We’ve all used artificial intelligence in one way or another, whether it's to finish that English assignment, ask about a math problem, or just ask everyday questions. AI is the fastest-growing industry in the world, outpacing any other. In fact, it is reported that over half of adults in the United States use chatbots such as ChatGPT, Gemini, or Claude, and 25% report using them daily.¹ Chatbots can complete tedious tasks in a matter of minutes, tasks that would’ve taken a human hours. However, this seemingly weightless transaction carries much greater costs under the illusion. AI models are software that have been trained on large amounts of data to recognise patterns, make decisions, and generate results based on the information they have been given. The whole process is run on large corporations’ servers, such as Google's or OpenAI's, which host these chatbots. So when a prompt goes into your computer and spits out an answer shortly after, in reality, that prompt is sent to a multi-billion-dollar server and returned with an answer straight to your computer in mere seconds. 

In the process of training and running these AI models, corporations use a great amount of electricity and water, and have recently been criticised for it. This begs the question: how much of these resources are they really using, and how does their usage compare to other industries?

The first step in the process is to train the AI models. The training of GPT-4 is reported to have cost over $100,000,000 alone, and while the cost of training for GPT-5, their newest model, has not been disclosed yet, it is speculated that it could be billions.² The overall energy consumption in the training process is calculated to have been 51,772,500 to 62,318,750 kilowatt-hours (kWh) of electricity.³ To put this into perspective, the average American household uses about 10,500 kWh a year, so this amount of energy could power about 5,000 homes for a whole year. With each new model introduced by these AI powerhouse companies, the number of parameters and the size of the data they are trained on continue to grow exponentially. Therefore, their energy consumption and training costs could grow at an uncontrollably fast pace. 

As for the process of prompting, these days, when you have a question, you face the choice between using AI and Google Search. AI has become better at personalising its answer to your exact needs, while Google gives you many different sources to look at. Both give an almost immediate answer, but what are the differences between the two energy-wise? Google search uses about 0.3-1.0 Watt-hours(Wh) per question, while GPT-4 uses about 3-5 Wh per question.⁴ This is almost a 10 times increase in energy usage, showing that even though the two methods may seem similar, they in fact are not. Overall, all data centres are consuming about 415 terrawatt-hours (TWh) of electricity a year, and are projected to consume 945 TWh by 2030, which would be greater than the total energy consumption of Germany and France combined and similar to the aviation industry annually. Out of this, AI will likely account for 20% of data centre energy consumption worldwide.⁵ All in all, an absurd amount of energy is used to power this revolutionary technology that is freely available at our fingertips, without most people actually knowing its costs.

Water usage in the context of AI has gained traction recently, but many do not know exactly how much water is actually being used. Data centres need water to help cool their machinery, and because of the size of many of these centres, a lot of water is used. By 2030, it is estimated that global water usage for AI will reach 450,000,000 gallons a day, equivalent to the daily water usage of 5,000,000 people.⁶ Looking from a macro perspective, however, AI’s water footprint is still a fraction of the global agriculture industry, which accounts for 70% of all freshwater withdrawals. The problem with these data centres lies in the fact that their water usage is hyper-focused, usually taking millions of gallons from one area. This can make areas around these centres drought-prone, if they were not already.⁷ Water usage is not the only environmental impact of AI, however; the immense amount of energy needed to power it results in a large carbon footprint that is widely underestimated. Data centres are responsible for around 1% of the world's greenhouse gas emissions, which may not sound like much until it is compared to the aviation industry. Air travel, often seen as a leader in greenhouse gas emissions, accounts for 2% worldwide, which means that AI has almost reached half of the carbon emissions of planes.⁸ AI’s energy usage is only on the rise, so accordingly we can expect the same of its carbon footprint until it rivals and surpasses the 2% set by aviation. Finally, as advancements are made, more centres will need to be constructed, and in fact, by the end of the decade, data centres are expected to reach a land mass of 14,500 square kilometres. 

AI is one of the hottest and fastest-growing industries in history,  and it seems like everyone wants a piece. It will continue to shape society for an indefinite period of time as the technology of the future. However, the costs hidden behind AI are usually larger than expected, and though this technology may seem unassuming, the repercussions are undoubtedly there. Now, if AI technologies are consuming a grandiose amount of energy, water, and land, and emitting carbon dioxide, is the technology really worth it? Large-scale solutions to these problems are yet to be discovered, but for now, it really makes you stop and think twice before typing in a prompt.

Notes

1. Jeffrey Gottfried et al., "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact," Pew Research Center, June 17, 2026, www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/. ↩
2. Will Knight, "OpenAI's CEO Says the Age of Giant AI Models Is Already Over," Wired, April 17, 2023, https://www.wired.com/story/openai-ceo-sam-altman-the-age-of-giant-ai-models-is-already-over/. ↩
3. Kasper Groes Albin Ludvigsen, "The Carbon Footprint of GPT-4," Towards Data Science (Medium), July 18, 2023, https://medium.com/data-science/the-carbon-footprint-of-gpt-4-d6c676eb21ae. ↩
4. Jason A. Riddell, "The Hidden Cost of Information: Google Search vs ChatGPT," Medium, July 13, 2025, https://jasonariddell.medium.com/the-hidden-cost-of-information-google-search-vs-chatgpt-31184cdf1582. ↩
5. Lauren Smart and Sam Hsu, "The AI-Energy Nexus Will Dictate AI's Future. Here's Why," World Economic Forum, December 1, 2025, https://www.weforum.org/stories/2025/12/ai-energy-nexus-ai-future/. ↩
6. Smart and Hsu, "The AI-Energy Nexus." ↩
7. Henry Throp, "Artificial Intelligence, Water Consumption and the Trillion-Radish Conundrum," Tech Policy Press, November 12, 2025, https://www.techpolicy.press/artificial-intelligence-water-consumption-and-the-trillionradish-conundrum-/. ↩
8. Lyudmila Chulinda, "Decarbonization with the Help of Artificial Intelligence—One of the Priorities of International Civil Aviation," Contemporary Issues in Artificial Intelligence 1 (2025), https://doi.org/10.69635/ciai.2025.8. ↩