Okay, here’s a breakdown of the key data from the provided text, organized for clarity. I’ll cover the main points, calculations, and comparisons.
Main Idea:
The article discusses the often-overlooked water footprint of Artificial Intelligence (AI) systems. While AI is seen as a digital technology,it requires significant energy to operate,and that energy production requires water (for cooling power plants,etc.). The amount of water used varies dramatically depending on the efficiency of the AI model and the energy source powering it.
Key Findings & Calculations:
AI and Water Connection: AI models consume energy, and energy production consumes water. Less efficient AI models and less efficient power grids use significantly more water.
Water Factor: The amount of water used per watt-hour of electricity ranges from 1.3 to 2.0 milliliters per watt-hour (ml/Wh).
1.3 ml/Wh = Highly efficient facilities/cleaner energy
2.0 ml/Wh = More typical operations
Energy Usage Examples (per prompt):
GPT-5 (medium length, 150-200 words): 19.3 watt-hours
GPT-4o (medium length): 1.75 watt-hours
Google Gemini (median prompt): 0.24 watt-hours
Water Footprint Calculations (per prompt):
GPT-5 (19.3 Wh):
With 2 ml/Wh factor: 19.3 wh 2 ml/Wh = 38.6 ml (approximately 39 ml)
With 1.3 ml/Wh factor: 19.3 Wh 1.3 ml/Wh = 25.09 ml (approximately 25 ml)
GPT-4o (1.75 Wh):
With 2 ml/Wh factor: 1.75 Wh 2 ml/Wh = 3.5 ml
With 1.3 ml/Wh factor: 1.75 Wh 1.3 ml/Wh = 2.275 ml (approximately 2.3 ml)
Google Gemini (0.24 Wh): 0.24 Wh 0.26 ml/Wh = 0.26 ml (roughly five drops)
Daily Water usage Estimates (based on prompt volume):
Google Gemini (2.5 billion prompts): Approximately 650,000 liters per day.
GPT-4o (2.5 billion prompts): Approximately 8.8 million liters per day.
GPT-5 (2.5 billion prompts): Approximately 97.5 million liters per day.
Crucial Takeaways:
Efficiency Matters: More efficient AI models (like GPT-4o compared to GPT-5) use significantly less water.
Infrastructure Matters: The source of electricity (clean vs. fossil fuels) and the efficiency of power plants greatly impact the water footprint.
Scale Matters: even small amounts of water per prompt add up to considerable volumes when considering the billions of prompts processed daily.
* Comparison is Challenging: direct comparisons between models are challenging because prompt lengths and the specific water factors used in calculations aren’t always consistent.
Let me know if you’d like me to elaborate on any specific aspect of this information!
