The hidden cost of AI: draining our Water resources

Estimated reading time: 3 minutes

In the current world of digital civilization, Artificial Intelligence (AI) stands out as an important technological invention that has triggered major changes to almost all current industries, including the health and entertainment sectors. However, beneath these revolutionary capabilities lies an environmental cost that is often overlooked: the massive volume of water that most AI systems must use to deliver their services suitably.

The water behind the code

AI systems, particularly those involved in machine learning like large language models, require vast computational resources. These systems are typically housed in data centres, essentially nerve centres for our digital lives. The operation of these data centres involves significant energy consumption, and that energy generates heat. To manage this heat, data centres employ cooling systems, and water is one of the most effective coolants available.

• Training AI Models: The process of training an AI model, like the famous language models such as GPT-3 or GPT-4, is incredibly water-intensive. For instance, training GPT-3 reportedly used approximately 700,000 litres (about 185,000 gallons) of water. This water is primarily used in the cooling process to prevent server overheating, but it also includes water used in the generation of electricity that powers these centres.

• Operational Water Use: Even after the training phase, the operational phase of AI systems continues to consume water. Every interaction or query with an AI, such as a user asking ChatGPT a question, involves water use. Estimates suggest that simple interaction with ChatGPT could use around 500 millilitres (0.13 gallons) of water for every 5 to 50 questions or prompts, due to the cooling needs of data centres in response to the computational load.

More like this

• Kwara Govt partners NALDA on food security 
• NBS reveals high inflation weakened Nigerians’ spending power in 2022
• Court sentences Guinean to two years in prison for wildlife trafficking 
• Tinubu: APC refutes claims of apology to Duke over usage of image in campaign
• The Body Keeps Count (Part 4): Importance of drinking water 

The scale of water usage

The scale at which AI consumes water is alarming, especially when you consider the cumulative impact of multiple data centres worldwide:

• Microsoft and Google: In 2022, Microsoft’s water consumption surged by 34%, reaching nearly 1.7 billion gallons annually, a spike largely attributed to the demands of AI research and development. Google, in the same period, reported a 20% increase, consuming about 5.56 billion gallons, highlighting how AI has pushed up water use in their data centres.

• Future Projections: With AI’s expansion, projections indicate that by 2027, the global water use by AI systems could reach between 4.2 billion and 6.6 billion cubic meters, which is nearly half of the UK’s annual water consumption. This underscores the need for sustainable practices in AI development.

It is our capability at work that has brought about this integration of AI into our daily existence and, at the same time, there is a need to ponder over the issue is costliness to the environment. However, as we evolve we need to shift towards policies that will reduce the use of water in the technology sector as we fuel the development of AI technologies. The process of becoming sustainable AI is not merely a development of advancement but also includes the development following ecological rationality.