ChatGPT's Impact on Water Usage
As ChatGPT and other generative AI tools gain popularity, concerns about their environmental impact are coming to light. It has been reported that while training GPT-3, Microsoft used 700,000 liters of fresh water. This is equivalent to the water needed to fill a nuclear reactor's cooling tower or used to produce 370 BMW cars or 320 Tesla vehicles. These numbers estimate that ChatGPT requires 500 ml of water (a standard 16.9 oz water bottle) for every 20 to 50 questions answered.
While this amount may not seem significant on an individual level, it can accumulate to a large water footprint due to ChatGPT's large user base. The water footprint includes on-site and off-site water usage and is still significant despite the small amount of water per user.
AI models like GPT-3 are hosted in data centers, which require significant amounts of energy and water to operate. Energy is used in the training process, and water is used to keep temperatures in check across the entire infrastructure. Freshwater is necessary for proper humidity control. Using saltwater can lead to "corrosion, clogged water pipes, and bacterial growth."
While Microsoft is investing in research to measure the energy use and carbon impact of AI and make large systems more efficient, it is up to individuals to reduce their water footprint. We can do this by being conscious of our use of generative AI tools like ChatGPT and limiting our usage where possible.
Ultimately, it is necessary to address water footprint alongside carbon footprint to enable truly sustainable AI.
