The natural cost of Artificial Intelligence

Rapidly increasing carbon footprint of AI

The explosion in use of Artificial Intelligence around the world over the past few years has led to a spike in demand for global energy and water as well as accelerated e-waste, leading to a sharp increase in carbon emissions due to AI. Data centres and AI hardware strain resources, threatening ecosystems and overwhelming waste management systems worldwide.

In an era where Artificial Intelligence (AI) is celebrated as the next frontier of human ingenuity, few pause to consider the silent toll it exacts on our environment. From powering virtual assistants to generating images and optimising logistics, AI is woven into the fabric of modern life. But behind every ChatGPT query and AI-driven recommendation lies a growing ecological crisis, one that threatens to accelerate climate change, deplete freshwater reserves and overwhelm waste management systems, both globally and in India.

AI’s environmental footprint is not just about data centres, though they are a major part of the problem. According to the International Energy Agency (IEA), global electricity demand is projected to grow by 4 pc over the next three years, driven in large part by the expansion of data centres and increased AI workloads.

The energy sector already produces close to 76 pc of all global emissions, with manufacturing and transportation accounting for 13 pc and 14 pc, respectively. In comparison, data centres currently contribute about 3.5 pc of global emissions, a figure that may seem modest but masks the broader impact of AI.

“AI’s environmental footprint is more than just data centre emissions. We also need to account for mining of raw materials, manufacturing of equipment, transportation of the same, water and energy use as well as e-waste generation. As AI use is significantly rising and more data centres are being built all over the world, the environmental impact of AI is only expected to go up. We cannot let this go unchecked in an age of climate change and triple planetary crisis,” Kavya Manoharan, an environmentalist based in Chennai, tells Media India Group.

The rapid adoption of generative AI, like ChatGPT and DALL-E, has led to a dramatic increase in data centre construction and electricity consumption. Massachusetts Institute of Technology (MIT) news reports that the electricity demands of data centres are a major factor contributing to the environmental impacts of generative AI.

Reportedly, data centre power consumption increased by 72 pc from 2019 to 2023 alone, largely due to AI’s rise. By the end of this decade, AI technology is expected to consume nearly as much energy as Japan currently does, with only about half of that demand likely to be met by renewable sources. In fact, by the end of 2025, AI will require almost twice the power needed by the Netherlands, 23 gigawatts compared to the Netherlands’ 12.4 gigawatts.

The environmental impact of AI extends far beyond electricity. Data centres are also voracious consumers of water, needed to keep servers cool. Microsoft, for example, consumed 6.44 billion litres of water in a single year, a 34 pc increase largely driven by AI research. Globally, data centres use about 560 billion litres of water annually, a figure projected to double by 2030 as AI workloads increase and global temperatures rise.

“Running of these data centres and cooling of the equipment require a lot of energy and water. More often than not, this energy comes from burning fossil fuels and the water comes from fresh water sources,” Manoharan adds.

The sheer scale of water use is alarming as AI-related infrastructure may soon consume six times more water than Denmark, a country of six million people. This is especially concerning given that over 25 pc of the global population already lacks access to clean water and sanitation.

The environmental toll of AI is not limited to energy and water. The equipment used in data centres eventually becomes e-waste as in their race to stay ahead of competition, tech companies are known to constantly upgrade their hardware.

“The equipment used in data centres eventually become e-waste and tech companies are known to constantly upgrade their equipment. e-waste contains hazardous substances like mercury and lead. While it is the fastest growing category of waste, e-waste recycling is still in its nascent stages and mushrooming of data centres and increased adoption and use of AI infrastructure can be expected to put undue pressure on waste collection systems and recycling infrastructure,” says Manoharan.

Graphics Processing Units (GPUs), the chips that make AI possible, are replaced rapidly and are made of rare earth elements, requiring more energy to manufacture than Central Processing Units (CPUs).

“Moreover, GPUs are replaced rapidly, too. These chips are made of rare earth elements and require more energy to manufacture, than a CPU. With rapid growth of data centres and AI infrastructure, mining and extraction of these materials would increase, and so would the resultant energy consumption. This can have multiple environmental, social and ethical concerns,” Manoharan adds.

The carbon emissions from AI are also surging. The United Nations International Telecommunication Union (ITU) and World Benchmarking Alliance report that the operational emissions of Amazon, Microsoft, Alphabet and Meta skyrocketed by 150 pc on average from 2020 to 2023. Amazon’s total operational emissions increased by 182 pc, while Microsoft’s surged by 155 pc.

Each ChatGPT query emits about 4.32 grams of CO₂e and a ChatGPT request uses 10 times as much electricity as a Google Search.

“The training process for a single AI model can consume thousands of megawatt hours of electricity and emit hundreds of tons of carbon. Furthermore, AI model training can lead to evaporation of large amounts of freshwater into the atmosphere due to data centre heat rejection, exacerbating stress on already limited freshwater resources. As for the impacts of using these models in our daily lives, a ChatGPT search requires more energy than a simple Google search. Researchers estimate that a ChatGPT query consumes 5-10 times more electricity than a simple web search. Thus, it is important to avoid using AI for non-essential uses,” Manoharan adds.

The environmental impact of AI is further compounded by the need for raw materials, the manufacturing and transportation of equipment, and the resulting e-waste.

According to the United Nations Environment Programme (UNEP), most large-scale AI deployments are housed in data centres and these data centres take a heavy toll on the planet. The electronics they house require significant amounts of raw material to make. The GPUs that power AI need rare earth elements which are often mined in environmentally destructive ways. Then there are also the emissions from manufacturing and transportation of these equipment.

Moreover, e-waste produced by data centres contain toxic and hazardous substances like mercury and lead and if disposed incorrectly, which is a common practice worldwide, this waste can poison the environment and communities while causing some seriously irreparable damage.

In India, where rapid digitalisation is underway, the environmental impact of AI is equally pressing. The country is home to a growing number of data centres, many of which are located in regions already facing water scarcity. With e-waste management systems still developing, the surge in AI-related hardware could overwhelm recycling and disposal capacities. Despite the scale of the problem, regulation remains woefully inadequate.

“World-over, while there are a few initiatives in place which are aimed at regulating AI, hardly any of them focus on AI’s environmental impact. It is my opinion that the current global policy/regulatory landscape is inadequate and severely falls short, given the scale of the problem. There should be standardised measurement and disclosure of AI’s environmental impact, disincentives for inefficient algorithms, mandates for renewable energy and recycled materials, and strict rules on water use and data centre location,” Manoharan concludes.