While AI is projected to boost global GDP by 0.5% annually between 2025 and 2030, its environmental footprint is growing at an unsustainable rate, driven by skyrocketing electricity and water demands.
Every time a user prompts ChatGPT, this triggers a process that can use as much as 10 times the electricity of a standard Google search. Scaled to billions of queries daily, the cumulative energy demand is staggering.
A 2025 report by the International Monetary Fund (IMF) warns that artificial intelligence (AI) could triple global electricity consumption by 2030, matching the annual energy use of entire nations such as India and potentially derailing global climate goals.
AI’s expanding energy appetite
Training large language models such as GPT-3 is energy-intensive. A single training cycle consumes 1,287 megawatt-hours, enough to power 120 U.S. homes for a year, and emits 502 metric tonnes of CO₂ which is equivalent to driving a petrol-powered car for one million miles, according to 2021 research by Google and the University of California, Berkeley.
However, the majority of AI’s electricity consumption now comes from inference, the energy used during daily interactions. AI tools, particularly image generators, require significantly more power per query than traditional search engines.
Goldman Sachs estimates that demand for energy by data centers will surge by 165% by the end of the decade as a result of AI. According to estimates from the International Energy Agency (IEA), global electricity usage by data centers is expected to exceed 945 terawatt-hours by 2030, rising from 415 terawatt-hours in 2024.
See also: The impact of artificial intelligence on the environment | Experts’ Opinions
Nearly 80% of this surge is anticipated to come from the United States and China. In contrast, developing economies will contribute only 5%. In the U.S. alone, it is predicted that data centers will consume approximately 606 terawatt-hours (TWh) of electricity annually by 2030, up from 147 TWh in 2023, with AI being a key factor driving this expansion. This would be sufficient to power nearly 57 million homes.
Emissions on the rise
The unchecked growth of AI is expected to considerably increase greenhouse gas emissions. In the U.S., AI could push electricity prices up by 8.6% and increase emissions by 5.5%, according to the IMF report. Globally, AI-linked emissions alone could raise the climate burden by 1.2%.
By 2035, IEA forecasts emissions from data center electricity use could reach 300 million tonnes under a baseline scenario or as high as 500 million tonnes in a high-growth ‘lift-off’ model. This would exceed the total industrial emissions of many countries.
Despite pledges of carbon neutrality, major technology companies are falling short on these promises. In 2024, Microsoft and Google reported significant increases in emissions and water use, with Google’s emissions rising by 50% over five years, largely due to AI operations.
These developments come amid worsening climate conditions. In 2024, the planet recorded an entire year of temperatures above the 1.5°C threshold, breaching the limit set by the Paris Agreement to avoid catastrophic climate tipping points.
Water use: The hidden toll
While electricity consumption dominates the headlines, AI’s water footprint is equally alarming. Data centers depend on water-based cooling systems that consume approximately two liters of water or more per kilowatt-hour.
Microsoft’s water consumption rose from 6.4 million cubic meters in 2022 to 7.8 million cubic meters in 2023, marking a 23% year-over-year increase.
A recent estimate suggests that AI could be responsible for consuming up to 6.6 billion cubic meters of water annually by 2027, almost half of that used in the United Kingdom every year.
This surge comes as more than 1.8 billion people are expected to face severe water scarcity in 2025, according to the Food and Agriculture Organization.
A coal-powered future?
In Omaha, Nebraska, plans to decommission a 1950s coal-fired power plant have been delayed until at least 2026 due to surging demand from the nearby data centers operated by Google and Meta.
According to The Washington Post in 2023, Meta’s data center consumed nearly as much electricity as the entire coal-fired plant could produce, while Google’s energy usage surpassed even that.
This resurgence in coal reliance underscores a growing contradiction: technologies touted as the cornerstone of a green future are, in practice, deepening the world’s dependence on fossil fuels.
Resource extraction and e-waste
Beyond electricity and water, AI systems rely on graphics processing units (GPUs), which require resource-intensive mining and hazardous chemicals. In 2023, GPU shipments to data centers increased by 44%, further intensifying the sector’s environmental impact.
The extraction of rare earth minerals such as lithium, cobalt, and tantalum which are central to GPU production carries a steep environmental toll. These mining operations frequently cause deforestation, soil degradation, and water pollution, raising questions of environmental justice.
