The Environmental Cost of AI
By Eli Ibanga
Edited By Victoria Sosa
A few years ago, I was a panelist for a discussion on the benefits of utilizing AI in support of emergency management operations. I proposed that AI could be used for analyzing financial and location data to better anticipate community needs in advance of natural disasters. While I don’t regret that take, I do think our collective jets should be cooled on AI. Within the last few years, there has been a noticeable push across industries to develop and implement AI, with key goals appearing to be an automation of tasks to decrease costs whilst developing new revenue streams for companies involved in the development and deployment of AI tools. Furthermore, the technology has become part of the average tech layman’s toolbox, with tools like ChatGPT being used across the world daily. While AI undoubtedly has immense potential for good, I believe it is currently being used irresponsibly, partly due to little oversight and regulation. There’s a myriad of potential issues such as biases in the training of algorithms, unethical AI deployment, and its sociological impact, but in the spirit of Earth Month I’d like to focus on the carbon footprint of AI.
I found statistics on AI tools’ climate impact were both underwhelming and jarring, depending on the context through which I considered the data. On one hand, AI’s energy consumption is only approximately 3% of Earth’s total emissions (Kemene et al., 2024). However, the typical ChatGPT inquiry uses approximately 10 times the energy used by a traditional digital inquiry method, like a Google search (Parshall, 2024). Generative AI uses around 33 times more energy to complete a task than “task specific” software (Kemene et al., 2024). Those figures are large, but they mean next to nothing without contextualization. For this, let us take a closer look at what is probably the most famous AI tool at the moment, ChatGPT.
ChatGPT produces around 4.32 grams of carbon dioxide per query, and receives around 50 million unique visits each day (Mittal, 2024). This works out to nearly 78,840 metric tons of CO2 production annually. Compared to the USA, that’s a drop in the bucket, as we generate nearly 18 million metric tons of CO2 per day (EPA, 2024). But what about the amount of power used? Many environmental activists and researchers have begun to realize that some “green” products may have a negative cumulative effect on the climate crisis. If an electric car is developed, or a wind turbine installed, we all feel good because it’s an environmentally conscious product. But what of the environmental cost? Is the supply chain environmentally sound? Are the net emissions to create and sustain such an item worth it? What is the human cost to produce these products? These are the questions we must consider when looking at emerging technologies to better assess their net impact on the environment.
Let’s examine ChatGPT’s climate impact by comparing its energy consumption and emissions to a country like Pakistan, a notable developing country with a large population. Pakistan ranks 5th for largest population worldwide, 32nd for CO2 emissions, and 26th for electricity consumption when compared to other countries and territories (Crippa, et al., 2024; Fulghum, 2024; United Nations, 2024). As ChatGPT is essentially a digital application, we can ignore supply chain emissions, as those are more attributable to smartphone and computer developers (just in case you are wondering though, there are a LOT of emissions there). ChatGPT generates 7.62 metric tons of CO2 in a year, compared to Pakistan’s annual CO2 emissions of 200.51 teragrams (Crippa, et al., 2024; Mittal, 2024). This pales even further when compared to a country like the USA’s 4,682.04 teragrams worth of CO2 emissions (Crippa, et al., 2024).
So, we’ve established that when compared to the biggest of polluters, AI is nowhere close to being one of the main culprits, but that’s only when considering CO2 emissions. ChatGPT annually uses 14.46B kWh, or 8.45% of the electricity that Pakistan uses over the same period, which is made more disturbing when we consider that Pakistan has the 5th largest population worldwide (Fulghum, 2024; United Nations, 2024; Wright 2025). It’s even more scary when you consider that the AI tool uses an equivalent amount of water per month to that of every person on Earth having two glasses of water (Wright, 2025).
Looking at some hopefully more digestible figures, when you drive just one mile in an average gasoline-powered vehicle you produce CO2 emissions equivalent to generating about 243 AI images (Heikkilä, 2023). Despite this, the AI tool could use approximately the same amount of electricity as it takes to charge your smartphone to generate just one of those images (Heikkilä, 2023). Within that context, the use of AI for trivial tasks seems irresponsible at best and unethical at worst (looking at you, AI photo enthusiasts). On a small scale, the aimless ChatGPT query is a drop in a bucket. But cumulative impact paints a grimmer picture.
So, it's clear that using AI won't send the Earth to a fiery grave, at least not on its own, but if AI remains unregulated then the public will remain uninformed on its true environmental impact. This ignorance could erode the very small gains the global community has made against climate change. When considering the environmental cost of AI, one should consider both the energy usage (how much energy will it use, and what else could that energy be used for) as well as its greenhouse emissions (what is the negative impact on the planet). Humanity has introduced a new energy consumption and emissions variable in AI, and it wasn’t created with the purpose of replacing a less sustainable system either. The disconnect between AI’s negative impact on our planet and its potential to aid our everyday lives must be rectified to guarantee that while the AI sector continues to grow, we prioritize Earth’s sustainability over convenience and profitability.
References:
Blue, M.-L. (2022, August 30). What is the carbon footprint of a plastic bottle?. Sciencing. https://www.sciencing.com/carbon-footprint-plastic-bottle-12307187/
Bureau of Labor Statistics (2025, January). Employed persons by detailed occupation, sex, race, and Hispanic or Latino ethnicity. U.S. Bureau of Labor Statistics. https://www.bls.gov/cps/cpsaat11.htm
Crippa, M., Guizzardi, D., Pagani, F., Banja, M., Muntean, M., Schaaf, E., Monforti-Ferrario, F., Becker, W., Quadrelli, R., Risquez Martin, A., Taghavi-Moharamli, P., Köykkä, J., Grassi, G., Rossi, S., Melo, J., Oom, D., Branco, A., San-Miguel, J., Manca, G., Pisoni, E., Vignati, E. and Pekar, F., GHG emissions of all world countries, Publications Office of the European Union, Luxembourg, 2024, doi:10.2760/4002897, JRC138862
Edwards, C., & Fry, J. M. (2011, July 25). Life cycle assessment of supermarket carrierbags: A review of the bags available in 2006. GOV.UK. https://www.gov.uk/government/publications/life-cycle-assessment-of-supermarket-carrierbags-a-review-of-the-bags-available-in-2006
Environmental Protection Agency. (2016, January). What If More People Bought Groceries Online Instead of Driving to a Store?. EPA. https://www.epa.gov/greenvehicles/what-if-more-people-bought-groceries-online-instead-driving-store#:~:text=What’s%20the%20big%20deal?,times%20the%20distance%20to%20Pluto!&text=All%20these%20car%20trips%20result,out%20of%20milk%20again.%E2%80%9D)
Environmental Protection Agency. (2024, April 11). Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2022 U.S. Environmental Protection Agency, EPA 430R-24004. https://www.epa.gov/ghgemissions/inventory-us-greenhouse-gas-emissions-and-sinks-1990-2022.
EUROCONTROL Business Cases (2024, June 18). Standard Inputs for Economic Analyses. EUROCONTROL. (10) 7-7.5. https://ansperformance.eu/economics/cba/standard-inputs/chapters/amount_of_emissions_released_by_fuel_burn.html
Fulghum, N. (2024, December 9). Yearly Electricity Data. Ember. https://ember-energy.org/data/yearly-electricity-data/
Heikkilä, M. (2023, December 1). Making an image with generative AI uses as much energy as charging your phone. MIT Technology Review. https://www.technologyreview.com/2023/12/01/1084189/making-an-image-with-generative-ai-uses-as-much-energy-as-charging-your-phone/
Kemene, E., Valkhof, B., & Tladi, T. (2024, July 22). AI and energy: Will AI reduce emissions or increase demand?. World Economic Forum. https://www.weforum.org/stories/2024/07/generative-ai-energy-emissions/#:~:text=How%20much%20energy%20does%20AI,just%20users%20on%20one%20platform.&text=How%20can%20business%20leaders%20strike,of%20the%20Annual%20Meeting%202025
Mittal, A. (2024, October 8). CHATGPT: How much does each query contribute to carbon emissions?. LinkedIn. https://www.linkedin.com/pulse/chatgpt-how-much-does-each-query-contribute-carbon-emissions-mittal-wjf8c/
Parshall, A. (2024, June 13). What do Google’s AI answers cost the environment? Scientific American. https://www.scientificamerican.com/article/what-do-googles-ai-answers-cost-the-environment/
Pearce, C. (2023, July 12). How many cars equal the CO2 emissions of one plane?. BBC Science Focus Magazine. https://www.sciencefocus.com/future-technology/how-many-cars-equal-the-co2-emissions-of-one-plane
United Nations (2024). World Population Prospects 2024: Summary of Results. UN DESA/POP/2024/TR/NO. 9. New York: United Nations.
U.S. Energy Information Administration. (2023, June). Frequently Asked Questions (FAQs) - U.S. Energy Information Administration (EIA). https://www.eia.gov/tools/faqs/faq.php?id=97&t=3
Wright, I. (2025, March 6). ChatGPT energy consumption visualized - beuk. Business Energy UK. https://www.businessenergyuk.com/knowledge-hub/chatgpt-energy-consumption-visualized/