This blog was written by Alex Walker, Program Manager, Climate Finance and Aliénor Rougeot, Senior Program Manager, Climate and Energy
In Part 1 of our series, we explored the rapidly expanding physical infrastructure behind artificial intelligence – the vast network of data centres that power our digital world. Now, let’s pull back the curtain on what these facilities require to operate and the environmental footprint they leave behind.
#1 Water
Data centres generate enormous heat. To prevent equipment failure, they employ extensive cooling systems that often rely on water – and the volumes they use are staggering.
In 2023 alone, Google consumed 23 billion litres of water in its data centres, which is equivalent to the yearly water use of 280,00 people in Canada for one entire year. Microsoft similarly withdrew 12.9 billion litres of water across its operations in the same year. Google and Microsoft have both reported year-on-year increases to their annual water use.
This consumption can create significant local impacts. In 2021, a Google data centres was responsible for almost 30 per cent of The Dalles, Oregon’s water consumption. For much of Google’s time in The Dalles, the area has experienced a multi-year drought. The state of Querétaro in Mexico is not only experiencing a boom in new data centre construction, but also depleting aquifers and drought, partly due to such construction.
Researchers estimate global AI demand may account for 4.2–6.6 billion cubic metres of water withdrawal by 2027 – more than the total annual water withdrawal from half of the United Kingdom.
#2 Energy
The electricity required to power and cool date centres represents another significant environmental concern. According to the International Energy Agency, data centres accounted for around 1.5 per cent of the world’s electricity consumption in 2024, while this is set to more than double by the end of the decade. AI applications specifically are driving much of this growth. Goldman Sachs analysis predicts that by 2028, AI will represent about 19 per cent of all data centre power demand.
While many tech companies are significant purchasers of renewable energy – Google, Meta, and Amazon are among the top five corporate buyers of wind and solar power globally – AI’s surging energy demand is also being used to justify extending the life of fossil fuel infrastructure. Coal plants are being kept online or converted to gas facilities specifically to meet this growing demand. In October 2024, BP’s CEO Murray Auchincloss explicitly cited tech “hyperscalers” as driving demand for natural gas, describing AI as “a major boon for the fossil fuel industry.”
The growing relationship between big tech companies, and the fossil fuel industry is cause for further energy-related concerns. Large tech companies are actively collaborating with oil and gas giants to increase oil and gas output. Machine learning and AI can help oil and gas companies to optimize their exploration and extraction processes. BP has a partnership with Microsoft using their Azure AI to determine the retrievability of hydrocarbons, while Google works with companies including Total to interpret subsurface imaging to aid extraction decisions. In particular, oil and gas companies use AI to cut down costs as they strive to compete with cheap renewable energy production.
This growing demand also poses risks for consumers. AI-induced electricity demand can lead consumers to pay significantly more for electricity. One study showed that electricity prices could increase by as much as 70 per cent in the next decade in North Virginia , which is currently the data centre capital of the world.
#3 Waste
Beyond water and energy, AI infrastructure has a substantial material footprint. The specialized chips, servers, and cooling systems used in AI and data centres require enormous quantities of resources to manufacture and are frequently replaced as technology advances.
This rapid turnover contributes to the growing global e-waste crisis. E-waste generation increased 82 per cent between 2010 and 2022, reaching 62 million tonnes annually. Only 22% of this waste is collected for processing, and a mere 1 per cent of the valuable metals and minerals are recovered for reuse.
The materials in AI hardware include heavy metals, forever chemicals (PFAS), and various plastics – many of which cannot be safely recovered with current recycling technologies. Current e-waste “recycling” often involves shredding and smelting, which can recover some metals but results in most plastics and chemicals being incinerated, creating air pollution.
The environmental justice implications are severe. Much of the world’s e-waste ends up in landfills, incinerators, or is handled by informal waste workers in the Global South, creating health hazards for vulnerable communities, including millions of children.
AI’s Environmental Benefits
Despite these substantial environmental costs, AI also offers potential environmental benefits when thoughtfully applied. Unlike AI tools that you might use at home or work, specialized AI tools are already helping address environmental challenges across various sectors.
In electrical grid management, AI is helping integrate renewable energy by better predicting supply and demand patterns. Google, for instance, developed an AI tool that improved the financial value of its wind turbine fleet by 20 per cent by more accurately predicting electricity generation output based on weather patterns.
Machine learning algorithms can identify inefficiencies in industrial processes, optimize logistics routes, and improve building energy performance. The Boston Consulting Group estimates that targeted AI applications could help mitigate 5-10 per cent of global emissions through such efficiency improvements.
The Balance Sheet
As we consider AI’s environmental impacts, we need an honest accounting of both costs and benefits. The water consumption, energy demand, and material footprint of AI infrastructure are substantial and growing. At the same time, thoughtfully applied AI offers real potential to address environmental challenges.
What’s clear is that without proper policies and safeguards, the environmental costs of AI could easily outweigh its benefits. In the final installment of our series, we’ll explore potential policy solutions and regulatory approaches to ensure AI development aligns with environmental protection and climate goals.
