The Thirsty Side of AI

I was today years old when I found out ChatGPT consumes water. Yes!! ChatGPT, the generative AI platform consumes water!! With the rise of AI models like ChatGPT, there’s growing interest in their environmental impact. While discussions often focus on electricity use and carbon emissions, one lesser-known factor is water consumption. ChatGPT consumes a 500 ml of water every time it answers about 5 to 50 prompts sent in by its users. This water is being used in cooling down the supercomputers and the servers that generate heat after using electricity. Heavy electricity is used just for analysis and then to process the human-written text and then it is given to large language models like ChatGPT or Gemini. 

It's not that hard to understand the fact that AI algorithms require tonnes and tonnes of energy, but recent studies have shown that AI consumes loads of water too. As you look into the numbers of how much water is required to call all data units and servers, even training AI like GPT-3 alone from OpenAI and Microsoft partnerships, use up to 700031.8 Liters of water, which is the amount of water that is more than sufficient to cool a nuclear reactor, or it was enough to produce around 400 BMWs. While 500 ml seems less, the total water footprint is still very humongous considering the billions of ChatGPT's users. More complicated next-generation AI models and tools like GPT-4, Google's Gemini, WhatsApp's Meta AI etc., would ideally consume even more water but it can't be estimated due to lack of statistical numbers. 

How AI Uses Water:

AI models run on powerful servers housed in data centers. These data centers generate significant heat and require cooling systems to maintain optimal temperatures. Two primary cooling methods are used:

  1. Air Cooling – Uses fans and ventilation systems to dissipate heat.
  2. Water Cooling – Uses water to absorb and carry away heat, often through evaporative cooling or direct liquid cooling.

Many large-scale data centers rely on water cooling because it is more efficient and sustainable than air cooling. This means that every time you use ChatGPT, some amount of water is used to keep the servers running.

What Can Be Done?

1. More Efficient Cooling Technologies-

  • Liquid Immersion Cooling: Instead of air or water-based cooling towers, some companies are submerging servers in special non-conductive cooling fluids. This reduces water usage and improves efficiency.
  • Direct-to-Chip Cooling: Some data centers are developing cooling systems that bring liquid coolant directly to processors, reducing waste.
  • AI-Optimized Cooling Systems: Using AI to monitor temperature and dynamically adjust cooling can reduce water use by up to 40%.

2. Using Recycled & Non-Potable Water-

  • Many data centers now use gray water (treated wastewater) instead of fresh drinking water.
  • Some facilities tap into industrial water sources that would otherwise go to waste.

3. Locating Data Centers in Cooler Climates-

  • Placing data centers in naturally cool areas reduces the need for artificial cooling.
  • Some companies are experimenting with underwater data centers, where the ocean itself helps dissipate heat.

4. Switching to Air Cooling When Possible-

  • In low-humidity climates, air cooling is a viable alternative to water-based cooling systems.
  • Large-scale AI models could be optimized to use hybrid cooling systems, which adjust based on weather conditions.

5. Improving AI Model Efficiency-

  • Smaller, more efficient models can reduce the energy and cooling demands of AI.
  • Distillation techniques (training smaller versions of AI models) could allow similar performance with less computing power.
  • Using specialized AI chips (like TPUs and FPGAs) instead of general-purpose GPUs can increase efficiency.

6. Investing in Renewable Energy & Sustainable Practices-

  • Pairing AI data centers with solar, wind, or hydroelectric power can lower the overall environmental impact.
  • Companies like Google, Microsoft, and Amazon are working toward water-positive goals, meaning they aim to replenish more water than they consume.

7. User Awareness & Demand for Sustainable AI-

  • Users and businesses can push AI providers to adopt sustainable data center practices.
  • Governments can implement water regulations that encourage companies to use low-water or waterless cooling.
I know that just coming up with solutions can't actually solve the issue, rather by implementation of these solutions, which is why I made this blog in the first place, so that together as a community we can implement many such solutions and make a difference globally :)


Comments

Post a Comment

Popular posts from this blog

Genetics: The blueprint of an individual 🧬

Ever wonder why you have your dad’s nose but your mom’s temper? Or why your sibling got the perfect curls while you're stuck with a frizzy mess? Welcome to the wild world of genetics—the science of inheritance, mutations, and why we are the way we are. Genetics is the study of how traits are passed from parents to offspring. It’s like a biological instruction manual that determines everything from your eye colour to hair colour to how your body works. It explains why children look like their parents, why some people are more prone to certain diseases, and how species evolve over time. This field of science revolves around DNA, genes, and chromosomes, which together form the instructions for building and running living organisms. The instructions on how a particular gene expresses itself is written in DNA, a molecule found in almost each and every cell of your body. 1) DNA -DNA or deoxyribonucleic acid, which is basically life’s instruction manual, but in a secret code of four lett...
Read more

STEM Education and Grassroots innovation - empowering communities through science 🧪🔭

If you think science is just for people in white lab coats shouting “Eureka!”—think again. STEM (Science, Technology, Engineering, and Math) is everywhere, from helping NASA and ISRO launch rovers to fixing stuff in your own backyard. STEM provides individuals with the knowledge and skills which are needed to solve real world problems. But in many communities, access to STEM resources is limited due to lack of funding, inadequate school programs or just lack of interest at a whole. That’s where grassroots innovation steps in, proving that science doesn’t have to come from billion-dollar research labs—it can also come from a garage, a community center, or that one guy down the street who built a wind turbine out of bicycle parts. Grassroots innovation is all about people using STEM to solve real-world problems, without waiting for big corporations or government funding.  Examples of Grassroots STEM initiatives include: 1)Community makerspaces: Local makerspaces equipped with 3D...
Read more