Should we implement AI prompt water consumption metrics to understand the environmental impact of AI services, similar to a car's gas mileage counter?
While individual prompt water usage is minimal, the aggregate consumption by data centers is substantial and visibility into these metrics can drive more sustainable practices.
AI Water Consumption: The Nuance
• Minimal Per Prompt, Significant in Aggregate: While a single AI prompt might use only milliliters of water, the sheer volume of AI operations globally leads to considerable overall water consumption. "One ChatGPT query uses about as much energy as a lightbulb running for 30 seconds."
• Cooling is the Main Driver: The primary reason AI data centers consume water is for cooling the powerful hardware that generates significant heat. "From what I understand the majority of water is used during cooling."
• Location Matters: The environmental impact of water usage by data centers largely depends on their location and the local water availability. "The problem is that they are being built in already water-stressed places."
The "Water Waste" Debate
• Not Always "Wasted": Many data centers use closed-loop cooling systems or non-potable water, and water used for cooling often re-enters the water cycle, albeit potentially warmed. "water used in data centres is not thrown out once used. basically they use water to cool down the servers but they keep recycling it again and again."
• Misinformation and Exaggeration: Some Redditors believe the environmental impact of AI's water usage is often overstated, especially when compared to other industries. "The water footprint of AI is miniscule compared to what people make it out to be, because big number sounds scary."
• Actual Local Impact: Despite global comparisons, local communities can experience real impacts if data centers consume water from already stressed regional supplies. "If you put a data centre in an arid climate like California, you could say the amount of water used for cooling is wasteful compared to say using it for drinking."
Comparison to Other Industries
• Agriculture and Other Digital Services are Larger Consumers: AI's water consumption, while growing, pales in comparison to industries like agriculture or even the total water usage of traditional data centers supporting streaming and social media. "If california stopped growing almonds, they'd save 2 trillion gallons of water a year ignoring AI - ALL DATA CENTERS in the US use 200~ billion gallons of water a year."
• Context is Key: It's crucial to put AI's water usage into perspective alongside other human activities and industrial processes. "In context, singling out AI for its water usage is motivated reasoning at best."
Do you think making these metrics visible would encourage AI service providers to adopt more water-efficient technologies?