The Invisible Cost of Smarter Machines
We turn upside down about the energy consumptions of AI: the amount of carbon emissions, the requirement of electricity but little people mention their thirst. And behind each ChatGPT output, each Midjourney image and every self-driving car algorithm is an industrial-scale use of water that absolutely horrifies me.
Consider that. Half a liter of water can already be consumed during backend cooling during a single chat with an AI chatbot. And then multiply by billions of searches a day. All of a sudden, it is clear that environmental impact of AI is not only about energy, but also about disappearing reservoirs, overstressed communities, and impending sustainability struggle.
How come this is not headlining? We also know that no sustainability report provided by Big Tech actually talks about the use of water, rather drowning it in misunderstood terms of efficiency. However, this silent crisis the world is experiencing is not going to remain a secret much longer now when AI models are becoming bigger and droughts are becoming even worse.
The Jaw-Dropping Numbers Behind AI’s Water Addiction
So, let us become specific. The process of training OpenAI GPT-3 consumed ~185,000 gallons of water: enough to create 300 Teslos or provide a small town with water during weeks. The current data centers being operated by Google across the globe and using Gemini and other AI implementations pulled in 15.8 billion gallons in 2022 alone, or more than the amount required by full-scale cities.
So Why So Much?
- In the data centers, evaporative cooling is predominant. Servers should also have a continuous supply of water like sweat athletes so that they do not overheat.
- Whereas conventional cloud tasks can be executed in days, the brute-force training of AI needs weeks of continuous computation.
- Location multiplies inefficiency: Arizona data centers consume 3-5 times more water than Swedish data centers because this is the climate.
Microsoft Iowa Problem Case Study
Mercifully, there was a drought in 2023 when a large amount of district water was consumed by Microsoft in West Des Moines data center (6 percent). Residents protested. The company aims at fixing this with water positivity by 2030- but is technology really the solution that ends the millions of gallons of water that are disappearing every day?
The Hypocrisy of “Green AI” Claims
Claiming to optimize cooling through AI, Google continues to use perfectly clean groundwater in its data centers in Oregon. Microsoft promotes its new term, sustainable campuses, when it outsources the creation of server farms to a region of Chile in the Atacama Desert where even now villages are already experiencing rationing of water.
The Gap of Transparency
- None of the big AI companies segregates the water use per model.
- There is also the existence of cooling technology (such as liquid immersion but it is not scaled because of cost).
- Just as an analogy, it is just like an SUV maker was proud of its fuel efficiency…and had a leak secret oil pipeline.
Useful Take: Warning by Dr. Ren
“The water footprint of AI developed at a higher rate than the carbon footprint. Instead, we are exchanging smarter chatbots with tired aquifers.”
The Human Toll: When AI Drinks a Town Dry
The implementation of the new data center by Microsoft in Goodyear Arizona has created a panic over drained wells. In Aragon, Spain the servers of Meta attacked farmers as the irrigation was diminishing.
Real-World Consequences
- Chile 2024: The droughts compelled data centres to truck in water and this increased the local costs.
- The monsoon failures and urban shortages may come in collision with the ambitions of India in AI.
- Coming crises: Will water inequality become worse in the future should AI centers be developed in Nevada or Saudi Arabia?
Conclusion: Time to Choose—AI Progress or Water Survival?
The dirty little secret? AI can not afford to continue growing in this way. Tech giants will have to invest in air-cooling, wastewater recycling, or desert-friendly data centres and/or regulators will make them make the investment.
What You Can Do
- Probe demand: Require information about water-per-query figures to remain open to AI companies.
- Encourage ethical AI: Prefer the companies that employ sustainable cooling.
- Be smart: Any email, picture or bit of code created by AI, uses water under the table.
Final Thought
“It would be like accepting a car which burns up a forest with each mile that is driven. Why would anyone tolerate AI that drawers reservoirs per search?“