Artificial Intelligence (AI) is one of the fastest growing technologies, with recent advancements resulting in wide scale adoption and integration into everyday tasks, like internet searches, email writing or fun image generation. While these advancements have created some useful tools, there are also many concerns associated with AI, including over-reliance on computers, data security, and potential replacement of jobs. Another major concern with the increased demand for AI technologies is the immense energy and water resources that it requires.

Energy Consumption  

The energy demands of AI computing is two-fold. First, a staggering amount of energy is needed to train many AI models, specifically generative AI models and with the increased demand for AI models, new models are trained every few weeks. Training an AI model requires a huge amount of data and computer processing, which leads to rapid fluctuations in energy use that occur over different phases of the training process. During this training period, AI data centers may consume seven to eight times more energy than typical computing. To accommodate the fluctuations in energy use, many power grid operators employ diesel-based generators to absorb those fluctuations and protect the grid.  

Once a generative AI model is trained, the energy demands don't disappear. Each time a model is used, such as asking ChatGPT to summarize an email, the computing hardware consumes energy. It has been estimated that a single ChatGPT query consumes about five to ten times more electricity than a simple web search. Because this energy consumption occurs offsite at these large data centers, many users don’t directly experience this energy use.   

Image

Water Consumption  

In addition to the amount of energy required for AI computing, a lot of water is needed to prevent these large computers from overheating. Like the fans used for cooling in most personal computers, chilled water is used to cool these data centers by absorbing heat from the equipment. Researchers estimate that for a simple conversation comprising of 20 to 50 questions, ChatGPT consumes the equivalent of a 500ml bottle of water. Multiple that water usage across all users, and it is estimated that by 2027 the annual water consumption of AI use may be equivalent to the annual water consumption of half of the United Kingdom.  

Many data centers are in areas where land is cheap and easy to develop, like deserts. This results in about a fifth of servers pulling water from moderately to highly stressed watersheds, putting further strain in those areas. Currently, about two-thirds of the world’s population experience at least one month of severe water shortage each year. This is expected to increase over the next few years due to many factors, including changes in temperature and precipitation as well as increased reliance on AI.

Other Environmental Impacts  

To accommodate the amount of data and processing involved, huge data centers are required to house all the computer infrastructure, including servers, data storage, and network equipment—imagine warehouse sized computers. These large data centers are temperature-controlled buildings, and their building footprint alone can take up several acres of land. Additionally, much like other electronics, the materials and equipment needed for these computing systems often require mined minerals and other materials, resulting in additional emissions and environmental impacts through the manufacturing process.  

Image

Possible Solutions and Mitigations 

So what can be done to alleviate these impacts? Investments in more sustainable technologies, such as transitioning to renewable energy sources like solar or wind, and prioritizing more energy efficient practices can mitigate the negative impacts related to fossil fuel consumption. Onsite renewable energy infrastructure can also alleviate demands on the electric grid.  

Regulations and policies can also be put in place at various levels of government to help contain the impacts. Whether that is requiring the use of renewable energy, reducing water consumption or ensuring these data centers are located in areas that can support the water and energy needs in a sustainable way.  

References:  

• AI’S Unsustainable Water Use: How Tech Giants Contribute To Global Water Shortages 
• Explained: Generative AI’s environmental impact | MIT News | Massachusetts Institute of Technology 
• AI has an environmental problem. Here’s what the world can do about that. 
• The Hidden Environmental Impact of AI 
• Can We Mitigate AI’s Environmental Impacts? 
• AI Is Accelerating the Loss of Our Scarcest Natural Resource: Water 
• How much water does AI consume? The public deserves to know - OECD.AI