Artificial Intelligence is generally considered a phenomenon of the digital world, but there is also an aspect of it that is directly related to the environment. A new 2025 research study claims that the water used to power AI systems could now exceed the total global consumption of bottled water. Additionally, carbon emissions from AI this year are estimated to be equivalent to that of a large city like New York. If these figures are even slightly accurate, then the increasing demand for AI poses a major threat to the environment.
How does AI use water?
You may have heard recently that AI is “drinking” water or that AI is depleting water resources. In fact, AI itself does not use water, but the machines and data centers that power it use a lot of water. That is why this issue is being discussed. When ChatGPT, Google, or any AI system works, it depends on large data centers. These data centers contain thousands of servers that constantly generate heat. To keep them cool, cooling systems are used, and this cooling process consumes millions of liters of water. In some places, this water goes directly to water cooling systems, while in others, water used to generate electricity also counts.
Training an AI model also requires a lot of energy. When large companies create new AI models, their servers run at full capacity for weeks or months. During this time, not only does electricity consumption increase, but a lot of water is also used to generate that electricity and cool the system. According to one estimate, some AI tasks indirectly use several milliliters of water for each question answered.
What the research says about the environmental impact of AI
This peer-reviewed study, titled “Data Center Carbon and Water Footprints and What This Could Mean for Artificial Intelligence,” was recently published. This research was led by Dutch researcher Alex de Vries-Gao. This study focused on the data centers that power AI systems and result in huge consumption of energy and water. The research also acknowledged that accurate data is difficult to obtain because companies do not differentiate between AI and non-AI workloads in their environmental reports.
How were the estimates made?
Due to the lack of direct data, the researchers took a different approach. They analyzed environmental reports, average emissions data, and water consumption data from data centers associated with big tech companies like Google, Meta, and Amazon. Based on this, they estimated how much electricity and water would be needed to handle the AI workload.
Carbon emissions equal to New York City, consumption of more water than bottled water
According to the study, carbon emissions from AI systems alone could range from approximately 32.6 million to 79.7 million tons of CO₂ in 2025. This quantity is considered equal to the annual carbon footprint of a large metropolis like New York City. The water consumption figures are even more shocking. AI-enabled data centers could use approximately 312 to 764 billion liters of water annually, which is more than the total bottled water used worldwide in a year. This clearly shows that AI is becoming a big problem not only for energy security but also for water security.
Daily use, not training, is the main reason
An important point of the research is that the biggest environmental impact is not from training AI models, but from their daily use, i.e. inference. When users ask questions, create images or videos, and digital assistants work around the clock, this places a heavy load on the data center. Millions of requests increase electricity and water consumption rapidly.
Better technology, but the effect is not decreasing
Surprisingly, despite efforts to make data centers more energy-efficient, the overall environmental impact is not decreasing. The reason for this is simple: the use of AI is growing so rapidly that all efforts to improve it fall short. Simply put, technology is getting better, but its use is increasing even faster.
It is no longer correct to consider AI as just software
Two main conclusions emerge from this study. First, AI should no longer be seen as just software. Just as environmental regulations apply to telecommunications, aviation, and heavy industry, the AI industry should be subject to the same level of scrutiny. The second important point is transparency.
The research says that if companies do not openly share energy and water consumption data related to AI workloads, it will be difficult to create effective policies. Without clear information, neither conservation nor proper planning for the future is possible. This study clearly shows that the future of AI will not only be determined by its intelligence, but also by its sustainability for the planet.
