Water, Power, and the Planet
AI’s Hidden Price

Imagine you ask ChatGPT to craft an article or compose a poem – everything pops up in seconds and you are amazed. But here is a lot more than meets the eye. Artificial Intelligence (AI) might be increasingly becoming an indispensable part of our lives, but these technology powering tools come with a hidden environmental cost. As AI continues to create shifts in industries and our daily lives, its impact on the environment—especially through its massive consumption of electricity and water—is becoming hard to ignore. The question is: can we afford this price for progress?

By Amrita Sengupta

Climate change is one of the greatest challenges facing the global community. This is apparent in ecologically sensitive regions like South Asia. India, for example, has seen a sharp increase in extreme weather events—rising from fewer than 50 annually in the 1970s to almost 200 per year in 2019. Amidst this, AI and its environmental impacts has become a subject of growing concern.

One of the concerns is the setting up of numerous data centers. Recently in Chile, Google had to halt its plan to set up a large data center over concerns of pressure on local power grids. Similarly, Uruguay witnessed protests from environmentalists concerned with the data centre releasing chemical hazardous waste in the area, negatively affecting communities. And that’s not all. All of these demands staggering amounts of energy and water, resources already under strain in many parts of the world.

The nuts and bolts: understanding AI infrastructure

To understand AI’s environmental footprint, we need to peek under the hood. At the heart of AI lies cutting-edge infrastructure that end up consuming electricity and water.

The first is AI chips—a critical component specializing in tasks like calculations, data processing and model training. But they require a huge amount of water in both manufacturing of the chips and subsequently cleaning them.  

The second are data centers. Think of them as AI factories, where all the processing takes place. These large buildings provide the essential infrastructure for the chips to work 24/7 for efficient AI application processing. First, to run them, you need to power it with significant quantum of electricity, Further, to prevent overheating, they require cooling systems—which are created by either using water, or air conditioning systems or via natural cooling.

In short, AI requires significant amounts of natural resources, including the mining of resources like copper, and the use of energy and water, thereby leaving a significant carbon footprint—the amount of greenhouse gases generated by various processes and actions that individuals or large-scale industries undertake.

Various climate researchers, looking at water and emissions-related risks, also postulate which regions could be particularly vulnerable to data center related resource use. Geographically, it therefore raises the question of the placement of these data centers, how might they impact local communities living in the vicinity of these data centers that generate waste and emissions and also use up important resources that may be scarce in certain areas.

AI and its need for water and electricity

AI applications rely on massive swaths of computational resources to perform large-scale data processing and decision-making. When we ask ChatGPT to create an email for us, it processes huge amounts of data it has been trained on, requiring computational resources that depend heavily on electricity. Quantifying the use of energy is a difficult task, given the various points at which energy is used, from training to deployment. In addition, whether the energy comes from renewable or non-renewable sources is another aspect to be considered. Currently, North America has the highest number of data centers, followed by Western Europe. This is because these centers need to be kept cool, and what more effective way than placing them in countries that are naturally cold!?

Water has been traditionally used for the cooling process, an issue that has now become a point of conversation amongst States and corporations developing these data centers.  This has called for using alternative processes of cooling which are not resource intensive. Ideas such as the use of refrigerants, techniques of immersion cooling (where servers are submerged in liquid that absorbs heat from the components), and geothermal cooling (underground cooling) are being trialled.

In conclusion: to AI or not to AI

Not all is lost. There are steps that can be taken to mitigate the environmental costs of AI.

To begin with, organisations need to make conscious efforts to develop and deploy AI where it is beneficial instead of training AI models indiscriminately.

Secondly, the supply chain should be scrutinised and efforts should be made to reduce their footprint across the chain instead of only at the point of data centers.
 
Thirdly, more funding should be allocated to researching and finding robust tools to estimate carbon footprint across different stages including emissions from training, operations, and usage.
 
Finally, governments need to consider the costs of AI-driven emissions and create guardrails to its unchecked use. This is not new and has been traditionally applied across various industries—from coal to aviation and beyond. The Dutch government, for instance, is seeking to apply controls on hyperscale data centers after determining that the cost of energy outweighs the benefits from the deployment of AI-technologies.

So, the next time you use AI, take a moment to think about what powers it. It’s not just data—it is water, electricity, and the balance of our environment.