Sign up to see the future, today
Can’t-miss innovations from the bleeding edge of science and tech
AI data centers will consume as much water as the water needs of 1.3 billion people by 2030, according to a new report from the United Nations University Institute for Water, Environment and Health.
The report, released this week, found that the environmental cost of AI is being “systematically mismeasured” because current assessments focus on the carbon emissions from training large language models while overlooking the tech’s broader water and land footprint. The water footprint comes from cooling and powering the data centers, and the land footprint comes from the energy infrastructure and supply chains that go into building and running them.
In reality, the report stresses, these initial training costs are dwarfed by inference costs, meaning the cost of running the models to answer prompts, which make up 80 to 90 percent of total AI energy use. Training OpenAI’s GPT-4 model consumed up to 70 gigawatt-hours of electricity, for example. But running ChatGPT, it’s estimated, uses a monstrous 383 GWh from answering billions of prompts per day.
Factoring in inference costs, data centers powering AI will use 945 terawatt-hours of electricity by 2030, the report found, which is triple the combined electricity use of Pakistan, Bangladesh, and Nigeria — which, altogether, are home to over 650 million people.
That brings us to the water footprint. By that same year, AI’s thirst will see it consume 9.3 trillion liters water — which is equal the basic annual water needs of all 1.3 billion people in Sub-Saharan Africa.
Addressing AI’s environmental impact won’t be as straightforward as transitioning to greener sources of power, the report warns. Ditching coal in favor of bioenergy could slash carbon emissions associated with electricity costs by 70 percent, but cause the water footprint to surge by 30 times, and the land footprint by 100 times.
“What surprised us most is how often the choices that look greenest from a carbon perspective end up worse for water or for land,” report lead author Miriam Aczel, an UNU-INWEH researcher, said in a statement about the findings.
“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,” she added.
And the hits just keep coming: paradoxically, making AI more energy efficient may actually increase its environmental footprint.
“A lot of people think that the environmental footprint of AI reduces, as technology improves and processes become more efficient. But that is only a partial picture of the overall problem,” coauthor Kaveh Madani, director of UNU-INWEH, said in the statement. “More efficient and affordable AI and energy mean more consumption of AI, making the overall footprint far bigger than what we save through efficiency gains.”
More on AI: Neighbors Horrified by Data Center Twice the Size of Manhattan
