The Environmental Cost of Artificial Intelligence

The hidden cost of Artificial Intelligence (AI) is the water consumed to run ChatGPT. As AI capabilities grow, so does the amount of natural resources, such as water, needed. The consequences of not addressing this issue could be dire. When considering where the most freshwater is allocated, many people consider daily human consumption for appliances such as the toilet, shower, laundry, and cooking. For large industries requiring water to succeed, textiles, chemicals, paper, energy production (oil, gas), and agriculture are examples. 

Let’s understand why AI requires water

A computer is comparable to a radiator. Each unit of electricity consumed by a computer releases an equal amount of megawatts of heat. If the equipment runs at that heat level for long periods, it will damage or potentially ruin the electronics. Water has the thermal capacity to transfer heat 25 times higher than air, meaning that using water to cool artificial intelligence equipment consistently is the most efficient resource for maintaining AI’s capabilities. 

One might wonder, if water is so crucial, why don’t we use just any type of water to maintain AI equipment? The answer lies in the necessity of clean, drinkable water. A system of evaporative cooling uses fresh water to maintain the temperatures in buildings such as university campuses. In AI Equipment Cooling, four types of water are used: deionized, distilled, chilled, and specially treated water. Clean, drinkable water is not just a luxury but a necessity for AI equipment to run smoothly and avoid deterioration. The water must be free from algae, bacteria, conductive properties, mineral deposits, or corrosion, which are water components that are unsafe to drink. 

Artificial intelligence is now an industry that demands large amounts of natural water resources. At the astronomical pace at which AI is growing, the hidden amount of water used in industries and human consumption raises concerns about what the long-term outcome will mean for this natural resource. As AI technologies continue to be developed, advanced, and deployed in all industries, the challenge is not just about growth, but about ensuring that the implementation is ethically and strategically used to benefit humanity while recognizing the economic effect of maintaining these systems. 

The astonishing rate at which AI consumes water

For around 10-50 questions imputed into Chat-GPT, an estimated 500 milliliters of water—equal to a standard bottle of water—are required to provide the answers. The silent and often overlooked cost of Big Tech’s water footprint is visually understood when recognizing that hundreds of millions of users interact with Chat-GPT daily. Looking at the water consumption of Big Tech organizations in 2022, Microsoft’s consumption was 1.2 billion gallons, and Google’s was 5.6 billion gallons. Both firms hope to replenish more water than they have used by becoming water-positive by the end of 2030. Their water footprint was released in 2022 before their AI integrations, Bing Chat and Google Bard, were published. Data centers typically generate the highest percentage of Big Tech’s water footprint, using water in evaporation to expel building heat. The United States operates over 5,300 data centers, ten times more than Germany, which houses the second most data centers worldwide. As of January 2024, the computing power required for AI was estimated to double every 100 days.

Water scarcity is a primary issue across all continents during the current climate crisis. It is essential to govern AI within planetary boundaries. This is possible by providing less energy-intensive solutions such as the following:

• Energy-Efficient Data Centers: renewable energy, advanced cooling solutions, efficient server utilization
• Algorithmic Improvements: efficient neural networks, removing unnecessary parameters and quantization, knowledge distillation
• Hardware Optimization: specialized AI chips, application-specific integrated circuits
• Data Management: data augmentation and compression to reduce storage and transmission energy costs
• Green AI Initiatives: transparency, benchmarking, promoting research in energy-efficient AI, and collaboration between industries to develop best practices
• Global effort by individual users: avoid using generative AI in unnecessary situations and instead do a simple search to obtain information which is less resource-consuming

Lastly, if market forces or innovation do not decrease water consumption, limitations should be put in place to ensure the water footprint does not continue to increase exponentially through the use of Big Tech. Additionally, more information should be publicized about this hidden cost of Artificial Intelligence, and companies should be required to disclose their use.

---

---