AI Threatens Water and Land Supplies as Data Centre Electricity Consumption Projected to Surge
The explosion in the use of artificial intelligence (AI) is expected to not only increase global electricity demand but also potentially pressure water supplies and land use on a large scale. Recent research from the United Nations University (UNU) estimates that global AI data centre electricity consumption could reach 945 terawatt-hours (TWh) per year by 2030, which is nearly three times the electricity needs of Pakistan, Bangladesh, and Nigeria combined, representing a population of approximately 650 million people.
The report indicates that the environmental impact of AI stems from more than just energy consumption. AI infrastructure also requires significant amounts of water for cooling data centres, as well as vast areas of land to support power plants and supply chains.
The study estimates that global AI water consumption by the end of this decade could be equivalent to the water needs of 1.3 billion people. Meanwhile, land requirements could reach 14,500 square kilometres, nearly twice the area of the Greater Jakarta (Jabodetabek) region.
According to the official UN website, the report highlights the environmental impacts of AI that have previously received little attention. Previous studies focused primarily on carbon emissions from data centres, while the pressure on water and land resources has been relatively rarely discussed.
Researchers suggest that efforts to reduce emissions through the development of AI infrastructure could potentially create new environmental pressures. While using certain energy sources may lower greenhouse gas emissions, it may simultaneously require more water and land. This situation is considered risky for regions already facing scarcity in water resources or productive land.
The report also reveals that daily AI usage is actually the largest contributor to energy consumption. AI operational activities are estimated to account for approximately 80 to 90 per cent of total energy needs, significantly higher than the process of training AI models.
“The scale is staggering: a single widely-used AI service is estimated to process around 2.5 billion prompts per day, consuming hundreds of gigawatts of electricity per hour annually,” the report stated.
Energy requirements also vary by type of AI service. AI image generation requires far more energy than text-based requests, while video generation requires even higher energy consumption.
The research warns that increased technological efficiency may not necessarily curb the surge in energy consumption. A phenomenon known as the ‘rebound effect’ means that AI usage may actually increase as the technology becomes cheaper and more efficient.
“In such a scenario, total resource consumption could continue to rise even as technology becomes more energy-efficient,” the report concluded.