Sustaining AI and the Environment

User icon
Data Society
Icon clock
November 2023
File icon

Humankind’s remarkable history of progress has often brought us face-to-face with unintended consequences. Just as the Industrial Revolution introduced new forms of workplace dangers, the dawn of the Internet ushered in a more insidious mechanism for bullying among peers, fraud, and misinformation. Such byproducts of technology exemplify the importance of constructing guardrails to make sure that, in the long term, these types of technological advances result in a net-positive outcome for humanity. 

We are now confronting that moment with generative AI. As we grapple with this technology’s social and ethical implications, we also need to address its considerable environmental impact. Fortunately, there are steps we can take to reduce its ecological footprint and use its transformative power to become better stewards of the planet.

Generative AI’s Gluttony for Natural Resources




AI will no doubt proliferate at a rapid clip, with its global market size projected to grow at a CAGR of 36.8% percent between 2023 and 2030. Generative AI, whose large-language models are especially demanding when it comes to consuming natural resources and generating greenhouse gas emissions, will likely drive much of this increase. Based on what we already know about the environmental impact of this nascent technology, we can only imagine its potential toll on the planet as it takes hold more expansively, as the volume of data continues to explode, and as models become increasingly complex.

Environmental threats are present in several processes related to generative AI. Areas of particular concern include:

The Carbon Emissions of Model Training and Inference - Generative AI requires tremendous computational power, and carbon emissions associated with AI training and inference are significant. Emissions from training a common model can be comparable to nearly five times the lifetime carbon dioxide emissions of an average American car, including emissions associated with the car’s manufacture. Further, according to OpenAI, the processing power used for deep learning research since 2012 doubled every 3.4 months through 2018, increasing by an estimated 300,000 times in that period.

The Resource Usage of Data Centers - The power stations driving many of today’s cutting-edge technologies, data centers have an especially voracious appetite for natural resources and have staggeringly large carbon footprints. Their processing, maintenance, and cooling needs lead to massive energy consumption, which equals 10 to 50 times the energy consumption of a typical office building per floor space, according to the US Department of Energy’s Office of Energy Efficiency and Renewable Energy. In addition, to meet tremendous cooling demands, a typical data center can chug an estimated three to five million gallons of water a day. 

The Contaminating Sprawl of E-waste - As adoption and development of AI technologies ramp up, so also will the production and eventual demise of related hardware left behind in the form of e-waste. Only 17.4 percent of the world’s e-waste is recycled in an eco-friendly way. The result is mounting volumes of residual toxins that present environmental challenges including soil contamination from materials such as lead and mercury and air pollution. E-waste and the failure to recycle or dispose of it properly can also release toxic gasses into the atmosphere and metal poisonings into water.

The Greening of Generative AI

Fortunately, there are two promising approaches that can enable us to embrace generative AI while caring for our planet. The first approach involves measures that organizations can take to limit the carbon footprint of their AI projects, and the second involves applications of AI itself that can mitigate negative environmental impact and lead to more efficient, greener use of natural resources. 


What Organizations Can Do

According to the McKinsey 2022 State of AI report, 40 percent of organizations say they are working to reduce AI environmental impact by limiting energy used to train and run models. For example, by monitoring a calculation’s carbon footprint and incorporating optimization algorithms into development projects, organization’s can limit the environmental impact of their models. 

Another way that organizations can accomplish this goal is by practicing good old-fashioned frugality when it comes to their use of these technologies. For example, they can make judicious use of large language models with large energy demands and, more generally, use generative AI only when it produces significantly more business value. As with any effort to reduce waste and conserve resources, organizations who value environmental sustainability should develop best practices for using only what they need when it comes to technology.

In addition, organizations can influence sustainability practices by prioritizing environmental considerations in their relationships with the large companies behind today’s generative AI technologies. Just as consumers are increasingly holding companies accountable for their environmental impact, companies should establish and enforce sustainability standards for their vendors and partners. The large technology juggernauts have the resources and scale to support more efficient practices and more ecologically sound sourcing choices, and many of today’s technology giants have plans in place to reduce their carbon footprints and replenish water in the years to come. As organizations select providers for services such as cloud computing and AI solutions, they have the power to shape their own environmental impact by choosing to work with companies whose sustainability practices align with their values.

It’s important to note that resource usage related to generative AI can vary widely depending on where they are sourced. For example, where there are more fossil fuels used, carbon emissions will be higher. By making strategic geographic choices about where various operations are located, the large technology companies can help limit emissions and distribute water demands to reduce the overall environmental burden. 


What AI Can Do 

The McKinsey 2022 State of AI report also says that organizations that lead in AI usage are 1.4 times more likely to report AI-enabled sustainability efforts and that they are working to decrease AI-related emissions. Of organizations using AI, the report states, 62 percent report improving environmental impact through such measures as increased energy efficiency and optimized transportation.

In addition to reducing fuel usage and emissions through route optimization and predictive maintenance, examples of generative AI at work on behalf of the environment are on the rise across industries. Some fascinating use cases include: 

  • AI algorithms that can monitor inventory levels and analyze performance data to extend the lifespan of technology products, classify components, and help organizations manage e-waste responsibly. 
  • Google’s application of its DeepMind machine learning technology to create super-efficient services, reducing its energy use for data center cooling by 40 percent.
  • AI-powered tools that monitor and measure emissions and air quality in real time.
  • A digital twin platform that helps cities simulate the impact of carbon emissions and create sustainable climate plans.  

Tackling Environmental Challenges Today for AI-Driven Sustainability Tomorrow

There is no question that today’s emerging technologies pose critical challenges in the form of a tremendous environmental footprint. However, understanding the magnitude of the challenge ahead equips us to make decisions about how we can benefit from this technology while helping to reduce its harmful effects on our planet. And, unlike many innovations throughout history, generative AI can also empower us to discover unprecedented solutions to its potential threats within the technology itself.

Subscribe to our newsletter

Subscribe to get the latest updates from Data Society, including tips for how to use your data better, real-life examples of leveraging analytics, and more.

cross linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram