Written by Malena Sanchez Moccero
Art by Jouli Di Marco

Every time we ask a digital assistant for help, we trigger a chain of events that consumes energy, water, and mineral resources. How sustainable are these technologies we now rely on every day? Who will foot the energy bill of the future? And most importantly—can we do anything about it?

When OpenAI launched ChatGPT in November 2022, it became the fastest-growing app in history, reaching one million users in just five days. Since then, millions of people around the world have adopted generative AI (GenAI) tools to write text, generate images, code, translate, brainstorm, and much more.

But while these tools may feel effortless and almost magical, their operations rely on real infrastructure, natural resources, and human labor. The environmental cost of generative AI is one of the least visible—but increasingly urgent—issues of our time.

Behind every AI-generated image, chatbot reply, or automatic email summary lies a global system of energy-intensive data centers, massive underwater cables, freshwater usage, rare mineral extraction, and significant carbon emissions.

It’s time we pay attention to the material footprint of our immaterial technologies.

AI Models Need Training—and That Requires Energy

At the heart of generative AI systems are large language models (LLMs) trained on vast amounts of data. These models, such as GPT-4, Claude, and Gemini, are powered by machine learning algorithms that require staggering computational resources.

Training these models is no small task. According to research, training GPT-3 produced over 500 metric tons of CO₂—equivalent to over 1 million miles driven by an average gasoline-powered car. 

As more powerful models are released, their carbon footprints grow accordingly. The AI race isn’t just a competition over accuracy or speed—it’s also one over environmental cost.

The Cloud Has a Physical Weight

Despite the name, the “cloud” isn’t weightless. It’s a network of physical data centers spread across the globe, housing thousands of high-performance GPUs (graphics processing units) and CPUs. These servers run 24/7 to support AI tools and online services, and they generate massive amounts of heat in the process.

To keep servers cool, companies rely on advanced cooling systems, many of which use large volumes of freshwater.

For example, in 2023 alone, Google used nearly 5 billion gallons (20 billion liters) of water globally to cool its data centers. 

And it’s not just water. These centers also consume enormous amounts of electricity, often powered by fossil fuels—especially in countries with coal-heavy energy grids.

One Chat, Half a Liter of Water

It may sound hard to believe, but every question you ask ChatGPT—or another generative AI model—uses water.

A joint study by researchers at the University of California, Riverside, and the University of Texas at Arlington found that a 20–25 question session with ChatGPT consumes about 500 milliliters (half a liter, or roughly 0.13 gallons) of water, mostly due to cooling needs at data centers.

Now multiply that by the millions of daily users of AI tools around the world, and the total freshwater impact becomes alarming—especially in regions already facing drought or water scarcity.

Making things worse, many hyperscale data centers are being built in arid regions of the western United States, where water stress is already high. 

From Minerals to E-Waste: The AI Supply Chain

The environmental footprint of AI doesn’t end with water or energy. Building the hardware that powers generative AI systems requires the extraction of finite minerals like lithium, cobalt, and rare earth elements. These are essential for chips, batteries, and servers—but mining them comes with high social and environmental costs.

Beyond mining, the disposal of electronics creates another problem. Obsolete GPUs, servers, and smartphones often become e-waste, containing heavy metals and toxic materials that leach into soil and water if not recycled properly.

Are Big Tech’s Emissions Being Underreported?

Many of the world’s largest tech companies have made public commitments to carbon neutrality and green energy. But recent investigative reports question the transparency and accuracy of their sustainability claims.

Why the discrepancy? Often, companies rely on carbon offset schemes, buy renewable energy credits, or only report direct emissions—ignoring the full scope of their environmental impact, including upstream supply chains and downstream usage.

Experts warn that, if current trends continue, the data center industry could emit over 2.5 billion metric tons of CO₂ by 2030—equivalent to the annual emissions of more than 500 million gasoline-powered cars.

Seeing the Invisible Infrastructure

Because the environmental cost of AI is not directly visible—unlike a landfill, smog, or oil spill—it can be easy to overlook. But artists, designers, and researchers are starting to expose this hidden layer of the digital world.

Projects like Anatomy of an AI System visually map the supply chain behind a simple Amazon Echo speaker, revealing the global flow of materials, labor, and data. By making the invisible visible, these kinds of projects help us recognize that every “digital” action we take—every prompt, image, and reply—relies on physical resources.

What Can We Do?

Generative AI is here to stay. Its creative and economic potential is immense. But that doesn’t mean we should accept its environmental toll as inevitable.

As users, we can also make choices. Not every task needs to be powered by AI. Sometimes a simple search or a lightweight app can get the job done. Maybe asking ourselves whether we really need a bot to write our emails or a smart assistant to plan dinner is, in itself, a way to care for the planet.

Still, it’s important to recognize that placing the entire burden on individual users would be unfair. Real change has to come from the top. Big tech companies—and the governments that regulate them—bear the greatest responsibility for reducing AI’s environmental impact.