Artificial intelligence (AI) is often discussed in terms of innovation, productivity, and economic growth. But a new United Nations University (UNU) report warns that its physical footprint is becoming comparable to that of entire countries.
The report, “Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints”, finds that global data centers powering AI could consume 945 terawatt-hours (TWh) of electricity by 2030. That is nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria—countries with more than 650 million people.
- If AI data centers were a country, they would rank among the top six electricity consumers in the world by 2030.
The report argues that the impact of AI cannot be measured by carbon emissions alone. It also includes large and growing demands for water and land. These pressures are reshaping how governments and companies think about digital infrastructure.
Kaveh Madani, the Director of UNU-INWEH and the professor who led the investigation, remarked:
“This report is not a case against artificial intelligence, a technological transformation that is improving the lives of billions of people around the world. It is a call for using it responsibly and addressing its unintended impacts proactively to make it sustainable and equitable. We have a narrow window to ensure that the backbone of the technological revolution of our era develops within planetary limits, and 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.”
Data Centers Already Consume Nation-Level Resources
The scale of AI infrastructure is already large today, and it is growing quickly.
In 2025, global data centers consumed about 448 TWh of electricity. That is more than the total electricity use of countries like Saudi Arabia. They also produced about 189 million metric tons of CO₂ emissions, similar to the annual emissions of Argentina.
The report finds that AI currently accounts for about 20% of total data center energy use, but this could rise to 40% by 2030 as AI applications expand. Goldman Sachs predicts that data center power use will climb by over 160% by the same period.
This shift is driven mainly by “inference,” which is the continuous use of AI systems after they are trained. The report estimates that inference accounts for 80–90% of total AI energy consumption, far more than model training.
A single widely used system shows the scale. ChatGPT processes around 2.5 billion prompts per day, which translates to about 383 GWh of electricity per year for one application alone.
The report highlights a key trend: AI is no longer a training problem. It is a continuous global electricity demand system.
The Triple Burden: Energy, Water, and Land Under Pressure
One of the report’s central findings is that AI’s environmental cost is multi-dimensional. It is not just about carbon emissions. By 2030, data centers are projected to use:
- 945 TWh of electricity,
- 9.3 trillion liters of water, and
- Over 14,500 square kilometers of land.
The water footprint alone is equal to the basic annual needs of 1.3 billion people in Sub-Saharan Africa. The land footprint is roughly twice the size of the Jakarta metropolitan area, a region home to more than 32 million people.
These impacts come from cooling systems, power generation, and infrastructure build-out. The report warns that focusing only on carbon can hide trade-offs. For example, switching to some low-carbon energy sources can reduce emissions but increase water and land use.
This creates a more complex challenge, according to the report. “Low-carbon” does not always mean “low-impact.” Dr. Miriam Aczel, the lead author, stated:
“What surprised us most is how often the choices that look greenest from a carbon perspective end up worse for water or for land. If we keep judging AI sustainability by carbon alone, we might think that renewables make AI infrastructure clean, but that is solving one problem while creating other problems, often in places that didn’t ask for it.”
Efficiency Gains Are Being Outpaced by AI Growth
AI systems are becoming more efficient, but demand is growing even faster. A typical AI image query can use about 1,450 times more energy than a basic text classification task. A single AI video can consume as much electricity as 200,000 simple queries.
Even small design choices matter. The report notes that changing output length, resolution, or model type can significantly alter energy use per request.
However, efficiency gains are often offset by rising usage. This is known as the rebound effect. As AI becomes cheaper and faster, people use it more frequently.
The report warns that this trend could cancel out many efficiency improvements unless stronger limits or design rules are introduced. It also highlights a growing environmental justice issue.
Only 32 countries host AI-specialized data centers, and more than 90% of global capacity is concentrated in just two countries. More than 150 countries have little or no access to AI computing infrastructure, even as they bear environmental costs linked to mineral extraction and e-waste.
The top 20 data center hubs and their distribution are as follows:
Local Resources Are Under Pressure From Global AI Demand
The environmental impact of AI is not evenly distributed. The report shows that data centers can place heavy pressure on local water and electricity systems.
In Ireland, they already account for 21% of total metered electricity use, exceeding household consumption in some regions. Authorities have paused new approvals in parts of Dublin until 2028 due to grid constraints.
In other regions, the pressure is even more direct. In Mexico and Uruguay, data center expansion has coincided with severe drought conditions, raising concerns about water availability for local communities.
The report also warns about downstream impacts. AI infrastructure could generate up to 2.5 million tons of electronic waste per year by 2030, much of which may be processed in countries with weaker environmental protections.
This creates a mismatch. The benefits of AI are global, but many environmental costs are local.
A Call for Multi-Factor AI Governance
The UN report does not call for slowing AI development. Instead, it calls for better governance and measurement, as Professor Madani said.
It argues that current environmental reporting is incomplete because it focuses mainly on carbon emissions. The report recommends tracking carbon, water, and land footprints together.
It also proposes several actions, including:
- Governments should include AI infrastructure in energy and water planning.
- Companies should design models for efficiency, not just performance.
- Data centers should consider local environmental limits when choosing locations.
- Investors should treat resource use as a financial risk factor.
- Users should be encouraged to reduce unnecessary computing load.
The key message is that AI must be built within planetary limits.
AI Infrastructure Is Becoming a Global Resource System
The report concludes that AI is no longer just a digital technology. It is becoming a physical infrastructure system that consumes electricity, water, land, and minerals at a national scale.
By 2030, AI data centers could use as much electricity as some of the world’s largest countries combined. At the same time, they could require water equivalent to billions of people’s needs and generate large volumes of electronic waste.
The UN framing is straightforward. The question is no longer whether AI will grow. It already is. The real challenge is whether that growth can be managed in a way that stays within environmental limits and distributes both benefits and burdens more fairly across countries and communities.
