Study Finds Alarming Water Usage By AI Data Centers Like ChatGPT

According to a recent study, AI data centers consume staggering amounts of water. The findings have sparked concerns about the environmental effect of tech companies’ operations.

OpenAI’s chatbot earned popularity last year for its unusual way of responding to queries. Over time, the bot has shown significant achievements, including fascinating abilities. It can answer logical queries and summarize research papers.

ChatGPT responds in a human-like manner but with informational details.

Two-thirds of the global population is expected to face fresh-water shortages by 2025. Data centers have traditionally relied on heavy water consumption for cooling, but can now reduce usage up to 50% annually with our hybrid cooling units. Learn more! https://t.co/5oRfxQna8K pic.twitter.com/AicjsfM5wg

— Johnson Controls (@johnsoncontrols) June 7, 2019

Water Consumption of ChatGPT

The study shows that prior research has focused on the carbon footprint of similar AI models. The water consumption to run them on a huge scale was left unnoticed.

According to the scientists, training GPT-3 alone, Microsoft has used 700,000 liters of water. This amount is enough to make 370 BMW cars. In addition, other AI models, such as Google’s LaMDA, can suck up millions of liters of water.

Furthermore, the new research was also posted in arXiv. In estimate, a single conversation with nearly 20-50 questions with ChatGPT in a single system can consume a “500ml bottle of water.”

You may also like: New Study Shows ChatGPT Could Displace 4.8 Million Workers in the US

The researchers have developed a framework. This calculates the amount of clean and fresh water companies use to generate electricity to supply power to data centers and cooling servers to run these AI models.

What are the implications of excessive water usage in AI data centers?

Excessive water consumption in AI data centers can have significant financial, environmental and operational consequences. The increasing demand for HIs has led to more data centers. These facilities rely on water for their system cooling.

However, the requirement for huge amounts of water can jeopardize local water supplies, particularly in places with an existing water shortage.

Moreover, the energy these AI data centers need requires the pumping and treating of water. These processes contribute to climate change and greenhouse gas emissions.

Thus, companies relying on AI may soon face an increase in costs as water becomes more expensive due to shortage.

Call for Social Responsibility

In response to worldwide water shortages, the researchers requested that companies address their water footprints and take social responsibility.

The scientists also mentioned n the study that “AI models can, and also should, take social responsibility and lead by example in the collective efforts to combat the global water scarcity challenge by cutting their own water footprint.”

How can we stem the tide of energy usage by data centers? Read how water is powering new energy innovation: https://t.co/d43FipCIuT #DoubleBottomLine pic.twitter.com/VDJ0ctFmeC

— Hitachi Vantara (@HitachiVantara) July 4, 2019

They also noted that even if “A 500ml bottle of water might not seem too much, the total combined water footprint for inference is still extremely large.” This is a reality considering the massive volume of ChatGPT users.

You may also like: Judge Employs ChatGPT to Determine Bail in Murder Trial

Researchers also confirmed these numbers might double up by “multiple times” on the newly-launched GPT-4 with its bigger model size.

The Need for Transparency

The scientists also called for more transparency in the water footprint of these AI models. They suggest more disclosure of data about operations and the effectiveness of water usage during the systems’ running time.

With the new ChatGPT4, there’s an alarming amount of water consumption. It is essential for companies to adhere to environmental and social responsibilities. They should also increase transparency in the actual water consumption of AIs.