By: Susie

Read time: 10 minutes

The world seems wrapped up in the possibilities of artificial intelligence (AI). From increasing our lifespans, helping eradicate diseases, reducing our need to work, and even resolving our most pressing environmental problems, the promises of AI seem never ending. But despite all these potential benefits, AI exists on a planet with finite resources, and the environmental cost of this technology seems to be ever increasing. 

What is AI?

Artificial intelligence refers to computer systems that are designed to simulate human learning and memory processes. Generative AI, which is a type of AI used to generate images, text, videos, and other content, is the AI that many people are familiar with. 

AI technology is sweeping across the globe and has already transformed aspects of our culture. Tools like ChatGPT are revolutionizing the way people interact with the internet. But at the same time, AI also undermines human knowledge due to its ability to generate and spread misinformation. 

It isn’t all bad, though. AI could potentially help us solve environmental problems and is widely celebrated for its ability to detect patterns in data and help up predict future outcomes. Researchers from the nonprofit Imazon, a Brazilian based organization, developed an AI tool that helps to identify and prevent logging and deforestation in the Amazon rainforest. Solutions such as these can be promising in the fight against climate change, but it’s important to know at what cost.   

So how does AI impact the environment?

On the front lines of its development, communities across America are grappling with the direct environmental costs of AI: noise pollution, enormous energy and water requirements that strain local resources, and loss of habitats like wetlands, forests, and farms, as large tracts of land are rezoned to make way for what’s considered a modern-day gold rush.  

In addition to the more noticeable environmental impacts, raw material is needed to create computing hardware for AI models to thrive, which requires mining that is often unsustainable and harmful to the planet. The lack of transparency surrounding the mining operations makes this environmental problem feel almost invisible, but the impact it has should not be understated. 

AI developers make bold claims about its benefits, but they often leave out the environmental impacts from the conversation. In order for AI to truly be a revolutionary technology, education surrounding the environmental cost of AI needs to be increased to improve efficiency and implementation of this technology and offset adverse outcomes.

AI requires environmentally demanding data centers to run

To understand the environmental cost of AI, it is important to know that AI technology requires data centers to run. According to IBM, data centers are “physical rooms, buildings or facilities that house information technology (IT) infrastructure for building, running and delivering applications and services. They also store and manage the data associated with those applications and services.”

Data centers are not a new invention, and it is estimated that there are around 9,000 data centers globally. They have been around for decades, and in the early 2000’s, they underwent a revival thanks to the development of cloud computing. The data centers that store and process information on the cloud are so large that they are referred to as hyperscale data centers. These hyperscale facilities are like the data centers required for AI technology, except they’re often larger and require more water and energy to run. 

Data centers have a multifaceted impact on the environment which can be thought of in four phases: mining and manufacturing for raw materials, construction of the actual data center, operation of the data center, and the end use of the computer hardware, or e-waste, that is generated because of the process.

  1. Mining and manufacturing 

The first and often one of the most exhaustive phases of data center development is mining for raw materials. Computer hardware such as processors, servers, hard drives, and mainframes require critical minerals and rare earth elements that must be mined from the Earth. 

Mining is one of the most environmentally destructive practices on the planet because it defaces entire ecosystems and exposes humans to hazardous conditions. The environmental and human health regulations of mining also vary significantly by country, leaving many gaps in our understanding of the true extent of harm. 

In addition to mining, the raw materials must be manufactured into a final product and transported across the globe so that they can be used in the construction of a data center. This process requires energy, largely in the form of burning coal, or fossil fuels, for transportation and manufacturing, which leads to air pollution, water pollution, and increases carbon dioxide in the atmosphere. 

  1. Data center construction

Once all the components for a data center are manufactured, the actual data center can be designed and constructed. Construction of a data center will inevitably require land, and while development footprints of data centers can vary widely, emerging AI data centers are unusually land consumptive.

For example, an Amazon data center designed specifically for AI scheduled for construction in New Carlislie, Indiana, sits on around 1,000 acres. That is nearly the same size of the entire town itself. 

The Amazon data center in New Carlise will destroy approximately 10 acres of wetlands and be primarily built on farmland. Habitat loss, in the form of wetlands, farmland, and forests, is a major impact of the construction phase. If not carefully designed and placed, data centers can have a negative impact on the underlying environment. In addition, water pollution, water flow disruption, and noise pollution are also environmental impacts of the construction phase. 

  1. Operation of data center: Electricity demand and water use

The operation phase of the data center is often credited with the most significant environmental impacts since it requires disproportionate amounts of energy and water to power data centers and cool the equipment. 

Researchers have estimated that AI tools such as ChatGPT require 5 to 10 times the amount of energy when compared to a traditional web search. The increased demand for data centers to power AI is expected to significantly increase the global electricity demand. The International Energy Agency (IEA) estimates that over the next couple of years, electricity demand from data centers could more than double.

The Amazon data center in New Carlislie is projected to use as much electricity as around 1.5 million households, or about half of the state of Indiana. Developers of the Amazon data center claim that it will be powered by renewables, but some researchers have their doubts about data centers being powered by renewable energy on a large scale. 

Noman Bashir, lead author of the research paper The Climate and Sustainability Implications of Generative AI, saysthat “the demand for new data centers cannot be met in a sustainable way. The pace at which companies are building new data centers means the bulk of the electricity to power them must come from fossil fuel-based power plants.”

The energy needs for data centers are largely being met by unsustainable methods such as coal fired power plants, which is a major contributor to climate change. In Indiana, burning coal also means generating more coal ash which is improperly disposed of, creating additional water pollution and human health risks. 

In addition to the energy needed to power data centers, large amounts of water are also required to cool the computer equipment. While water usage requirements vary, some researchers estimate that for every kilowatt hour of energy that a data centers consumes, it would need about a half-gallon of water for cooling.  

Citizens Action Coalition, an Indiana based nonprofit that is currently advocating for a moratorium on data centers, estimates that a single data center may use anywhere between one to five million gallons of water a day. The proposed Amazon data center and General Motors Battery Plant signed an agreement with the City of New Carlise that caps their combined water use at 24 million gallons of water per day from the Kankakee River aquifer, or about half of the aquifer’s safe usage limits, though it is unclear how much of that water will be used directly by the Amazon data center. 

In Dalles, Oregon, Google operates three data centers. In 2022, the company filed a lawsuit to keep their water use a secret, but public records now show that Google’s data centers use more than a quarter of the city’s water supply. 

Water use is already a major concern among cities. Poor planning, overuse, and the looming prospect of water stress and drought create challenges for maintaining adequate levels of water in aquifers. Without proper allocation and planning, data centers can threaten groundwater resources, even in regions with seemingly abundant water resources. 

  1. Electric waste 

Many of the computer hardware components required to keep AI running have an expiration date. For example, hard disk drives, which are data storage devices, have typical lifespans of three to five years. Russ Ernst, Chief Technology Officer of data security company Blancco, writes “the need for constant upgrades to support evolving AI capabilities can shorten the lifespan of data center devices and drive faster cycles of upgrading equipment. With an estimated lifespan of just three to five years, between 20 and 70 million expensive hard disk drives in the US reach their end of life each year – most of which are shredded and sent to landfills, creating vast amounts of waste.”

Where do we go from here?

AI can yield many positive benefits for humanity, but during this early phase of its development, it is critical to proceed with caution. Our environment is already suffering on a global scale: climate change is increasing weather irregularity and intensity, bringing waves of intense heat, cold, drought, rain, and storms. We are dealing with widespread pollution, species extinction, and habitat loss. On our current path forward, the planet cannot sustain our rates of consumption for much longer. Therefore, it is imperative that AI development strikes a sustainable balance.

Researchers at the Massachusetts Institute of Technology argue that this “will require a comprehensive consideration of all the environmental and societal costs of generative AI, as well as a detailed assessment of the value in its perceived benefits.”

The good news is that with our ever-increasing awareness of environmental problems, many people aren’t looking away. According to writers with the United Nations, “more than 190 countries have adopted a series of non-binding recommendations on the ethical use of AI, which covers the environment. As well, both the European Union and the United States of America have introduced legislation to temper the environmental impact of AI.”

While these policies are at the early stages of development, it’s a positive sign that countries are already aware of the environmental impacts of AI and they are beginning to do something about it. In the meantime, here are some strategies we can implement to lessen AI’s environmental impact:

  • Increasing transparency of AI’s environmental impacts
  • Limiting the construction footprint of AI data centers and protecting sensitive ecosystems from development 
  • Stricter regulation of mines and recycling materials to reduce electric waste 
  • Optimizing water and energy use 
  • Balancing data center development with other forms of economic development

Have a comment about this article?

Email the author, Susie McGovern, Water Science and Sustainability Specialist at smcgovern@hecweb.org.


Categories: Climate Change, Sustainable Development and Green Infrastructure