Netural Carbone Zone
Welcome to our new website!

The Climate Cost of AI

21 May, 2025

Artificial Intelligence is changing the way we live, work, and solve problems. From helping doctors diagnose diseases earlier, to optimising how we manage national power grids. And while it holds real promise in accelerating climate action, there’s a catch.

AI itself is becoming a growing contributor to environmental harm.

In a world racing towards net zero, we can’t afford to ignore the carbon, water and material cost of our digital choices. A new report from the National Engineering Policy Centre (NEPC) sets out the facts, and a five-step plan to make sure AI is part of the climate solution, not the problem.

 

The Environmental Cost of AI: What’s Really Happening?

Let’s break it down.

AI runs on data centres, vast warehouses of servers that require massive amounts of electricity to process, store, and transfer data. That electricity often comes from non-renewable sources. These same servers also require constant cooling, which in most cases involves large volumes of potable water.

The demand is soaring. Microsoft, for example, reported using nearly 7.8 billion litres of water in 2023 alone, up from just over 4 billion litres in 2020. Much of that rise is linked to the growth of AI workloads and infrastructure expansion.

Globally, the training of a single large language model (like GPT-3) can generate over 500 metric tonnes of CO₂e—equivalent to driving a petrol car more than a million miles.

But the real footprint starts after the model is trained. That’s the inference phase, the stage when users interact with AI, asking questions, generating images, running recommendations. This ‘use phase’ can account for up to 80% of an AI model’s total energy consumption over its lifetime.

And it’s only accelerating. Since 2010, the computing power used to train state-of-the-art AI models has grown over 4x per year. Meanwhile, data centre energy demand in the UK alone is forecast to increase sixfold by 2035, according to the CEO of National Grid.

If these trends continue, AI workloads could push local electricity grids to breaking point and outpace the growth of clean energy. Some projections even suggest that AI-related energy demand could outstrip new renewable generation capacity within a decade in developed countries.

Water: The Hidden Cost

We often focus on electricity, but water is just as critical. In 2023, Google revealed that 78% of its global water withdrawals came from potable sources, water that could otherwise support agriculture or human consumption.

In the UK, water scarcity is already a growing concern. By 2030, seven English regions including London, home to over 75% of the UK’s data centre capacity are predicted to face severe water stress. Data centres withdrawing from local supplies only compounds the issue.

To make matters worse, semiconductor manufacturing, the beating heart of AI, also consumes water at an astonishing rate. A single chip fabrication facility can use upwards of 37 million litres of ultrapure water per day.

Critical Materials and E-Waste

AI isn’t just energy and water intensive. It’s also resource hungry. The chips and systems powering AI depend on critical raw materials like gallium, indium, and rare earth elements. Mining these materials creates environmental damage, destroying ecosystems, releasing toxins, and consuming more water and energy.

Once these components are outdated, they often end up as e-waste. In the UK, we already produce nearly 24kg of e-waste per person every year, the second-highest rate globally. And AI-driven infrastructure is a major contributor. One study estimates that generative AI systems could be responsible for generating 1.2 to 5 million tonnes of e-waste between 2020 and 2030.

So, What Do We Do?

The NEPC’s report lays out five critical steps that can help governments, businesses and institutions steer AI in the right direction.

  1. Mandate Environmental Reporting

We need hard data on AI’s environmental cost. That means requiring developers and data centres to report energy and water use, carbon emissions, and hardware reuse. Without transparency, we can’t manage what we don’t measure.

  1. Bridge the Information Gap

Most users and enterprises don’t know how much their AI usage is costing the planet. Developers should be required to disclose the environmental footprint of their models (through model cards and transparency frameworks), and sustainability should be embedded into digital education across schools, universities, and corporate training.

  1. Set Green Standards for Data Centres

New and existing data centres should meet minimum sustainability standards: zero potable water for cooling, 100% carbon-free energy backed by auditable certificates, waste heat reuse, and ambitious targets for hardware reuse and recycling.

  1. Smarter Data Practices

We must incentivise leaner data collection, transmission, and storage. This means revisiting data retention policies, embedding best practice in AI procurement, and encouraging model development that balances capability with resource use.

  1. Government Must Lead

The UK government can lead by embedding sustainability into all AI policy and investment, particularly in procurement, research funding, and AI Growth Zones. Public funding should prioritise smaller, task-specific models and edge computing that reduces data centre loads.

An Opportunity for UK Leadership

The UK is well-placed to lead the global conversation on AI sustainability. With strong academic expertise, a world-class AI assurance market, and a net zero 2050 target, we have both the responsibility and the potential to act.

Smaller, purpose-built AI models, tailored to specific tasks, are already showing comparable performance to their larger cousins, while consuming far less power and requiring less data. Businesses, especially SMEs, stand to benefit too, with lower operational costs and reduced risk of ‘technical debt’.

Embedding sustainability into AI is no longer optional. It’s essential. And the decisions we make today will lock in consequences, good or bad, for decades.

So…

AI isn’t inherently bad for the planet. But its unchecked growth could undermine the very climate goals it promises to support.

This report is a wake-up call and an opportunity. If we build responsibly, transparently, and sustainably, AI can truly help us solve humanity’s biggest challenges.

But, of course… only if we act now.

 

Alan Stenson, CEO

Neutral Carbon Zone

 

Related Blogs

Speak to our team

Start integrating sustainability now into your business, one step at a time with independent third party verification from NCZ.