Water usage has become one of the main topics in the discussion about AI data centers.
The overall global demand for water is, perhaps, not the main problem. But water usage near these data centers is a real problem for those who live nearby. 40% of the world's data centers are located in regions of high or extreme water stress. And the demand for water for cooling peaks exactly in summer - the same moment when farmers and local residents are already facing drought and shortage.
The report reminds us of the old Jevons paradox: when a technology becomes more efficient and cheaper, people use more of it, and total resource consumption only rises instead of falls.
We've already been through this with coal and plastics - each time, efficiency didn't reduce consumption; it accelerated its growth by lowering the barrier to entry. With AI, the same dynamic applies: chips become more efficient - video generation grows. The water footprint of a single query goes down - the number of queries multiplies. A model is trained to be more efficient - and immediately a thousand new models are launched on top of it.
But the trend is clear: water is becoming as strategic a resource for building data centers as cheap electricity once was. And businesses whose "digital" products sit on physical infrastructure will increasingly have to prove that their growth doesn't leave neighboring cities without water.
According to a new UN report, by 2030, data centers will consume 945 TWh of electricity - almost three times more than Pakistan, Bangladesh, and Nigeria combined. That's 650 million people.
The water footprint - 9.3 trillion liters of water per year will go toward cooling and electricity generation. That's enough to cover the basic domestic needs of 1.3 billion people - nearly the entire population of sub-Saharan Africa.
The overall global demand for water is, perhaps, not the main problem. But water usage near these data centers is a real problem for those who live nearby. 40% of the world's data centers are located in regions of high or extreme water stress. And the demand for water for cooling peaks exactly in summer - the same moment when farmers and local residents are already facing drought and shortage.
The report reminds us of the old Jevons paradox: when a technology becomes more efficient and cheaper, people use more of it, and total resource consumption only rises instead of falls.
We've already been through this with coal and plastics - each time, efficiency didn't reduce consumption; it accelerated its growth by lowering the barrier to entry. With AI, the same dynamic applies: chips become more efficient - video generation grows. The water footprint of a single query goes down - the number of queries multiplies. A model is trained to be more efficient - and immediately a thousand new models are launched on top of it.
But the trend is clear: water is becoming as strategic a resource for building data centers as cheap electricity once was. And businesses whose "digital" products sit on physical infrastructure will increasingly have to prove that their growth doesn't leave neighboring cities without water.