Artificial intelligence is often presented as a climate solution. AI can help manage power grids more efficiently. AI can predict when crops will fail. AI can model new ways to capture carbon.

These are real applications. But they come with a cost that rarely makes the headlines: building and running AI systems requires enormous amounts of energy, water, and raw materials. Most discussions about AI and climate ignore this side of the equation.

The Numbers Nobody Talks About

One AI data center uses as much electricity as an entire town. Training ChatGPT—the AI chatbot millions of people use—used 210,000 to 700,000 liters of water. That’s enough to fill 280 to 900 Olympic swimming pools. Just to create one model.

And that’s not a one-time cost. Every update, every new version requires more water and more energy.

Building AI chips also requires mining rare metals like cobalt and lithium from places like Congo, Indonesia, and Peru. This mining damages land and pollutes water. Meanwhile, the finished computers go to data centers in wealthy countries.

The Real Problem: Every Question Costs Energy

Training an AI model is expensive. But you only do it once.

Using it is different. Every time someone asks ChatGPT a question, it uses energy. Every time. Asking an AI a question uses about 10 times more electricity than doing a Google search.

Think about it: millions of people use AI every day. If just 1 billion people ask AI a question once, that’s 1 billion energy-hungry queries in a single day. And the number keeps growing.

The International Energy Agency says AI data centers will use nearly 3 times more electricity by 2030 than they do today. That’s a huge jump. And much of that energy comes from coal and gas, not wind or solar.

The Geographic Dimension

Data centers are often located where electricity is most affordable. That frequently means countries in South Asia, Africa, and Southeast Asia. In many of these regions, the power grid relies heavily on coal and gas rather than renewables. This means that expanding AI infrastructure in these regions can increase local emissions and air pollution.

The labor side also matters. Data annotation—labeling images and text so AI can learn—is often done by workers in the Global South at much lower wages than equivalent work would cost in wealthier countries.

E-waste from discarded computer equipment is another issue. Older hardware frequently ends up in developing countries where recycling standards are less strict, potentially exposing workers and communities to toxic materials.

The Paradox

AI does have climate benefits. It can improve weather prediction. It can optimize power grids. It can reduce energy waste in some sectors.

At the same time, the infrastructure that powers AI uses enormous amounts of energy, water, and materials. Some of that energy still comes from coal and gas. The mining required for materials damages ecosystems.

This creates a paradox: a technology meant to help with climate change is itself a significant driver of resource consumption and emissions.

The industry is working on solutions. Companies are investing in renewable energy for data centers. Engineers are designing more efficient chips. These are real efforts. But efficiency improvements haven’t slowed the overall growth in AI’s resource use. Demand for new systems keeps increasing.

What Needs to Change

AI won’t become a climate solution until companies change how they build it.

First: Tell the truth. Companies should say exactly how much energy, water, and materials their AI uses. Show the real numbers. Not just marketing claims.

Second: Take responsibility for the whole chain. If cobalt is mined in Congo under harsh conditions, the company using that cobalt owns that problem. It doesn’t matter if the data center runs on solar power.

Third: Stop building unnecessary systems. Not every company needs its own AI. Not every problem needs an AI solution. If fewer systems existed, they’d use less energy. But that requires governments to make rules. Tech companies prefer to move fast.

Fourth: Protect people in the Global South. Countries like Pakistan, Bangladesh, and Nigeria shouldn’t pay the price for AI used in the West. That means fairer wages for workers. Real environmental protections for mining areas. Using renewable energy, not coal. And sharing the money these countries earn from hosting data centers.

The Bottom Line

AI might help fight climate change. But the way we’re building it right now is making things worse.

This isn’t an argument to stop using AI. It’s an argument to build it honestly. Companies need to admit what it costs. They need to take responsibility for the damage. They need to make sure people in poor countries don’t pay for technology that benefits the rich.

Because a technology that claims to save the planet while poisoning it is a lie. And the planet doesn’t care about good marketing. It only cares about what’s actually happening.