By 2030, Will AI Drink Up Water Equivalent to 1.3 Billion People? The Truth About Environmental Impact!

📰 News Overview

• Massive Resource Consumption: By 2030, AI-related water consumption is projected to reach that of 1.3 billion people (equivalent to Sub-Saharan Africa), with land use hitting 14,500 square kilometers (double that of the Jakarta metropolitan area).
• National-Level Energy Demands: The current power consumption of AI data centers (448 TWh) is comparable to that of France. By 2030, it will require three times the total consumption of Pakistan, Bangladesh, and Nigeria combined.
• “Inference” as the Major Contributor: Traditionally thought of as “training,” it turns out that generating user responses (inference) accounts for 80% to 90% of total energy consumption.

💡 Key Points

• Usage Disparities: One chat costs 200 times more energy than classifying emails, 1,400 times for image generation, and a whopping 200,000 times for generating short videos.
• Complex Trade-offs: Pushing for low carbon options (like biofuels) could paradoxically increase water consumption by 30 times and land use by 100 times, necessitating a multifaceted evaluation.

🦈 Shark’s Perspective (Curator’s Insight)

What’s truly jaw-dropping about this news is how it delivers a sharp blow to the notion that AI’s costs can be summarized by just “carbon emissions (CO2)”!

It’s not as simple as saying, “Using renewable energy makes it clean!” The report highlights a hellish trade-off: switching to bioenergy could devour land and water at alarming rates. Notably, the “inference” we casually use daily accounts for 90% of the overall burden. Using the power of one video generation could equal that of sending 200,000 emails, turning it into a literal “resource-devouring monster”! We must confront this physical impact before the efficiency of technology breaks planetary limits!

🚀 What’s Next?

The evaluation criteria for AI models will likely shift from merely “performance” to “overall efficiency in water, land, and energy.” Dramatically reducing inference costs will become a top priority for tech companies by 2030.

💬 Haru-Same’s Takeaway

The waters where sharks dwell might end up as cooling water for data centers! Let’s rein in unnecessary generation and use AI wisely! Shark shark!

📚 Glossary

• Inference: The process of using a trained AI model to generate predictions or answers for new data—essentially the act of responding to user queries.

• Water Footprint: A measure of the total volume of freshwater used and polluted directly and indirectly in the production and consumption processes of goods and services.

• Carbon Footprint: A representation of the total greenhouse gas emissions produced throughout the lifecycle of a product or service, expressed in CO2 equivalents.

• Source: AI will consume as much water in 2030 as 1.3B people