Data centers running artificial intelligence already consume as much electricity as Saudi Arabia, the world’s 11th largest electricity consumer. By 2030, that figure could double, reaching 945 terawatt-hours of electricity per year, or roughly 3 percent of the entire planet’s projected energy use. The carbon footprint of that electricity would require 6.7 billion trees grown over a decade to offset.
These are the central findings of a new report from the United Nations University Institute for Water, Environment and Health, published June 3. The report is the first comprehensive attempt by a UN body to quantify the full environmental cost of AI across four dimensions: energy, water, land, and material resources.
The numbers are sobering, but the report’s deeper argument is not that AI should be abandoned. It is that the current trajectory is unsustainable and that the standard solution, making AI more efficient, will not solve the problem on its own.
The Jevons paradox
The report explicitly invokes an economic principle from the 19th century: the Jevons paradox, named after the English economist William Stanley Jevons. In 1865, Jevons observed that improvements in the efficiency of coal-fired steam engines did not reduce Britain’s total coal consumption. Instead, cheaper steam power enabled new industrial uses, and total coal demand rose.
The same dynamic, the UN report argues, is now at work with AI. As models become cheaper and more efficient to run, they become economically viable for more tasks, attracting more users and driving higher overall energy consumption. Efficiency gains may slow the rate of growth, but they are unlikely to reverse it. Under the report’s central projection, AI’s energy use doubles by 2030 regardless of anticipated efficiency improvements.
This is a fundamental challenge to the narrative that better technology will solve the environmental problem created by new technology.
The scale: energy, water, land
The report’s headline numbers are built from primary data aggregated across multiple sources, including the International Energy Agency, the International Telecommunication Union, and direct industry reporting.
On energy, data center electricity consumption was approximately 415 TWh in 2024, representing about 1.5 percent of global electricity use. The projection of 945 TWh by 2030, roughly 3 percent, represents more than a doubling in six years. For context, the entire country of Japan consumes roughly 900 TWh annually.
On water, the report estimates that data centers could consume 9.3 trillion liters of water per year by 2030, predominantly for evaporative cooling of server racks. That volume exceeds the projected annual drinking water needs of the entire global population. The water footprint is concentrated in regions where data centers are clustered: the southwestern United States, northern Virginia, Singapore, and parts of Europe, many of which already face water stress.
On land, the physical footprint of AI infrastructure is projected to reach nearly 10 times the area of Mexico City by 2030, driven not only by data centers themselves but by the associated mining, manufacturing, and energy generation facilities that support them.
A divided world
The environmental burden is not distributed evenly. The report finds that only 32 countries currently host AI-specific cloud infrastructure, and 90 percent of that capacity is concentrated in the United States and China. Countries that consume AI services but do not host the infrastructure bear none of the direct environmental cost of running models but may bear a disproportionate share of the upstream burden, including mineral extraction for hardware manufacturing and the disposal of electronic waste.
This structural inequity, the report warns, risks widening the digital divide. Nations that build and control AI systems capture the economic benefits while the environmental costs are externalized to regions with weaker regulatory frameworks and less bargaining power.
Responsible AI: what it would look like
The report lays out a framework for what it calls “responsible AI,” built around six principles: transparency, efficiency by design, equity and justice, lifecycle responsibility, global cooperation, and sustainable use.
At the operational level, this means making environmental disclosures a routine part of AI development. The report calls for standardized reporting of energy consumption, carbon emissions, water use, and material sourcing at both the model level and the individual task level, so that users can compare the environmental cost of different AI systems the way they compare the fuel economy of vehicles.
It also calls for integrating projected AI energy demand into national climate and energy planning frameworks, which currently do not account for it. Many countries, the report notes, have national AI strategies that promote adoption and innovation but contain no requirement for environmental reporting and no regulatory mechanism for enforcement.
The report is not an argument against AI development. Its authors emphasize that AI has significant potential for climate modeling, energy grid optimization, and environmental monitoring. But they argue that the current approach, in which capability is pursued with little regard for environmental cost, is not sustainable and that waiting for efficiency improvements to solve the problem is a misreading of economic history.
Source: United Nations University Institute for Water, Environment and Health (UNU-INWEH). “Environmental Cost of AI’s,” June 2026.