Study: AI Data Centers' High Energy Consumption May Worsen Local Environmental Problems

A new study by the United Nations University warns that the high consumption of water, electricity, and land by AI data centers could exacerbate environmental issues for local communities. Scholars suggest planning 'energy parks' in industrial zones to co-locate computing facilities, renewable energy, energy storage, and reclaimed water systems.

The United Nations University Institute for Water, Environment and Health released a report titled "Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints" on the evening of the 3rd, Taiwan time.

The report estimates that by 2030, global data center electricity consumption could exceed 945 billion kilowatt-hours, accounting for about 3% of global electricity use. The water footprint (direct and indirect water consumption) could be about 9.3 trillion liters, the land footprint (actual land area occupied) could exceed 14,500 square kilometers, and up to 2.5 million tons of e-waste could be generated.

The report points out that while training AI consumes electricity, the continuous 'inference' interactions between users and AI, billions of times daily, account for 80-90% of AI's energy consumption.

The report argues that assessing the environmental cost of AI should not only look at carbon emissions but also quantify carbon, water, and land footprints, and pay attention to the consequences of e-waste, because 'low carbon emissions do not equal low impact.' Footprints will also be unevenly distributed depending on the local power structure and the location of computing centers, and the transferred burden may exacerbate environmental problems for local communities.

The Taiwan Science Media Center stated that Taiwan currently has about 36 data centers, ranking 9th globally in number. In response to the UN report, it invited experts in power governance and regional governance to provide insights.

Zheng Anting, a professor at the Department of Land Economics at National Chengchi University, said the report indicates that the size of the land footprint depends on each country's power generation structure. If the proportion of biomass, onshore wind, and hydropower is higher, the land footprint will also increase.

Zheng believes that Taiwan's current energy policy is centered on solar photovoltaic and offshore wind power, which have already triggered large-scale land-use competition and social conflicts in Taiwan. Even if Taiwan's land footprint figures seem unremarkable in international comparisons, when the same land pressure is applied to a place with a relatively limited land area, the actual impact may be far more severe than the numbers suggest.

Zheng suggested that as resources have gradually been delegated to local governments in recent years, future mechanism design should strengthen the relevant powers and responsibilities of local governments. He proposed using the development review of the national spatial plan to require local governments to bear certain net-zero obligations and power generation responsibilities.

Zhao Jiawei, Director of the Taiwan Climate Action Network Research Center, stated that for the siting of AI data centers, industrial zones with renewable energy supply, low water stress, and grid capacity should be encouraged to plan 'energy parks' where computing facilities, renewable energy, energy storage, and reclaimed water systems are co-located. Large data centers should also participate in energy storage, and non-real-time tasks such as AI training should be shifted as much as possible to periods when green electricity is abundant or grid pressure is low, so that the new AI electricity demand can support the integration of renewable energy into the grid and enhance grid resilience.