Connor Wright
Partnerships Manager,
Montreal AI Ethics Institute
Renjie Butalid
Co-Founder & Director
Montreal AI Ethics Institute
AI’s transformative power comes with a cost. The Montreal AI Ethics Institute sheds light on the environmental footprint of artificial intelligence and advocates for sustainable solutions.
Artificial Intelligence (AI) often feels intangible, existing somewhere “in the cloud” as a seamless helper for writing emails, generating images, or automating processes. But beneath this virtual veneer lies an energy-hungry, resource-intensive infrastructure that demands real-world materials and impacts. At the Montreal AI Ethics Institute (MAIEI), where we focus on demystifying AI’s ethical and environmental challenges, we’ve observed how AI’s material reality is too often ignored.
AI’s Environmental Footprint
The environmental toll of AI is immense, spanning water, electricity, heat, and carbon emissions. While these impacts are familiar in industries like transportation or manufacturing, AI’s unseen infrastructure can make them easy to overlook.
Water Usage:
Cooling (typically via using water) is critical to prevent overheating in the data centers that house the millions of servers powering AI systems. Microsoft, for example, houses four million data servers globally in over 300 data centers, while Amazon has over 100 data centers to Meta’s 26. This reliance on water is profound. In Taiwan, where the world’s largest chip manufacturer TSMC is based, a 2021 drought forced the government to truck in water for factories, even as farmers were subsidized to stop growing water-intensive rice. In the U.S., residents in Bluffdale, Utah, face water shortages courtesy of the nearby Utah Data Center. Meanwhile, Microsoft and Google’s water usage has skyrocketed due to their focus on AI, with research estimating that a single ChatGPT query for a 100-word email consumes over 500 milliliters of water—more than a standard bottle. As one research study puts it, “AI is thirsty.”
Electricity Consumption:
Keeping data centers cool also requires enormous amounts of electricity. Reports project that by 2026, data centers will consume a third of Ireland’s electricity. Training advanced AI models, like large language models (LLMs), uses electricity comparable to that of hundreds of thousands of households annually. Even a single ChatGPT query can require ten times the electricity of a basic Google search. In January 2023 alone, it was estimated that OpenAI’s electricity consumption (the organization that created ChatGPT) roughly translated to what 175,000 Danish families use in one year. As LLMs grow larger, these energy demands are only expected to increase.
Carbon Footprint:
As energy use increases, so does AI’s carbon footprint. Research estimates that training LLMs can produce almost as much carbon (1438 lbs) as one trans-American flight (1984 lbs). While these estimates are improving as AI systems become more efficient, carbon accounting has yet to become mainstream in AI practices. Research continues to prioritize accuracy improvements over hardware efficiency.
A Path to Sustainable AI
Efforts to address AI’s environmental toll are beginning to take shape. Google has pledged to replenish 120% of the water it consumes by 2030 and partnered with Kairos Power to explore nuclear energy for reducing emissions. Meta and Amazon have made similar pledges, while Microsoft experimented with underwater data centers to improve cooling efficiency, but discontinued this effort in mid-2024. Researchers are also advocating for smaller, more resource-efficient AI models, partly in response to diminishing performance gains from ever-larger systems.
The Reality Behind AI
AI’s environmental impact is not abstract; it is real, material, and consequential. Its infrastructure competes for resources—water, energy, and land—already under strain from climate change. As AI continues to transform industries, integrating sustainability into its design and development is not optional; it’s critical.
At MAIEI, inspired by our late founder Abhishek Gupta’s award-winning work, “The Imperative for Sustainable AI Systems,” we advocate for ethical and sustainable AI practices. Recognizing AI’s material reality helps us make smarter decisions, ensuring this transformative technology aligns with the planet’s future.
AI may be artificial in name, but its footprint is undeniably real.
To learn more, visit montrealethics.ai.
