Recent research has highlighted the significant environmental impact of AI models like ChatGPT.
The energy consumption of data centres powering these models contributes to greenhouse gas emissions and strains water resources. ChatGPT alone produces around 260,930 kilograms of CO2 monthly globally based on the energy consumption of data centres and the number of users interacting with ChatGPT worldwide. This is comparable to the emissions from 260 flights between New York City and London. Each query generates about 1.59 grams of CO2, equivalent to burning a single candle for an hour, or 1 standard inflated ballon.
A typical ChatGPT query uses nearly ten times more energy than an average Google search. The cooling of network servers for AI requires substantial fresh water.
Shaolei Ren, an associate professor at the University of California Riverside, estimates that GPT-4 produces a quarter to half a pound of carbon emissions to write a 100–250-word email. In 2020, data centres and data transmission networks accounted for 0.9% of energy-related GHG emissions and 0.6% of total GHG emissions globally, according to the International Energy Agency (IEA).
A study by Ren and colleagues predicts that by 2030, the public health impact of U.S. data centres alone could exceed $20 billion annually—double that of U.S. coal-based steelmaking and comparable to California’s on-road emissions. Training an AI model like Llama3.1 can produce air pollutants equivalent to over 10,000 round trips by car between Los Angeles and New York City.
A report from International Energy Agency (IEA) 2024 found that “electricity consumption from data centres, artificial intelligence and the cryptocurrency sector” could double from 2022 to 2026, “roughly equivalent to the electricity consumption of Japan.
Google and Microsoft released sustainability reports in 2024 that also reported significant increases in greenhouse gas emissions, in part due to AI.
Google’s report said its greenhouse gas emissions rose by 48% from 2019 to 2023 — “highlighting the challenge of reducing emissions while compute intensity increases and we grow our technical infrastructure investment to support this AI transition.” Microsoft reported its emissions rose 29.1% from 2020 to 2023 because it was investing “in infrastructure needed to advance new technologies.” In the introduction paragraph, the Microsoft report referred to “new technologies, including generative AI.”
According to the United Nations Environment Programme, AI systems “rapidly accelerate the speed and scale of warfare in terms of inflicting harm on both civilians and the environment.”
The United Nations Environment Programme warns that AI systems can accelerate environmental damage through conflicts involving AI-enabled warfare, which may harm critical resources like water supplies and natural reserves—exacerbating climate change.
AI can also incite war of inflicting harm on both civilians and the environment.
Conflicts involving AI-enabled warfare can result in unintended, large-scale damage or destruction to critical resources, such as water supplies, agricultural land and natural reserves, exacerbating the environmental toll of warfare. Destruction of natural reserves and land can also mean the destruction of natural carbon sinks that absorb carbon from the air, which can contribute to climate change.
While scientists are still debating AI’s overall environmental impact, Ren emphasizes the complexity, saying, “We can measure carbon emissions, but the benefit part—that depends on the usage scenario.”