A row of servers in Google’s data centre at Douglas County, Georgia, in the US. Photo from Google Gallery.

A new report from the United Nations University Institute for Water, Environment and Health (UNU-INWEH) provides sobering statistics about the massive environmental costs of artificial intelligence.

The report, entitled ‘Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints’, states that every AI interaction draws on finite resources, and the total environmental footprint depends on how AI systems are designed, how often they are used, and what tasks they perform.

“Real progress depends on embedding sustainability at every level, from hardware and model design to deployment, governance, and public use,” the report’s authors conclude.

“Every kilowatt-hour of electricity used to train or run an AI model carries environmental footprints, including a carbon footprint from the generation mix; a water footprint from electricity production and cooling; and a land footprint from energy infrastructure, reservoirs, and fuel extraction.”

AI’s carbon footprint is primarily determined by multiplying the total energy required to train and run the model by the carbon intensity of the local electricity grid.

The carbon intensity of the power grid fluctuates drastically depending on the energy sources (the ‘generation mix’) used in the location where the data centre is operating

Training new AI models requires immense energy, the new report states.

No training energy data are yet available for ChatGPT-5.0, but, according to the new report, scaling up from GPT-3 and GPT-4 suggests that a next-generation model could plausibly require about 100 GWh (Gigawatt-hours), with associated footprints of roughly 42,000 tonnes of CO₂e (carbon dioxide equivalent), one billion litres of water, and a land footprint of 1.5 km².

“This level of electricity use is equivalent to the annual residential electricity consumption of approximately 770,000 people in Sub-Saharan Africa and the water footprint is sufficient to meet the minimum annual water needs of roughly 137,000 Sub-Saharan residents,” the new report states.

Offsetting a carbon footprint of 42,000 tonnes of CO₂e would require about 700,000 tree seedlings grown for ten years.

One GWh of electricity equals 1 billion Wh (watt-hours), 1 million kWh (kilowatt-hours), and 1,000 MWh (megawatt-hours). One TWh (Terawatt-hour) equals 1,000 GWh.

UN scientists warn that the footprint of AI’s daily use is far bigger than that from training new AI models.

ChatGPT alone is estimated to process around 2.5 billion prompts per day. At a conservative 0.42 Wh per text prompt, that translates into about 383 GWh of electricity per year. Offsetting the associated carbon emissions would require 2.6 million tree seedlings grown for 10 years.

The related annual water footprint would be equal to the minimum annual domestic water needs of some 500,000 people in Sub-Saharan Africa.

The new UN University report notes that Google processes an estimated 5 trillion searches annually and a conventional search uses about 0.3 Wh. An AI-enhanced generative search uses up to 3 Wh, a 10-fold increase.

“One of the most consequential dimensions of AI that remains comparatively under-examined is its environmental footprint and the justice implications that follow,” the UNU-INWEH report states.

It says that AI’s expansion involves “physical infrastructure and supply chains, including data centres, chips, electricity generation, cooling systems, water withdrawals, land occupation, critical minerals, and eventual e-waste”.

The Google data centre at The Dalles in the US state of Oregon. High-voltage transmission lines feed directly into on-site substations, converting grid power to the stable, continuous electricity required to run thousands of servers. Photo by Visitor7 (Wikimedia Commons).

Aerial view of the Google data centre at The Dalles. The operation of the centre relies on local water infrastructure, with usage increasing over time as the company has expanded its facilities. Data source: Airbus image, 2026.

NU-INWEH director and the report’s lead investigator Kaveh Madani said: “The future of artificial intelligence should not be measured only by what machines can do, but by whether humanity can deploy those capabilities within planetary boundaries.

“Though often described as weightless and virtual, the reality of AI is profoundly physical. Behind every prompt, image, or video lies a growing infrastructure of energy systems, water withdrawals, land use, mineral extraction, and electronic waste. This report is a call to make those hidden environmental costs visible before they become unmanageable.”

UN under-secretary-general, UN University rector, and co-author of the new report, Tshilidzi Marwala, said: “The promise of AI is immense, particularly in areas such as healthcare, education, scientific discovery, and climate resilience. But innovation without stewardship risks deepening inequality and intensifying pressure on already stressed planetary systems.”

According to the new report, expenditure on AI this year is projected to exceed USD 2.5 trillion and the global market is predicted to grow from USD 189 billion in 2023 to nearly USD 5 trillion by 2033, which is a 25-fold increase in less than a decade.

The new UN report says that, to understand the energy and environmental footprints of AI, we need to examine the infrastructure that enables it: data centres.

“These sprawling facilities – often housing thousands of high-performance processors – consume vast amounts of electricity to operate, cool, and transmit data across networks,” the report states.

Data centres are the engines of the digital age, powering an ever-expanding range of services, the report’s authors note.

These centres underpin nearly every aspect of modern life, the report states, and, increasingly, they also serve as the backbone of AI, supporting the training and deployment of large-scale models that power recommendation systems, generative chatbots, search engines, and autonomous technologies.

“As the global appetite for AI grows, so does the demand for energy-intensive infrastructure, driving rapid expansion in both the number and size of data centres, especially hyperscale sites that consume immense amounts of power,” the UN report states.

AI-related workloads accounted for roughly 20% of total data centre electricity use in 2025, it added. If that share rises to an expected 40% by 2030, AI-related electricity consumption could reach approximately 374 TWh.

Roof of a data centre showing cooling equipment and backup generators. These systems support temperature control and power continuity for high-density computing operations by managing heat and providing on-site electricity during outages. Photo by Rsparks3 (Wikimedia Commons).

An interior view of MareNostrum 5, a flagship European supercomputer, inside the historic chapel at the Barcelona Super¬computing Center in Spain. The system is among Europe’s most powerful high-performance computing installations, deliver¬ing around 314 petaflops of computing power.
MareNostrum 5 is being upgraded to expand Europe’s AI capacity, with next-generation GPUs, expanded storage, and energy-efficient cooling to support large-scale AI training and advanced machine learning for startups, researchers, and public institutions across Europe. Photo by Steve Jurvetson (Flickr).

The new report states that, if data centres were a country, their estimated 448 TWh (448 billion kWh) of electricity consumption in 2025 would rank them 11th globally, roughly on par with France.

The projected electricity consumption of the world’s data centres could exceed 945 TWh by 2030, the UN report says.

This would account for nearly 3% of projected global electricity use – enough to supply residential electricity to all 1.3 billion people in Sub-Saharan Africa for more than five years.

The carbon footprint of producing this much electricity averages about 399 million tonnes of CO₂e, which would require 6.7 billion trees grown over ten years to offset, the report adds.

The average water footprint of generating the required amount of electricity would total about 9.32 trillion litres, enough to satisfy the annual basic water needs of the entire population of Sub-Saharan Africa, while the associated land footprint would total about 14,500 km².

Depending on how the electricity was generated, associated emissions could reach 400 million tonnes of CO₂e, which is comparable to the UK’s emissions from all sectors in 2025.

The new report notes that the estimated 9.3 trillion litres of water used by data centres would meet the drinking water needs of Earth’s 8.1 billion people for about 1.6 years.

The report highlights how, in many cases, the withdrawal of huge amounts of water occurs in regions already facing drought or groundwater depletion, amplifying both environmental and social stress.

It says that, even when some withdrawn water is returned, “large-scale withdrawals can strain aquifers and river systems, particularly in arid or groundwater-depleted regions”.

Google’s Mesa data centre in Arizona, for example, holds a permit to use 5.5 million cubic meters of water annually.

In Querétaro, Mexico, expanding compute infrastructure is drawing on water supplies amid prolonged droughts. In Uruguay, plans for a water-intensive data centre coincided with a 2023 drought that depleted Montevideo’s freshwater reserves.

Protests in Montevideo, Uruguay, in May 2023, after severe drought depleted freshwater reserves and rising salinity made tap water unsafe to drink. Public anger intensified as plans for the Google data centre raised concerns that industrial demand was being prioritised over the population’s need for drinking water. Photo by Vivir.solo.cuesta.vida (Wikimedia Commons).

AI video and hardware waste

A single high-resolution AI video clip can require more than 415 Wh, making it more energy intensive than the creation of hundreds of AI images.

The new UN report notes that, when resolution and frame count are factored in, energy requirements rise quadratically (double the output quadruples the energy used).

“As video gets embedded in mainstream platforms, this quickly becomes an infrastructure-scale problem,” the UN report states.

The report also highlights the growing problem of AI hardware waste.

“At the end of life, poorly managed e-waste can expose frontline communities to hazardous substances,” the report states.

“By 2030, AI infrastructure could generate up to 2.5 million metric tons of e-waste each year …”

An uneven distribution of benefits and burdens

The minerals powering AI hardware are often extracted in ways that cause concentrated environmental and social harm, particularly in the global south and in regions with weak regulatory oversight, the UN report says.

“Frontier AI infrastructure is concentrated in a small number of locations. Countries that lack domestic compute capacity depend on external providers, giving them little control over access, pricing, or data governance,” the report notes.

“The result is a widening digital divide between nations that build and control AI systems and those that simply consume them while often bearing a disproportionate share of the environmental costs.”

‘Low-carbon does not equal low-impact’

The new UN University report points out that the carbon footprint of Brazil’s hydro grid is about 77% below the global average, but its water and land footprints are nearly triple the global mean.

The UK’s grid has a land footprint more than four times the global average.

The new report challenges the assumption that renewable-powered data centres are always green or sustainable. This finding goes against much of the current industry messaging.

The report underlines that efficiency gains alone will not reduce AI’s total environmental footprint. Lower costs drive higher volumes of use, potentially erasing all savings, it says. It calls for resource budgets, not just better hardware.

AI computing is concentrated in two countries

Aerial view of Alibaba’s Zhangbei data centre cluster in Hebei Province, China. Three adjacent sites developed between 2016 and 2025 form part of China’s Eastern Data Western Compute strategy. Data sources: Epoch AI; Sentinel-2 false-colour imagery, February 2026.

Only 32 nations host AI-specialised cloud infrastructure, and 90% of that capacity is in the US and China.

More than 150 countries have no sovereign AI computing at all.

The new report frames this not just as an economic divide, but as an environmental justice issue. It says that excluded countries bear mineral extraction and e-waste burdens while the strategic benefits flow elsewhere.

The report shows how the massive global expansion of AI is creating intense local pressures.

Ireland is cited as “a concrete, documented example of what happens when AI infrastructure growth outpaces energy planning – and a preview of what other countries are heading toward”.

By 2023, data centres accounted for 21% of Ireland’s total metered electricity, up from 5% in 2015 and exceeding all the urban household consumption combined. The national grid operator has paused new data centre approvals around Dublin until 2028.

Efficiency gains ‘will not reduce environmental footprint’

The UNU-INWEH report says efficiency gains alone will not reduce AI’s total environmental footprint.

“Lower costs drive higher volumes of use, potentially erasing all savings,” it states.

The report calls for “a responsible AI ecosystem built on six principles: transparency; efficiency by design; equity and environmental justice; lifecycle responsibility; global cooperation; and sustainable use”.

It makes the following recommendations:

• Governments should integrate AI infrastructure into energy planning, water governance, and land-use permitting, and require standardised environmental footprint reporting.

• Industry and AI developers should treat model selection, default outputs, and routing decisions as footprint determinants, and improve efficiency by design.

• Users and deploying organisations should adopt fit-for-purpose use, selecting the lightest model and lowest-energy format that meets the task.

• Investors should treat electricity, carbon, water, and land footprints as material risks in AI infrastructure portfolios.

• Communities and civil society should be involved early on in decisions about siting data centres and there should be enforceable transparency along with grievance mechanisms.

• International institutions should support harmonised measurement standards, reduce incentives for cross-border burden shifting, and build compute capacity in excluded regions.

The report says that even the language used by AI users can make a huge difference.

“Simply getting rid of politeness by not saying ‘please’ and ‘thank you’ can reduce the overall footprint significantly by making the prompts more concise,” it states.

“For example, a concise response mode can reduce ChatGPT token output by 30%, saving 87–98 GWh of electricity per year, equivalent to the annual residential electricity of nearly 760,000 people in Sub-Saharan Africa.”

The report reframes user behaviour and product design as environmental governance tools, not just convenience features.

“Technological advancement must remain environmentally manageable,” it states, adding that this requires measuring, disclosing, and acting on the full footprint, not just the carbon portion.

Aerial view of Microsoft’s data-centre complex in Midden¬meer in the Hollands Kroon municipality in the Netherlands. The complex consumed about 84 million liters of water in 2021 during a year of severe drought, far exceeding earlier estimates of 12–20 million litres, which has led to sustained opposition from local farmers over water use. Data Source: Sentinel-2 imagery, August 2025.

An increasingly polarised global workforce

The UN University report warns that, “without deliberate intervention, the global workforce could become increasingly polarised, divided by access to AI technologies and related workforce skills”.

It states: “Those with fewer training opportunities are especially vulnerable to the changes AI is bringing. While job disruption is a visible consequence of AI deployment, the technology’s influence extends far beyond the workplace, into realms of warfare, ethics, and even existential risk.”

It concludes that AI “offers remarkable potential, but fulfilling this promise responsibly requires systemic change”.

It adds: “By committing to transparency, engineering for efficiency, choosing wisely as users and institutions, protecting communities that face disproportionate burdens, and cooperating across borders, society can ensure that progress in intelligence is matched by progress in care.

“Responsible AI is possible when capability and stewardship grow together within planetary limits.”

Manager of UNU-INWEH’s Geospatial, Climate and Infrastructure Analytics Programme, and co-author of the new report Mir Matin said: “AI’s environmental footprint is not just an outcome of physical infrastructure; it is the cumulative result of countless daily decisions.

“Every prompt, default setting, generated image, video, and query accumulates when multiplied by billions of users and thousands of operators worldwide.

“Behaviour change across this entire decision chain – from individual users to corporate planners – is one of the most powerful and underused levers we have for keeping AI within planetary limits.”

AI-generated conceptual image of AI’s environmental footprint, generated using OpenAI’s ChatGPT/DALL·E, May 2026.
Estimated footprints per standard-resolution AI image: 2.9 Wh electricity, 1.22 g CO₂e, 28.6 mL of water, and 0.45 cm² of land, based on literature benchmarks and global average electricity footprint factors.

DONATE TO CHANGING TIMES VIA SIMPLE PAYMENTS

1= 5 euro, x 2 = 10 euro, X 3 =15 euro, etc.

€5.00