As artificial intelligence reshapes industries, economies and everyday life, a new concern is emerging beneath the excitement surrounding the technology’s rapid growth. While AI is increasingly viewed as a catalyst for innovation and productivity, a recent United Nations report asserts that the infrastructure powering this digital revolution may come at a significant environmental cost.
According to the report, the world’s data centres, the digital backbone supporting AI systems, are on track to consume 945 terawatt-hours (TWh) of electricity annually by 2030. The projected figure is nearly three times the current combined annual electricity consumption of Nigeria, Pakistan and Bangladesh, three countries with a combined population exceeding 650 million people.
The findings, contained in a report titled “Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints,” released by the United Nations University Institute for Water, Environment and Health (UNU-INWEH), highlight the growing pressure AI infrastructure is placing on global energy, water and land resources.
The report arrives at a time when governments and businesses worldwide are racing to expand AI capabilities, invest in large-scale computing infrastructure and position themselves competitively in an increasingly digital economy. Yet researchers argue that the environmental implications of this expansion remain significantly underestimated.
For years, discussions around the environmental impact of AI have largely centred on greenhouse gas emissions generated by data centres. However, the UN researchers contend that focusing solely on carbon emissions presents an incomplete picture.
The report argues that every unit of electricity consumed by AI systems carries three interconnected environmental burdens namely carbon emissions, water consumption and land use.
According to the researchers, these impacts do not necessarily move in the same direction. Measures designed to reduce carbon emissions can sometimes increase water demand or land requirements.
For example, shifting electricity generation from coal to bioenergy may reduce carbon emissions by approximately 70 percent. However, such a transition could simultaneously increase water consumption more than thirtyfold and expand land requirements by as much as one hundred times.
The report warns that evaluating AI sustainability using only carbon metrics risks obscuring environmental trade-offs and transferring pressures to regions already experiencing water scarcity or land-use challenges.
“Low-carbon does not automatically mean low-water or low-land,” the report noted, urging policymakers to adopt a more comprehensive framework for measuring AI’s environmental footprint.
Data centres becoming electricity giants
The scale of energy demand associated with AI infrastructure is already becoming significant.
Researchers estimate that global data centres consumed approximately 448 TWh of electricity in 2025. If considered a country, data centres would rank as the world’s eleventh-largest electricity consumer, surpassing nations such as Saudi Arabia and trailing only major industrial economies.
By 2030, electricity demand from these facilities is expected to more than double.
The accompanying environmental impacts are equally substantial. The report projects that by the end of the decade, data centres could require about 9.3 trillion litres of water annually while occupying more than 14,500 square kilometres of land.
To put this into perspective, researchers estimate that the projected water footprint would equal the minimum domestic water requirements of all 1.3 billion people living across sub-Saharan Africa for an entire year.
The hidden cost of everyday AI use
One of the report’s most significant findings challenges a widely held assumption about where AI consumes most of its energy.
Public attention has traditionally focused on the enormous computational resources required to train advanced AI models. Training OpenAI’s GPT-3 model, for example, was estimated to require approximately 1.3 gigawatt-hours of electricity, while GPT-4 reportedly consumed between 50 and 70 GWh during training.
However, researchers say that training is only part of the story.Once deployed, AI models consume far more energy through “inference”—the process of responding to user requests in real time.
According to the report, inference accounts for between 80 and 90 percent of an AI model’s lifetime energy consumption.
With billions of users increasingly relying on AI tools for writing, searching, coding, image generation and business operations, energy consumption is accumulating at an unprecedented pace.
ChatGPT alone is estimated to process around 2.5 billion prompts daily. Researchers estimate that servicing those requests requires roughly 383 GWh of electricity annually.
The report calculates that offsetting the associated carbon emissions would require approximately 2.6 million tree seedlings grown over a decade, occupying a land area roughly equivalent to Manhattan.
Its water footprint, meanwhile, would be enough to meet the minimum annual domestic water needs of approximately half a million people in sub-Saharan Africa.
AI-generated images and videos emerge as new pressure points
The environmental burden becomes even more pronounced when users move beyond text-based interactions.
The UN report found that energy requirements vary depending on the type of AI task being performed.
A standard conversational AI query may consume around 200 times more energy than a simple text-classification task. Generating an AI image can require approximately 1,450 times that baseline amount.
Video generation represents an even greater challenge.According to the report, producing a short AI-generated video can consume as much electricity as processing 200,000 spam classification tasks.
The researchers noted that most users remain unaware of these environmental costs because decisions affecting energy use such as model selection, prompt length, image quality and video resolution are often embedded within default system settings.
The report estimates that generating a typical AI image consumes enough electricity to power a 10-watt LED bulb for 17 minutes.
A complex AI-generated video, by comparison, could power the same bulb for approximately 42 hours.
Water consumption follows a similar pattern. While a single AI-generated image may require around 29 millilitres of water through associated electricity production and cooling processes, a high-complexity video can require more than four litres—equivalent to nearly two days of drinking water for one individual.
Efficiency gains may not solve the problem
The report also challenges the notion that technological improvements alone will reduce AI’s environmental impact.
Researchers cite the “rebound effect,” often referred to as the Jevons Paradox, whereby efficiency improvements lower costs and encourage greater usage.
As AI systems become more efficient and affordable, demand tends to rise faster than efficiency gains can compensate.
Kaveh Madani, a co-author of the report, explained that while technological advancements can reduce the environmental footprint of individual AI tasks, overall consumption continues to increase as more users adopt AI-powered applications.
“A lot of people think that the environmental footprint of AI reduces as technology improves and processes become more efficient. But that is only a partial picture of the overall problem.More efficient and affordable AI and energy mean more consumption of AI, making the overall footprint far bigger than what we save through efficiency gains,” Madani said.
Growing concerns over environmental inequality
Beyond energy and water consumption, the report highlights what it describes as a widening environmental justice issue.
While AI infrastructure brings significant economic, technological and strategic advantages, access to these benefits remains highly concentrated.
Researchers found that only 32 countries currently host AI-specialised data centres, while more than 150 countries have little or no sovereign AI computing capability. More than 90 percent of global AI computing capacity is concentrated in just two countries.
This imbalance, the report argues, means that many developing nations bear the environmental costs associated with AI including critical mineral extraction and electronic waste processing,without receiving comparable economic or technological benefits.
The researchers estimate that AI infrastructure could generate as much as 2.5 million tonnes of electronic waste annually by 2030.
Much of this waste is expected to be processed in lower-income economies that often lack adequate environmental safeguards and waste-management systems.
Similarly, many of the critical minerals required to manufacture AI hardware are sourced from regions where environmental regulations remain weak, exposing local communities to environmental degradation while the economic gains are realised elsewhere.
Call for a responsible AI ecosystem
Against this backdrop, the United Nations is calling for a more balanced and sustainable approach to AI development.
The report proposes a six-principle framework centred on transparency, efficiency by design, environmental justice, lifecycle responsibility, global cooperation and sustainable use.
Governments are encouraged to incorporate AI infrastructure planning into broader energy, water and land-use policies while mandating standardised environmental reporting.
Technology companies and AI developers are urged to design systems with environmental impacts in mind, ensuring that model selection and default outputs reflect sustainability considerations.
The report also calls on investors, utilities, data centre operators and local communities to play more active roles in assessing and managing environmental risks associated with AI expansion.
For countries such as Nigeria, where electricity shortages remain a persistent challenge and digital transformation ambitions continue to grow, the findings offer an important reminder that the future of artificial intelligence is not solely about computing power and innovation. It is also about how the world manages the resources required to sustain that future.
As nations pursue the economic opportunities presented by AI, the report suggests that balancing technological progress with environmental sustainability may become one of the defining policy challenges of the coming decade.
