What the UN report says: jaw-dropping numbers

The Institute for Water, Environment and Health at the United Nations University (UNU-INWEH) published on June 3, 2026 a comprehensive report The Environmental Cost of Artificial Intelligence: Carbon, Water, and Land Footprints. And its conclusions are not pleasant reading for anyone hoping that technological progress alone would solve AI's environmental footprint.

Key figures from the report:

• Energy: By 2030, AI-related electricity consumption could double from current levels and reach 3% of global electricity production. For comparison — that's roughly as much as all of Japan consumes today.
• Emissions: The carbon footprint of AI infrastructure could match the emissions of the entire United Kingdom — approximately 300 to 400 million tonnes of CO₂ annually.
• Water: Cooling data centers could require more water than the total global annual drinking water consumption. Already today, some large data centers consume as much water daily as a city of 100,000 residents.

The report also highlights the impacts of raw material extraction for chips and batteries — from lithium to rare earth elements — which burden ecosystems in developing countries and often escape attention in debates about "clean" AI.

Jevons paradox: Why more efficient AI doesn't mean savings

We often hear the argument that new AI models are more energy-efficient, so overall resource consumption will decline. The UN report, however, convincingly shows that this is a misconception — and explains why using the Jevons paradox.

This economic principle, named after British mathematician William Stanley Jevons, states that when technological progress increases the efficiency of using a resource, its total consumption does not fall, but instead rises. Jevons observed this in 19th century England: the invention of a more efficient steam engine did not reduce coal consumption — quite the opposite, coal became cheaper and started being used everywhere.

With AI, the mechanism is the same. When a model becomes cheaper and more accessible, new use cases emerge that no one had thought of before. Billions of queries per day, AI agents running in the background, generating videos and images for social media — all of this increases total consumption faster than the efficiency of individual computations can grow.

The report's authors therefore warn: relying on the idea that "technology will solve it" is a dangerous illusion. Without targeted regulation and transparent reporting, AI's environmental footprint will continue to grow regardless of technical innovations in chips and cooling.

What Europe and Czechia are doing about it

The European Union is already responding to the situation. The EU AI Act, which entered into force in 2024 and is gradually becoming effective, introduces transparency requirements including the energy intensity of high-risk AI systems. From 2026, the scope of the Energy Efficiency Directive (EED) is expanding, requiring data center operators in the EU to report key indicators — including energy and water consumption and the share of renewable sources.

For Czechia, the topic is relevant for several reasons. The European Czech AI Factory hub is being established in Ostrava, which will operate computing infrastructure for AI. And major players like Google and Microsoft are considering building data centers in Central Europe — partly due to relatively cheap electricity and available water. The question of how much water and energy these centers will consume and whether the region will get adequate value in return will be a key political topic in the coming years.

Companies promise "green AI", but the numbers tell a different story

Tech giants are racing to make green promises. Google claims it will run on carbon-free energy 24/7 from 2030. Microsoft promises carbon negativity by 2030. Amazon targets 2040. The reality, however, is less optimistic.

According to available data, Google's emissions have grown by 48% since 2019 — primarily due to expanding AI infrastructure. Microsoft has seen a 29% increase in emissions since 2020. Water consumption in Google's data centers increased by 17% between 2019 and 2023. And according to Microsoft itself, training GPT-3 in its data centers consumed 700,000 liters of water.

The problem is that data centers need enormous amounts of water not only for direct server cooling, but also indirectly — the power plants that feed them themselves consume water for cooling. In water-stressed areas such as the southwestern US or Southern Europe, this creates direct competition between AI and agriculture or drinking water for residents.

What to do about it: The UN's plan for responsible AI

The UN report doesn't just bring warnings — it also offers a concrete plan for the responsible use of AI, built on three pillars:

Transparency. Operators should be required to report energy consumption, water usage, and emissions — at the level of individual models, not just entire data centers. Without detailed data, the problem cannot be effectively addressed.

Equity. The environmental impacts of AI are geographically uneven. While the benefits of AI models are primarily enjoyed by wealthy countries, the negative impacts — raw material extraction, water consumption, emissions — often fall on developing regions. The UN calls for a fairer distribution of costs and benefits.

Sustainability. Innovation in AI must go hand in hand with sustainable resource use. This means investments in renewable energy for data centers, development of more efficient cooling systems, and thorough environmental impact assessments before approving new projects.

For the average user, the lesson is simple: every interaction with AI has a real environmental footprint. That doesn't mean we should stop using ChatGPT. But it's worth thinking about whether we really need to generate ten variants of a company logo or let AI run in the background as a "second monitor" when it's not necessary.