The E-Waste Column no. 204

Jul 1
3 min read

Today, we are diving into the findings of a recent UN report on the environmental impacts of artificial intelligence (AI).

🌱 What does the UN report say?

On 3 June 2026, the United Nations University Institute for Water, Environment and Health (UNU-INWEH) published a report titled “Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints”. According to the report, “AI's environmental cost [has been] systematically mismeasured” because the focus has been incorrectly placed on the carbon emissions from training AI models. The researchers who wrote the UN report therefore instead looked to quantify the impacts of every kilowatt-hour of electricity used by AI. This allowed them to better understand the carbon, water, and land footprints of AI's electricity use across the globe.

🌱 How much energy do the datacenters powering AI use?

It is estimated that 2.5 billion ChatGPT prompts are used daily, which collectively consume roughly 383 GWh of electricity per year. In 2025, datacenters globally “consumed an estimated 448 terawatt-hours of electricity”, making them “the world’s 11th largest electricity consumer” if they were treated as a nation. By 2030, the projected global electricity demand for datacenters is 945 TWh. This is roughly twice the amount of energy than France consumed in 2025 and amounts to nearly 3% of the world’s projected electricity use for 2030. In this context, it is worth noting that 80 to 90% of the total energy consumed through AI is used for “inference” or running deployed models through the day-to-day use of AI – rather than for training AI models. It is also worth noting that producing a typical AI-generated image uses 1450 times more energy than a basic text classification uses, and generating video content requires even more energy. According to the UN report, “[t]he energy required to generate a typical AI image is enough to power a 10-watt LED bulb for 17 minutes, and the energy required for a high-complexity AI video is sufficient to run that same bulb for 42 hours”.

🌱 How much carbon do the datacenters powering AI release?

399 million tonnes of carbon are expected to be released alone through the electricity consumption of the datacenters powering AI in 2030. To offset this carbon footprint “would require approximately 6.7 billion trees grown over 10 years”. This is “roughly twice the estimated number of trees [currently found] in the United Kingdom”.

🌱 How much water do the datacenters powering AI use?

9.3 trillion litres of water are expected to be used for the cooling of the datacenters that power AI in 2030. This is “equal to the basic annual domestic water needs of 1.3 billion people in Sub-Saharan Africa”. It is worth noting the datacenters powering AI sometimes operate “amid drought conditions”, meaning that the water use has a real and immediate impact on the communities and ecosystems in the area. According to the report, “the electricity-associated water footprint is about two tablespoons (29 mL) for a single [AI-generated] image, but jumps to 4.1 liters for a complex video—almost equivalent to a two-day drinking water need for one person”.

🌱 What does the land footprint of the datacenters powering AI look like?

By 2030, the land footprint associated with the electricity needed to power all the datacenters globally for AI use is 14 500 km². This is “about twice the Jakarta metropolitan area, home to more than 32 million people”.

🌱 How much e-waste do the datacenters powering AI produce?

It is projected that by 2030 2.5 million tonnes of e-waste will be produced annually in relation to AI. This is “equivalent to discarding [the weight of] nearly 250 Eiffel Towers each year”.

🌱 What are the researchers calling for now?

The authors of the report are calling for responsible AI use. They say “urgent action [is needed] to ensure that the technology develops within planetary limits”. They also say that it is key that we ensure “that the communities who provide the critical minerals for advancing AI and the ones that host its infrastructure and e-waste are also among those who benefit from it”.

💡 In next week’s column, we will be taking a closer look at the researchers' recommendations to build a more “responsible AI ecosystem” globally – so stay tuned.