The global rise of artificial intelligence has reshaped the pace and scope of innovation. From medical breakthroughs to creative content generation and supply chain automation, AI’s influence has been both disruptive and empowering. Governments are integrating it into policy, companies are restructuring entire business models around it, and individuals are interacting with AI-powered tools in their everyday lives, often unknowingly. This technological transformation, however, has created a new kind of pressure: not on labor markets or institutions, but on the very environment that sustains us.

As AI models become more complex and demand ever-larger computational power, the infrastructure behind them like data centers, cooling systems, and energy-intensive training processes, has triggered a sharp increase in electricity usage, water consumption, and greenhouse gas emissions. These impacts are often obscured behind sleek user interfaces and futuristic headlines, yet they are unfolding rapidly and at scale. The environmental footprint of AI is no longer theoretical; it is measurable, tangible, and growing. As societies embrace the benefits of artificial intelligence, they must also confront the escalating cost it imposes on the planet’s finite resources.

The backbone of modern AI, data centers packed with high-performance chips, requires an astonishing amount of electricity to function. Training large language models, running complex simulations, and serving real-time AI applications demand continuous computing power at scale. This surge in demand is not linear but exponential, driven by the increasingly large and layered nature of AI models and the near-constant appetite for faster, more intelligent services.

The International Energy Agency (IEA) reported that data centers globally consumed approximately 1.5% of total electricity use in 2024, with growth rates hovering around 12% annually over the past five years. The lion’s share of this consumption is now being attributed to AI-specific workloads. If current trajectories continue, the IEA forecasts that this demand could balloon to 945 terawatt-hours by 2030—comparable to Japan’s entire electricity consumption today.

This level of growth is reshaping infrastructure planning in both the private and public sectors. Tech giants like Microsoft, Google, and Amazon are racing to secure access to clean energy sources, while utility companies are rethinking grid expansion strategies to accommodate AI’s voracious needs. A single hyperscale data center optimized for AI can consume as much electricity as 100,000 households, with new centers under development projected to draw 20 times more power than traditional sites. What was once a background service is now emerging as a dominant force in global energy markets—altering not only how we compute, but how we power the future.

While electricity powers AI, water cools it. Behind every chatbot response or image generation, there’s a sprawling server farm that must remain at stable operating temperatures 24/7. This cooling process, which is vital for maintaining the performance and longevity of high-performance computing systems, relies heavily on water, often in staggering quantities.

AI workloads are significantly intensifying this demand. According to projections by researchers at the University of California, training a single large model like GPT-3 can consume upwards of 700,000 liters of fresh water—about the same as producing 320 Tesla electric vehicles or 370,000 plastic water bottles. That’s just for one model’s training phase; the water footprint continues to grow during the thousands of daily inferences made as users engage with the model post-deployment.

Globally, data centers already withdraw an estimated 292 million gallons of water per day, a number expected to rise to 450 million gallons daily by 2030. Recent estimates project that global AI‑related water withdrawals could reach 4.2–6.6 billion m³ by 2027, surpassing the annual water use of more than half the United Kingdom. This encompasses not just evaporative cooling on site but also water used in power generation and chip manufacturing, components integral to AI infrastructure that are often overlooked. 

Europe is already feeling the pressure. A 2025 joint investigation by SourceMaterial and The Guardian revealed that Amazon’s proposed data centres in Spain’s arid Aragon region were licensed to use about 755,720 m³ of water annually—enough to irrigate around 233 hectares of corn fields. The development sparked protests from local farmers who warned that the centres could divert critical resources from agriculture and local communities. 

This water burden is exacerbated by climate variability and hotter, drier summers across Southern and Western Europe. European regulators are responding: the EU is developing minimum performance standards for data-centre water usage, to be proposed by 2026 under its Water Resilience Strategy. At the same time, industry actors under the Climate Neutral Data Centre Pact are committing to new Water Usage Effectiveness (WUE) thresholds, such as ≤ 0.4 L/kWh in water-stressed zones.

While AI holds promise for accelerating climate solutions, it paradoxically contributes to the very problem it aims to solve. The carbon footprint of training and deploying large AI models is significant and growing, especially when powered by electricity derived from fossil fuels. As the arms race to build ever-larger models intensifies, so too does the environmental toll.

The International Energy Agency projects that AI workloads could generate 300 million tons of CO₂ annually by 2035 if powered by today's energy mix. These emissions are often front-loaded—meaning most of the environmental damage occurs during training, rather than inference.

But the carbon cost is just one side of the story. Studies warn that pollutants released by fossil fuel plants powering data centers could cause up to $20 billion annually in public health damages by 2030 in the United States alone—due to respiratory and cardiovascular conditions linked to fine particulate matter, sulfur dioxide, and nitrogen oxides. This figure rivals the health impact of coal-fired steelmaking and reflects a shift in how digital infrastructure is shaping not just the environment, but human well-being.

The paradox is striking: as AI becomes an increasingly essential tool for modern life, it is also intensifying some of the very systems that undermine long-term sustainability and health. Without aggressive shifts toward clean energy and transparent environmental reporting, AI risks locking in a high-emissions trajectory that could offset its benefits elsewhere.

While AI’s environmental toll is undeniable, meaningful steps are being taken across both industry and policy landscapes to mitigate the damage. From clean energy investments to water-saving standards, these early initiatives mark the beginning of a global effort to make AI infrastructure more sustainable.

• Big Tech bets on cleaner energy: Microsoft, Google, and Amazon are accelerating their transition to renewable and low-carbon power sources. In 2024, Microsoft announced a nuclear energy deal with Constellation Energy to support its AI-driven data centers. Meanwhile, Google expanded its geothermal and solar energy footprint to align its operations with its net-zero targets.

• European regulation gains momentum: The European Commission is drafting environmental performance standards for data centers, including caps on water consumption and carbon emissions. These measures are part of the EU’s broader Digital Decade framework and could be formalized by 2030.

• Emerging technologies offer resource efficiency: Companies are piloting new methods to reduce AI’s environmental burden, such as immersion cooling and AI-aware scheduling that aligns high-energy tasks with periods of low-carbon grid availability. While still nascent, these innovations could become standard in next-generation data center design.

• Skepticism around enforcement and scale: Environmental experts caution that voluntary efforts may not keep pace with AI’s rapid expansion. Without binding accountability mechanisms, there’s a risk that sustainability will remain a side note rather than a core operational priority.

As AI continues to accelerate innovation, it’s also redefining what sustainability must mean in the digital age. The environmental costs of intelligent machines, from rising carbon emissions to resource-intensive infrastructure, can no longer be treated as an afterthought. Without structural accountability, industry transparency, and global regulation, AI’s growth may come at the expense of the very future it hopes to improve. The challenge now is clear: build responsibly, or risk undermining the intelligence we strive to achieve.