Artificial intelligence (AI) is revolutionising industries, streamlining businesses, and even shaping how we interact with technology daily. But beneath its sleek digital veneer lies a growing environmental footprint. AI demands vast amounts of energy, depletes water resources, and relies on mineral-intensive hardware. As AI becomes more embedded in our lives, we must address its sustainability challenges while harnessing its benefits for a greener future.

Power Consumption: The Growing Energy Appetite of AI

AI’s energy needs are staggering. Training large language models (LLMs) such as GPT-3 requires an estimated 1,287 megawatt-hours (MWh) of electricity—comparable to the annual energy use of 120 U.S. homes. And that’s just the training phase. Once deployed, these models continue to consume power for inference, answering queries, and running real-time applications.

Data centres, the backbone of AI, already account for nearly 200 terawatt-hours (TWh) of global electricity consumption—roughly 1% of the world’s total energy demand. And this is only expected to rise. Projections suggest that AI-driven data centres could require 68 gigawatts of power globally by 2027. With energy-hungry GPUs and TPUs running continuously, AI data centres now consume as much electricity as small cities.

To put it into perspective, a single ChatGPT query is estimated to use nearly 10 times the power of a standard Google search. If all search engine queries globally were handled by large AI models, the energy demand would skyrocket, consuming approximately 9 terawatt-hours annually, or 4% of all U.S. data centre energy use.

Water Usage: The Hidden Cost of AI

While energy consumption is a well-known issue, AI’s water footprint often flies under the radar. Data centres rely on water for cooling, preventing servers from overheating. A small 1-megawatt data centre can use up to 26 million litres of water per year, equivalent to the annual water consumption of 62 U.S. households. With AI workloads increasing, this water demand is set to rise.

Microsoft, for example, saw a 34% jump in water consumption from 2021 to 2022, reaching 1.7 billion gallons. The reason? Increased AI workloads and expanded data centre operations. And it’s not just about cooling—producing a single semiconductor chip requires over 2,200 gallons (8,300 litres) of water, making microchip manufacturing another significant water-intensive process.

Mineral Extraction: The Resource-Heavy Backbone of AI

AI hardware depends on a cocktail of minerals—copper, silicon, gallium, and rare earth elements—all of which require intensive mining. The International Energy Agency (IEA) warns of an impending copper shortage, with current mining projects expected to meet only 80% of global demand by 2030. Meanwhile, China, a key supplier of gallium and germanium, has restricted exports, raising concerns over AI’s supply chain stability.

Mining these minerals comes at a heavy environmental cost. Copper mining, for instance, can lead to deforestation, soil erosion, and water contamination. While some deposits contain trace amounts of uranium or thorium, most copper mines do not pose a significant radioactive waste risk—a common misconception. Still, mining’s broader environmental impact, from habitat destruction to CO₂ emissions, is a challenge the AI industry cannot ignore.

Can AI Help Solve Its Own Environmental Crisis?

The good news? AI itself is part of the solution. Companies like Google and NVIDIA are leading efforts to reduce AI’s energy demands through low-power inference algorithms and edge computing. Google’s DeepMind AI cut energy use in data centre cooling by 40%, showcasing AI’s ability to improve efficiency in its own infrastructure.

Moreover, many hyperscale data centres are transitioning to 100% renewable energy. Microsoft, Google, and Amazon are heavily investing in wind and solar power to offset their carbon footprint. Some firms are even exploring liquid immersion cooling and closed-loop cooling systems, which can significantly reduce water usage.

The Road Ahead: Balancing AI Growth with Sustainability

AI’s environmental impact is undeniable, but it’s not an insurmountable challenge. We can strike a balance between innovation and sustainability by prioritising energy-efficient AI models, optimising hardware, and investing in responsible mineral sourcing. AI also has immense potential to drive climate-positive applications, from optimising supply chains to improving energy grid efficiency.

However, there’s a catch—the rebound effect. The more efficient AI becomes, the more we tend to use it, potentially offsetting any environmental gains. To avoid this, ongoing research, policy interventions, and industry-wide cooperation will be crucial.

If we get it right, AI can be a force for environmental good rather than a burden. However, it will take conscious effort, innovation, and accountability to ensure AI’s future is as sustainable as it is intelligent.