AI might feel digital. Its environmental impact isn’t.

Most conversations about AI focus on what it can do. Faster answers. Better tools. More automation.

But usage is scaling fast. OpenAI’s coding tools alone are reportedly seeing around 4 million weekly users, and growing.

What gets talked about less is what it takes to run it. Not in terms of money, but in terms of energy, water, and infrastructure.

Because AI doesn’t just live in apps or interfaces. It runs in data centres. Large, physical buildings filled with hardware that need to be powered, cooled, and maintained constantly. And that has a footprint.

The part people don’t see

When you use an AI tool, it feels instant. You type something, you get a response. It’s easy to assume the cost is minimal. But behind that interaction is a chain of processes that require significant resources.

Data centres use large amounts of electricity to run and train models. They generate heat, which then needs to be managed. Cooling systems, often water-based, are used to keep temperatures within safe limits. That’s where things start to add up. Not in a single interaction, but at scale.

As usage grows, so does the demand for:

• electricity
• water
• physical infrastructure

And unlike software, those things don’t scale invisibly.

Scaling AI means scaling infrastructure

Right now, data centres are expanding quickly to meet demand. New facilities are being built, existing ones are being upgraded, and regions are competing to attract investment. But that growth isn’t frictionless.

Communities are starting to push back. Not necessarily against AI itself, but against what comes with it. Increased energy use, environmental impact, and the local effects of large-scale infrastructure projects. That tension is starting to surface more often, not just as a technical issue, but as a political and planning one.

A growing gap between visibility and impact

There’s also a question of transparency. Some large tech companies have pushed to limit how much detail is publicly shared about data centre emissions and environmental impact. In some cases, regulators have agreed to keep certain data less visible.

Which creates a gap. Between how visible AI feels to users, and how visible its environmental impact actually is. That doesn’t mean there’s something inherently wrong, but it does make it harder to understand the full picture.

Where this could go next

At the same time, there’s work being done to reduce the cost of running AI systems. Some researchers are exploring approaches that combine neural networks with more structured, logical reasoning. Early indications suggest these systems could use significantly less energy, potentially by large margins.

If that holds up, it could change the trajectory. Lower running costs, reduced environmental pressure, and wider access for smaller organisations. But for now, demand is still growing faster than efficiency gains.

So what does this mean?

AI is often framed as something fast, digital, and lightweight. But the reality is more physical. It depends on energy. It depends on water. It depends on infrastructure that has to be built, powered, and maintained somewhere.

And as AI adoption accelerates, those dependencies become harder to ignore. Not necessarily as a reason to slow down, but as something that needs to be understood alongside the benefits.