UN University Report warns of surging environmental burden from AI expansion, stating, "mitigation measures are urgently needed"

The advancement of Artificial Intelligence (AI) has been remarkable. In particular, the expanding utilization of generative AI is rapidly transforming our daily lives and working styles. While some view this transformation as an "AI Revolution" comparable in scale to the Industrial Revolution of the late 18th century, the massive electricity consumption of data centers required to support this vast information processing has emerged as a new critical challenge. The International Energy Agency (IEA) projects that global electricity demand from data centers will more than double by 2030.

Against this backdrop, the United Nations University Institute for Water, Environment and Health (UNU-INWEH) has compiled a report highlighting that this surge will increase not only greenhouse gas emissions from power generation but also water and land footprints. The report emphasizes the importance of implementing environmental mitigation strategies and outlines ways to utilize generative AI more efficiently to save energy. It underscores the need to look beyond AI's evolution and convenience and closely examine its often-overlooked environmental impact.

Global electricity demand rising 4% annually

In 2025, the IEA released two pivotal reports detailing the rapid proliferation of AI and its impact on global power demand. The first, titled "Electricity 2025" and published on February 14, estimated that global electricity demand will grow at an annual rate of approximately 4% through 2027. It predicted that the total increase over the three years from 2025 will reach about 3,500 terawatt-hours (terawatt is 1 trillion, and 1 terawatt-hour is 1 billion kilowatt-hours) in total, and the electricity demand in 2027 will be 32,542 terawatt-hours.

The report attributes this trend to the widespread adoption of AI and Electric Vehicles (EVs), alongside the expansion of 5G communications infrastructure, electrified industrial equipment, and a surge in air conditioning needs. These factors indicate the arrival of an "Age of Electricity," where societies worldwide require significantly more power. For example, in China, where power demand is surging, consumption is being driven by the manufacturing of EVs, solar panels, batteries, and semiconductors, alongside the operational needs of air conditioning units and data centers.

The IEA estimates that 85% of this additional electricity demand will originate from emerging economies such as China and India. Meanwhile, developed countries, including Japan, are expected to account for 15% of the total growth. While demand in developed nations has stagnated in recent years due to energy efficiency initiatives, it is on track to expand once again. The increase in the United States is primarily driven by the data center requirements of a booming AI sector.

The increase in demand through 2027 is expected to be met almost entirely by "low-carbon power sources." This supply growth will be driven by the expansion of renewables like solar and wind, alongside a resurgence of nuclear power in countries such as France. The report noted that by 2025, renewable energy will surpass coal-fired power, and the share of coal-fired power will fall below 33% for the first time in the 21st century. However, it also pointed out that it will still take time for China and India to achieve full decarbonization.

Electricity supporting the "AI Revolution"

The second IEA report, "Energy and AI," was released on April 10, 2025. It drew attention as the first comprehensive analysis of the nexus between AI and energy. The beginning of the report pointed out that "AI is emerging as a general-purpose technology like electricity once did," and "AI, which was an object of academic research, has transformed into a huge industry accompanied by trillions of dollars in market value and venture investment in the past few years."

Immense computing power is necessary for the training and operation of AI models. For that reason, data centers that aggregate many servers and house data storage devices and network equipment are becoming essential. The report predicted that the electricity demand of data centers will more than double by 2030 to become 945 terawatt-hours.

The United States accounts for most of this increase, ranking top. China follows behind. In the United States, it has been pointed out that data centers account for half of the increase in electricity demand up to 2030. It is said that a scenario where the world's data center electricity demand in 2035 reaches about 1,200 terawatt-hours, mainly in various advanced countries including the United States, is influential.

In addition, it stated that behind the rapid proliferation of AI, there are dramatic improvements in the performance of supercomputers, an explosive increase in data volume, and innovations in algorithms. It explicitly describes that AI has entered the stage of societal implementation especially with the appearance of Large Language Models (LLMs). It can be said that the "AI Revolution" is already progressing, and the demand for electricity supporting it also increases.

According to the report, for example, generating text consumes about 1 to 2 watt-hours, advanced reasoning models consume several watt-hours, and generating a short video consumes dozens of watt-hours of electricity. Along with the expansion of such use, it pointed out that one large-scale data center consumes electricity equivalent to about 100,000 households, and the power consumption of ultra-large centers under construction in advanced countries will become up to 2 million households.

The feature of this report is that it showed a perspective that the AI revolution is not merely a digital revolution but also leads to an electricity revolution. It emphasized that AI and energy have an inseparable relationship, stating that how much AI develops and progresses in the future depends on infrastructure policies such as power grids and power generation equipment, in addition to new research and development of semiconductors.

Problems of greenhouse gases, water, and land, too

Many AI experts both inside and outside the country cite the two IEA reports, and opportunities where the use of AI is discussed in connection with energy and environmental problems are increasing. Under such circumstances, UNU-INWEH released a report on June 3 titled "Environmental Cost of Artificial Intelligence: Carbon, Water, and Land Footprints," sounding an alarm that the environmental impacts of AI are overlooked.

The feature of this report is its emphasis that the problems with AI and energy/environment are not limited to electricity consumption. In addition, we must also not forget greenhouse gases emitted during power generation, the use of water and land supporting the huge infrastructure necessary for the explosive expansion of AI, and the problem of waste.

The report also addresses the electricity consumption of AI and data centers. According to the report, even at the stage of 2025, the electricity consumption was an estimated 448 terawatt-hours (448 billion kilowatt-hours), which is comparable to the annual consumption of France. It is also predicted to exceed 945 terawatt-hours in 2030 (the same value as the IEA's prediction). On top of that, it estimates that greenhouse gases emitted by power generation rise to about 400 million tons of CO₂ equivalent. This is said to correspond to the annual emissions of the entire United Kingdom.

It stated that water used to support AI-related electricity demand will become 9.3 trillion liters. This corresponds to nearly six times Tokyo's annual water supply. In addition, it is said that in 2030, land needed for power generation equipment and the like will be more than 14,000 square kilometers. It is almost the same as the area of Northern Ireland. It also pointed out that AI-related infrastructure has the potential to generate up to 2.5 million tonnes of electronic waste annually by 2030.

How to save energy with generative AI

Generative AI does not respond from within the user's smartphone or PC. In response to requests from terminals, server groups in huge data centers set up by respective businesses in various places around the world do vast processing and reply to the terminals. Every time the AI of your terminal answers your question, a server equipped with GPUs is operating in a data center. Naturally, it uses electricity.

What attracts attention in the report of UNU-INWEH is the point that reducing unnecessary inputs to generative AI also leads to energy saving, and as an example, it calculates the energy-saving effect when additional wording such as "thank you" or "please" is reduced. It stated that the interactive generative AI "ChatGPT" is used an estimated 2.5 billion times a day in the world, and assuming the electricity consumption per time is 0.42 watt-hours, 383 gigawatt-hours of electricity is consumed annually.

It is said that just configuring ChatGPT to answer concisely leads to a reduction of 87 to 98 gigawatt-hours annually. This is equivalent to the amount of electricity that about 760,000 people living in sub-Saharan Africa use in one year. Even when one wants to make interactions with generative AI polite, it seems better to think "efforts to reduce unnecessary wording and more-than-necessary interactions as much as possible lead to the mitigation of environmental burden."

Upon the publication of the report, Director Kaveh Madani of UNU-INWEH, commented, "This is a report showing the UN's determination to ensure that technological progress improves human well-being while respecting environmental limits."

There is also a view that AI "worsens the environment" because it increases electricity and energy consumption. However, slowing down the rapid expansion of AI is not realistic. Therefore, we should harness the power of AI to reduce environmental impact and implement environmental measures. For example, we can fully utilize AI to operate renewable energy, adjust the amount of power generation that fluctuates depending on the weather by analyzing weather data, and optimize the operation of power generation equipment and power grids. While the global environment has limits, borrowing the miraculous power of AI for the energy saving of future society makes sense.