AI Data Centers Face Scrutiny Over Escalating Water Consumption
Sam Altman said that ChatGPT uses 0.3ml of water per query. Morgan Stanley said that AI data centers could consume 1,068 billion liters by 2028. AI will consume 1 trillion liters of water by 2028, every year. What is the actual cost? You will be blown away. Splash zone! How much water do they use? Water consumption in cubic meters. Google 24M m cubed, Microsoft 7.84M m cubed, Meta 3M m cubed, Apple 6.1M m cubed, Nvidia 134K m cubed. An Olympic-sized swimming pool requires 2.5 million liters of water to fill, based on its standard dimensions. 9,618 Olympic pools, 3,138 Olympic pools, 1,231 Olympic pools, 2,438 Olympic pools, 54 Olympic pools. Google's annual water consumption: 120 million bathtubs - enough for Japan to take a bath! These five major consume 1.7 trillion liters. We should know how does AI even use water? Of course, we know that AI is doing trillions of calculations every day. And guess where these calculations happen? You guessed it right, on high-end GPUs and TPUs. And these GPU units, they are stacked on the top of each other. So together, they produce heat equivalent to Utah's AI data center will produce heat equivalent to 23 atomic bombs every day, spans 40,000 acres, consumes more power than entire state. of three houses. How does the cooling of these GPUs happen? It happens through the wet cooling method. Water is left to evaporate on the top of these data centers so that they could cool down. Does it occur to you what kind of water you guessed it right, our fresh water. So when Sam Altman said, a single query uses 0.3 ml equivalent to 1/15 of a tablespoon. But I read some papers and did some calculations myself and found out that this 0.3ml wala Sam Altman figure has to be used for cooling for GPUs and TPUs and training of these AI models nonstop. So the real question is, what is the actual water consumption of these AI tools? ChatGPT uses 148 million liters of water every single day, which is 6 million UK households' much water every single day. Well, this is all fresh water that we are talking about. We are also living in water stress zones, some places like Texas, Arizona, they are already going through groundwater depletion because of these data centers. Places like Iowa had to stop agriculture, they needed water for these data centers so that they can cool them down. Should we give this water for AI to do our homework or make that stupid video? Or should we keep this water for agriculture so that we could eat? AI is not just expensive for companies or employers, it's expensive for the environment as well. So in the end, we'll have AI to do our homework, but we will not have water to drink or food to eat. What problem design kar liya na humne?
What's right
What's wrong
What's debatable
Breakdown
Summary of Claims: The video asserts that AI data centers are consuming vast amounts of water, with significant increases projected by 2028, leading to impacts on water-stressed regions and conflicts with agriculture. It cites Sam Altman's figures for ChatGPT's per-query consumption but disputes their adequacy, presenting much larger figures for overall data center water usage and daily ChatGPT consumption.
Water Consumption Projections: Sources confirm that AI data centers are indeed driving a substantial increase in water consumption. Projections suggest this usage could double or quadruple by 2028 [2].
While specific figures vary, one source estimates U.S. AI-related data centers could require up to 32 billion gallons annually by 2028 [8], and another projects a 300% increase to 68 billion gallons by 2028 [7].
The video's figures of 1,068 billion liters (approx. 282 billion gallons) and 1 trillion liters (approx.
264 billion gallons) are considerably higher than these, potentially encompassing broader indirect water footprints [7][8]. **Per-Query vs.
Overall Consumption:** Sam Altman did publish figures in June 2025 stating that an average ChatGPT query uses about 0.34 watt-hours of electricity and 0.000085 gallons (0.32 ml) of water [3][4][9]. However, the video correctly points out that this per-query figure does not account for the substantial water needed for cooling the massive infrastructure and training the AI models, which is a significant part of the overall water footprint [3][7].
The claim that ChatGPT uses 148 million liters of water daily is not supported by the provided sources and appears to be a significant exaggeration [7]. Impact on Water Resources and Agriculture: There is strong evidence that data centers are located in water-stressed areas, leading to concerns about groundwater depletion and competition for resources [7][8].
The claim that Iowa had to stop agriculture due to data center water needs is a specific instance not directly confirmed in the provided references, though the general conflict between data center water demands and other needs like agriculture is a recognized issue [7][8].