How much water do ChatGPT data centres consume to address our curiosity? Results will blow your mind.

 

It's no news that the artificial intelligence chatbot ChatGPT requires large amounts of data to become smarter. But how much water do data centres used to run such chatbots consume?

OpenAI's AI chatbot ChatGPT has demonstrated massive progress in a short period with its uncanny ability to summarise research studies, crack business school and medical exams, and basically respond to a range of user questions with human-like language.

But a new study has elaborated on the sacrifices the environment needs to make every day to help such chatbots satisfy our curiosity. According to yet-to-be-peer-reviewed research, a simple conversation of roughly 20-50 questions and answers with the AI chatbot in a single system likely requires a "500 ml bottle of water".

While the amount might not seem too much, the total combined water footprint is immensely large, considering ChatGPT's billions of users worldwide, according to researchers from the University of California, Riverside, and the University of Texas, Arlington.

Just to train GPT-3 alone, Microsoft potentially consumed an alarming 700,000 litres of clean freshwater, enough to produce 370 BMW cars, the scientists noted in the paper titled "Making AI Less Thirsty", urging companies running AI models to "take social responsibility" and cut their own water footprint.

The tech giant has partnered with OpenAI and invested a whopping $10 billion in the company.

While the amount of water Microsoft requires to cool its data centres in the US to train GPT-3 is already staggering, the numbers would have almost tripled if the location was somewhere in Asia. Crazy, right?

But there is more. Researchers believe the figures will likely increase by "multiple times" for the newly-released GPT-4 AI system which features a larger model size. But they noted that the absence of almost any public data was hindering a reasonable estimation of the water footprint for GPT-4.