The AI Energy Paradox

Written by Vedanti Rawal

Artificial intelligence has become an indispensable, multifaceted tool in modern life, boasting contactless applications ranging from virtual assistants to predictive analytics for climate change. However, there is a hidden cost to this virtual intelligence. While the software seems weightless, the hardware is literally thirsty.

Training a model like GPT-4 can consume enough water to fill an entire nuclear cooling tower. According to a study from the University of California, Riverside, a single 100-word prompt on Gemini uses a total of 16 oz of drinking water. AI requires this because it operates on powerful chips that generate substantial heat. To keep entire data centers from overheating, millions of gallons of nonrecyclable water are used for cooling. Moreover, the electricity required to power these data centers also reduces the available drinking water. 

The Paradox: Can AI Save the Planet?

Despite its significant footprint, AI may be one of our most effective tools against climate change. 

Smart Grids: AI is used to manage energy grids to reduce waste by directing electricity to precisely where it's needed. 

Precision Farming: AI helps farmers use 20% less water and fertilizer by predicting crop growth and identifying when additional care is required. 

Carbon Capture: Scientists are using AI to accelerate the discovery of new materials for Carbon Capture technology.

The Path Forward

The paradox is clear: scientists and engineers are using a resource-heavy tool to solve a resource-scarcity crisis. To make this "blade" less dangerous, the technology industry is shifting toward an environmentally friendly version of AI. This includes moving data centers to colder climates to reduce cooling needs and developing smaller networks that require a fraction of the power.

What Can You Do? 

For users, mindfulness is essential. Before asking an AI to write a joke or a trivial email, remember the "water bottle" cost. AI is a powerful resource, so let’s treat it like one.

1. https://news.ucr.edu/articles/2023/04/28/ai-programs-consume-large-volumes-scarce-water

2. https://www.eesi.org/articles/view/data-centers-and-water-consumption#:~:text=Similarly%2C%20as%20energy%20use%20increases,usage%20of%20approximately%201%2C000%20households.