The Growing Environmental Footprint of AI: Understanding Both the Challenge and the Solutions
- The White Hatter
- Jun 28
- 5 min read
Just because artificial intelligence exists in the digital world does not mean it is environmentally invisible. Every prompt we type, every image we generate, and every AI-powered search we perform relies on physical infrastructure that consumes electricity and, in many cases, significant amounts of freshwater. As AI adoption continues to accelerate, researchers are beginning to ask an important question, “What is the environmental cost of our growing reliance on artificial intelligence?”
A 2025 study titled “The Carbon and Water Footprints of Data Centres and What This Could Mean for Artificial Intelligence” provides one of the more comprehensive estimates to date (1). The researchers concluded that:
“The carbon footprint of AI systems alone could be between 32.6 and 79.7 million tons of CO₂ emissions in 2025, while the water footprint could reach 312.5–764.6 billion litres.”
At first glance, those numbers are difficult to comprehend. However, when we translate them into everyday language, the implications become much clearer.
A carbon footprint refers to the amount of greenhouse gases released into the atmosphere because of the electricity required to train and operate AI systems. Artificial intelligence does not create carbon dioxide on its own. Rather, the computers that power AI require enormous amounts of electricity, and depending on how that electricity is generated, significant carbon emissions can result.
To appreciate the scale, 32.6 million metric tonnes of CO₂ is roughly equivalent to the annual emissions produced by about 7 million gasoline-powered passenger vehicles. At the upper estimate, 79.7 million metric tonnes is comparable to the emissions from approximately 17 million gasoline-powered vehicles operating for an entire year. Those estimates vary because researchers modelled several possible futures based on how quickly AI adoption expands, whether data centres transition toward renewable energy, and how much computing efficiency improves over time.
The study also highlights another resource that often receives far less attention, water.
Most people are surprised to learn that many “legacy” data centres rely on freshwater cooling systems. Thousands of servers operate around the clock, generating substantial heat that must be removed continuously to prevent equipment failure. Cooling that hardware requires significant volumes of water, particularly in facilities that use evaporative cooling technologies. The researchers estimated that AI systems could require between 312.5 and 764.6 billion litres of freshwater in 2025.
Those numbers are equally difficult to visualize. At the lower estimate, that amount of water would fill approximately 125,000 Olympic-sized swimming pools. At the upper estimate, it would fill roughly 306,000 Olympic-sized swimming pools.
Another comparison helps put this into perspective. The average Canadian household uses approximately 220 to 250 litres of water per person each day. At the upper estimate, 764.6 billion litres would represent well over 3 billion days of household water use for one Canadian.
To help put these numbers in further perspective, according to the U.S. Environmental Protection Agency, Americans use nearly 9 billion gallons of water each day for residential landscape irrigation, which works out to approximately 12.49 trillion litres every year (2). That is enough water to fill about 5 million Olympic-sized swimming pools, or roughly 16 to 40 times greater than the annual water footprint projected for AI in 2025 (312.5–764.6 billion litres) mentioned above.
It is important, however, not to misinterpret what these findings are saying. The researchers are not claiming that AI itself “creates” carbon dioxide or directly consumes water. Instead, AI relies on enormous computing infrastructure housed inside data centres. Those computers require electricity to operate. If that electricity comes from coal or natural gas generation, greenhouse gas emissions increase. At the same time, many of those facilities require cooling systems that consume freshwater to keep equipment operating safely.
The study also does not suggest these numbers are inevitable. Instead, they represent projections across multiple scenarios. The final environmental footprint will depend on several factors, including how widely AI is adopted, how computationally intensive future AI models become, improvements in hardware efficiency, where data centres are built, and how much renewable energy is used to power them.
This is where an important part of the conversation often gets overlooked. Much like many other industries, the data centre sector is actively investing in technologies designed to reduce both energy consumption and water use. Recognizing growing public concern, companies are exploring several approaches that can substantially reduce environmental impacts without sacrificing computing performance.
One of the most promising developments is the use of closed-loop cooling systems. Rather than continually drawing fresh water, these systems recycle water repeatedly through cooling towers and heat exchangers. Depending on the design, freshwater consumption can be reduced by as much as 70 percent or more. These types of closed-loop systems are also being used to heat communities (3).
Some facilities are also adopting free-air cooling systems, where naturally cold outside air is used to cool servers. This approach works particularly well in cooler climates, making parts of Canada especially attractive locations for future data centres.
Other operators are expanding the use of air-cooled systems, particularly in regions where electricity is plentiful but water resources are limited. While air cooling can increase electricity demands, it significantly reduces freshwater consumption.
Perhaps the most innovative approach is immersion cooling systems. Instead of cooling servers with water or circulating air, computer components are submerged in specially engineered, non-conductive liquids that absorb heat far more efficiently. Although these systems require greater upfront investment, they dramatically reduce water consumption while improving energy efficiency and allowing higher-density computing.
Another important strategy involves changing how data centres are powered.
Electricity generation itself can consume tremendous amounts of water, particularly when produced by coal-fired or natural gas power plants that rely on steam-based cooling systems. By comparison, renewable energy sources such as solar and wind require little to no water during electricity generation. As more data centres transition toward renewable energy, both their carbon footprint and their indirect water footprint can decline significantly.
This highlights an important point that often gets lost whenever new technologies emerge.
Artificial intelligence undoubtedly presents environmental challenges that deserve careful attention. However, these challenges should not automatically lead us to conclude that AI development should stop. Rather, they emphasize the importance of continuing to invest in cleaner energy, more efficient computing hardware, and innovative cooling technologies that reduce resource consumption while allowing society to benefit from technological advances.
Here at The White Hatter, we believe this is another example of why critical thinking matters. It is easy to focus on dramatic headlines about AI’s environmental costs, just as it is easy to ignore them altogether. The evidence suggests that neither extreme provides the full picture.
Artificial intelligence does consume significant resources. That is a legitimate concern. At the same time, engineers, researchers, and industry leaders are actively developing solutions that are already reducing those impacts. As with so many conversations surrounding emerging technology, the most informed position lies somewhere between optimism and alarmism.
Rather than asking whether AI is “good” or “bad” for the environment, perhaps the better question is this:
“How do we ensure that the remarkable benefits AI can provide are developed responsibly, while continuing to reduce the environmental footprint that makes those benefits possible?”
Digital Food For Thought
The White Hatter
Facts Not Fear, Facts Not Emotions, Enlighten Not Frighten, Know Tech Not No Tech
Reference
