufacturing reach several percent.
• Training a single large model (GPT-3) was estimated at about 1,287 MWh of electricity and about 552 tonnes of CO2e (Luccioni et al., 2022); over a model's life, inference from millions of queries can exceed training energy.
• Data centres also withdraw water for cooling: Google reported around 5.6 billion gallons (about 21 million cubic metres) of water withdrawal in 2023, roughly 17% more than in 2022.
• AI is also applied to climate-relevant problems — weather and extreme-event forecasting, data-centre and grid cooling optimisation, battery and materials discovery — but published estimates of the emissions these applications save vary widely and are not settled.

The environmental impact of AI is well-documented and includes significant energy use, greenhouse gas emissions, water consumption, and electronic waste. Data centers supporting AI are among the fastest-growing consumers of electricity globally, and their cooling systems strain local water supplies. The production and disposal of AI hardware also contribute to environmental degradation. However, AI is not solely harmful to the Earth; it is also used to improve energy efficiency, optimize industrial processes, and advance climate science. The overall impact is a mix of costs and benefits, making the answer 'partly' rather than a simple 'yes' or 'no.'

• AI data centers are projected to consume up to 1,050 terawatt-hours of electricity by 2026, comparable to the energy use of major countries like Japan or Russia.
• AI's carbon footprint comes from electricity use for training, inference, and cooling, as well as hardware manufacturing and mining for critical minerals.
• AI data centers can strain local water supplies and create 'heat islands,' raising temperatures in surrounding areas.
• AI contributes to electronic waste, with hardware lifecycles shortening due to rapid technological advances.
• AI is used to optimize energy grids, improve climate modeling, and reduce waste in industries, potentially offsetting some environmental costs.

1. Environmental impact of artificial intelligence en.wikipedia.org
2. AI’s environmental costs threaten water, land and climate | UN News news.un.org
3. Artificial intelligence controversies en.wikipedia.org
4. Text - S.4727 - 119th Congress (2025-2026): Artificial Intelligence Environmental Impacts Act of 2026 | Congress.gov | Library of Congress congress.gov
5. Artificial intelligence en.wikipedia.org
6. Environmental Cost of Artificial Intelligence: Carbon, Water, and Land Footprints | United Nations University unu.edu
7. Environmental impact of AI - Wikipedia en.wikipedia.org
8. AI has an environmental problem. Here’s what the world can do about that. unep.org