A Dive into the Water Footprint of AI.

Half of the world’s population could be living in areas facing water scarcity by as early as 2025. (UNICEF)

2027 Projections Match Lifetime Water Use of 110 Million Humans. (Privette, 2024)

TL : DR

AI is thirsty—and not just for data. Behind every chatbot conversation and model training lies a hidden environmental cost: water. From cooling data centers to generating electricity, AI’s water footprint is surging, with global usage projected to exceed that of some entire nations. This article dives into where and how AI consumes water, why it varies by region, and what companies and everyday users can do to curb its impact. Want to use AI more sustainably? Start by asking smarter, lighter, and more intentional questions.

Abstract
The rapid expansion of artificial intelligence (AI), particularly large language models (LLMs), has introduced significant environmental concerns, notably regarding water consumption. This paper examines the direct and indirect water usage associated with AI, encompassing data center operations, energy production, and model training and inference. By synthesizing recent empirical studies and industry reports, we provide a comprehensive analysis of AI’s water footprint and propose strategies to mitigate its environmental impact.

1. Introduction
Artificial intelligence has become integral to various sectors, from healthcare to finance, due to advancements in machine learning and data processing capabilities. However, the environmental implications of AI, especially its water consumption, have garnered increasing attention. Water is essential for cooling data centers and generating electricity, both critical for AI operations. Understanding AI’s water footprint is crucial for developing sustainable practices in technology deployment.

2. Understanding AI’s Water Footprint
The water footprint of AI encompasses both direct and indirect consumption:

Direct Water Use: Data centers require substantial water for cooling systems to maintain optimal operating temperatures.
Indirect Water Use: Electricity generation for powering data centers often involves water-intensive processes, especially in thermoelectric power plants.
These combined factors contribute to AI’s significant water footprint.

3. Quantifying AI’s Water Consumption
Recent studies have highlighted the substantial water usage associated with AI:

Model Training: Training large AI models like GPT-3 can consume approximately 700,000 liters of water, primarily for cooling during intensive computational processes (Li et al., 2023).
Data Center Operations: A 100 MW hyperscale data center can directly consume around 2.5 billion liters of water annually for cooling purposes (Impax Asset Management, 2024).
Global Projections: AI’s global annual water consumption is projected to reach between 4.2 and 6.6 billion cubic meters by 2027, surpassing the annual water usage of countries like Denmark (University of Illinois, 2024).

4. Regional Variations in Water Usage
AI’s water footprint varies significantly across regions due to differences in climate, energy sources, and cooling technologies:

Water Usage Efficiency (WUE): WUE measures the liters of water used per kilowatt-hour (kWh) of energy consumed. In Microsoft’s global data centers, WUE ranges from 1.8 to 12 liters per kWh, with variations attributed to local environmental conditions and infrastructure (OECD, 2023).
Case Study – Iowa: Microsoft’s data centers in Iowa, used for training GPT-4, significantly increased the company’s water consumption by 34% from 2021 to 2022, highlighting the regional impact of AI operations (AP News, 2023).

5. Strategies for Mitigating AI’s Water Footprint
To address the environmental challenges posed by AI, several strategies can be implemented:

Optimizing Training Schedules: Scheduling AI model training during cooler periods or in regions with lower water stress can reduce cooling requirements (Li et al., 2023).
Enhancing Cooling Technologies: Adopting advanced cooling methods, such as liquid immersion cooling, can improve efficiency and reduce water usage (Submer, 2023).
Improving Energy Sources: Transitioning to renewable energy sources that require less water for electricity generation can indirectly decrease AI’s water footprint.
Transparency and Reporting: Encouraging companies to disclose water usage data promotes accountability and informs sustainable practices.

6.1 User-Level Strategies

Use AI Judiciously: Avoid redundant or exploratory prompts that require significant computation unless necessary. Streamline requests to reduce model processing time.
Opt for Local or Lightweight Models: When available, prefer using smaller, less resource-intensive models or local inference systems (on-device AI) that don’t require energy-intensive cloud computing.
Consolidate Tasks: Instead of multiple fragmented queries, plan and structure a single prompt that achieves the end goal more efficiently.
Promote Transparency: Use AI models that disclose their environmental impact or provide efficiency metrics. Favor platforms that demonstrate commitment to sustainability.
Engage During Off-Peak Hours: Some models may be more efficient or less water-intensive during cooler nighttime hours or in specific geographies.

6.2 Sample Prompts to Use AI Sustainably

“Summarize this 20-page report and highlight only the risks, dependencies, and mitigation strategies in bullet points.”
“Highlight only the differences between these two documents—no full comparison needed.”
“Generate a week’s worth of content ideas in one list, categorized by theme.”
“Give me a summary of how I’ve used this AI in the past week and suggest how I can reduce its environmental impact.”
“Create a low-resolution storyboard with 3 key frames to visualize this scene.”

7. Conclusion
As AI continues to evolve and integrate into various aspects of society, its environmental impact, particularly concerning water consumption, cannot be overlooked. By understanding and addressing the water footprint of AI, stakeholders and individual users alike can develop and implement strategies that balance technological advancement with environmental sustainability.

References
AP News. (2023). Artificial intelligence technology behind ChatGPT was built in Iowa – with a lot of water.
Impax Asset Management. (2024). Addressing the hidden water footprint of data.
Li, P., Yang, J., Islam, M. A., & Ren, S. (2023). Making AI less “thirsty”: Uncovering and addressing the secret water footprint of AI models. arXiv preprint arXiv:2304.03271.
OECD. (2023). How much water does AI consume? The public deserves to know.
Submer. (2023). Datacenter water usage: Where does it all go?
Ana Pinheiro Privette,University of Illinois. (2024). AI’s challenging waters.