Due to the speed of Artificial Intelligence (AI) improvements, we as a society haven’t caught up with our abilities to measure and understand the tradeoffs of its use. The ease of utilizing generative AI and the lack of information regarding the environmental impacts of user actions means we, as consumers, don’t have much incentive to reduce our use.
AI consumes an enormous amount of water, electricity, and energy. The industry produces significant carbon dioxide emissions that exacerbate climate change and an increased amount of toxic technological waste. The creation of technology requires the extraction of rare earth minerals, which depletes natural resources and contributes to environmental degradation.
To train an AI model and adjust billions of parameters through repeated computations requires immense processing power. Thousands of graphic and tenor processing units (GPUs and TPUs), specialized electronic circuits that improve the speed of machine learning, run continuously for each training session. These sessions can take weeks or even months, consuming massive amounts of energy.
In a 2021 research paper, scientists from Google and the University of California, Berkeley estimated, the process to train a model like that of OpenAI’s GPT-3 alone consumes 1,287 megawatt hours of electricity, generating about 552 tons of carbon dioxide. That’s enough energy to power about 120 average U.S. homes for a year.
These generative AI models have short shelf-lives, resulting from high demand for new applications. Companies release new models frequently, and the energy used to train prior versions goes to waste. New models consume more energy for training, often possessing more parameters than their predecessors.
The short lifespan of the GPUs and other high-performance computing (HPC) hardware components used to train models results in a growing problem of electronic waste and further impacts the environment through their manufacture and transport. Manufacturing the necessary components requires rare earth minerals, the extraction of which depletes natural resources, and is often done in destructive and unethical ways. Damaged or obsolete hardware containing hazardous substances like mercury and lead is discarded. Infrastructure failures and software inefficiencies increase the strain, making AI training one of the most resource-intensive computing tasks in the modern era.
This training is done in large data centers, temperature-controlled warehouses that house computing infrastructure like servers, data storage drives, and network equipment. The AI boom fueled an increased pace of construction for these institutions, surging from 500,000 in 2012 to 8 million in 2025. Before the AI boom took off in 2023, the International Energy Agency estimated that data centers already accounted for 1–1.5% of global electricity use and around 1% of the world’s energy-related carbon dioxide (CO₂) emissions. Experts anticipate the AI percentage of total U.S. electricity use to jump 7-12% in the next three years.
A single AI word-based query uses approximately 0.3 watt-hours. OpenAI’s ChatGPT receives over two and a half billion prompts or requests per day. That’s about 850,000 Wh, which could power the average American home for a month.
However, the energy impact of an AI query is not an understandable value with a shared methodology for calculating. Countless variables beyond the user’s control, like the type and size of the model, the type of output generated, which energy grid is connected to the data center a request is sent to, or what time of day it’s processed, can make one query thousands of times more energy-intensive and emissions-producing than another.
Each time an individual uses ChatGPT, the computing hardware that performs those operations consumes energy. Researchers estimate a ChatGPT query consumes about five times more energy than a typical web search engine request.
To cool the hardware used for training, deploying, and fine-tuning generative AI models, data centers use copious amounts of water. In 2021, Google’s U.S. data centers alone consumed around 12.7 billion liters of fresh water. That’s almost 3.4 billion gallons. To compare, five gallons of water can sustain a four-person family for two days.
AI-related infrastructure could soon consume six times more water than Denmark, a country of six million, when a quarter of humanity already lacks access to clean water and sanitation. The data centers strain municipal water supplies and disrupt local ecosystems. Light pollution could disrupt dynamics between predator and prey species, and anthropogenic (man-made) sounds could affect reproduction in birds, and a newly identified species of fish in central Alabama faces extinction at the hands of a center. Individuals living near newly constructed data centers, notably Beverly Morris in Mansfield, Georgia (as reported by the British Broadcasting Corporation [BBC]), cannot drink the water because sediment runoff escapes into the water system.
Given the rapid growth of AI use, it’s likely that our AI footprint today is the smallest it will ever be. Look for sustainable options, perhaps those that utilize more efficient infrastructure, rainwater harvesting, smarter cooling systems, and results of long-term thinking. Think of your actions’ outcomes before asking ChatGPT to summarize something when you could read it yourself. Don’t let a machine think for you at the expense of our world. “Chat, what should I do?” You should use your brain instead of relinquishing your cognitive abilities. A Google search provides enough information to help with a homework answer. You don’t need to make idiotic AI-generated videos of your friends doing weird stuff. It’s not necessary. Water, air, and earth are. Understand what you’re doing when you use AI. Know your impact.
