A recent study conducted by The University of California has put ChatGPT-4’s huge resource consumption under scrutinyone of the most advanced artificial intelligences of the moment. Although many focus on energy consumption or the impact on hardware, the use of water by data centers is also significant. This fact has given rise to an increasingly relevant debate on the sustainability and environmental impacts of these technological tools.
How much water does it take to generate just 100 words?
According to the study, For ChatGPT-4 to generate a text of just 100 words, approximately three bottles of water are needed. This fact, although at first glance it may seem trivial, takes on substantial importance when considering the scale of AI use around the world. For an individual user, writing a few sentences may seem like a marginal expense, but If we extrapolate this figure to millions of users who interact daily with this type of technology, the amount of water becomes astonishingly high.
Water consumption is mainly related to the cooling systems of data centers, where the servers are housed. that allow these AI models to operate. These centers require large amounts of energy to operate and avoid overheating. The use of water, either for direct cooling or to generate electricity in thermoelectric plants, is essential in this process.
Why water is used as a reference measure
One of the most striking elements of the study is the use of water bottles as a reference measure. Although it may initially seem like an unusual choice, there is a powerful logic behind it. The impact of resource consumption such as water and energy is difficult for most people to visualize.since these concepts are usually expressed in technical terms or in large units (such as cubic meters or kilowatt-hours).
Water bottles, on the other hand, are an everyday object that anyone can easily understand.So when we talk about “three bottles of water” for every 100 words generated, the impact becomes much more tangible.
Such references also facilitate dialogue around the sustainability and efficiency of artificial intelligence. Instead of arguing in abstract terms, experts can resort to comparisons that make the issue more accessible to the general public.
The background of the problem: why so much water?
ChatGPT-4’s huge water consumption is largely due to the infrastructure that supports the processing of massive data required to train and run these models. Artificial intelligences like GPT-4 require colossal amounts of energy to process the volume of data they work with, which generates heat. To avoid overheating, servers in data centers must be kept at appropriate temperatures, and this is where water comes into play.
Data center cooling systems use water to keep temperatures under control, especially those that use evaporative cooling or hybrid systems. This process, which may seem secondary, is actually one of the major pain points in the use of energy by the technology giants. The greater the processing demand, the greater the need for cooling, and therefore, the greater the water consumption.
In areas where water is a scarce resource, this fact could generate additional tensions.Data centers are strategically located in regions where energy is cheaper or the climate allows for lower cooling costs, but that does not mean that environmental impact is not a cause for growing concern.
Water consumption by technologies like ChatGPT-4 raises important questions about the sustainability of artificial intelligence on a large scale. As the technological world advances by leaps and bounds, it must also face the challenge of minimising its environmental impact, especially with regard to the use of essential resources such as water.
This study reinforces the need to continue exploring more efficient and sustainable technologies that allow balancing technological progress with environmental preservation.
Continue reading:
– US government will have priority to test new AI models
– OpenAI will not facilitate the detection of texts created with ChatGPT
– OpenAI does not stop and is already working on ChatGPT 5