The Hidden Cost of AI: How it’s Draining our Water

AI has revolutionized technology, transforming industries and enabling advanced data analysis. However, this progress comes with an overlooked environmental cost—water consumption. Although OpenAI has not publicly disclosed the exact training dates, GPT-3 was officially released in June 2020. During its training in Microsoft’s U.S. data centers, it is estimated to have consumed around 700,000 liters of clean water—a figure based on Microsoft’s reported Water Usage Effectiveness (WUE) of 0.55 liters per kilowatt-hour and GPT-3’s energy use of 1,287 megawatt-hours. By 2027, the global AI-related water usage could reach 4.2–6.6 billion cubic meters, which is more than Denmark’s annual consumption.

This rising demand is especially concerning in water-scarce regions like Arizona, USA and Maharashtra, India. Since only 2.5% of Earth’s water is freshwater—and an even smaller portion is accessible for human use—AI companies must adopt sustainable practices to avoid worsening global water scarcity.

The Environmental Impact of AI Data Centers' Water Use

Generative AI operates virtually but has a significant environmental impact, particularly in terms of water consumption. Each AI prompt requires thousands of computations on servers housed in massive data centers. These centers generate substantial heat, requiring cooling systems to prevent overheating. Water-based cooling is widely used, where heat is absorbed and released via cooling towers, similar to how sweating cools the human body.

Producing a 100-word AI-generated email consumes about a bottle of water. Training AI models over months requires even more resources, including vast amounts of electricity and water. As AI adoption grows, companies are rapidly expanding data centers, intensifying concerns about their ecological footprint.

The Scale of the Problem

The water usage of AI data centers is immense. Google’s data centers consumed 6.4 billion gallons of water in 2023, up from 4.3 billion gallons in 2021. Microsoft’s usage rose from 1.7 billion gallons in 2022 to 2 billion in 2023. Amazon reported a WUE of 0.18 liters per kilowatt-hour, in the year 2023, but did not disclose total consumption. Training large AI models alone can require millions of liters of water. As AI advances, these figures are expected to increase, raising serious environmental concerns.

A key challenge in addressing AI’s water footprint is the lack of transparency. Many tech firms prioritize innovation over sustainability, making it difficult to track actual water use. Without accurate data, AI researchers struggle to develop sustainable solutions, policymakers cannot enforce regulations, and the public remains unaware of AI’s role in water consumption. This lack of accountability hinders environmental mitigation efforts.Some companies have started disclosing their water usage. While Google reports using non-potable water for cooling in approximately 25% of its data centers, the majority—around 75%—still rely on potable water. Amazon highlights efficiency improvements but lacks detailed water usage reports. Microsoft has pledged future data centers will operate with zero water cooling, though specifics remain unclear.

AI Data Center Cooling Systems

Water is essential for cooling AI data centers, ensuring high-performance computing systems don’t overheat. Two primary cooling methods are used:

1. Water-Based Cooling: This method circulates water through cooling towers or heat exchangers to absorb and dissipate heat, requiring a continuous water supply. It is energy-efficient but increases water consumption.

2. Air-Based Cooling: This system relies on fans and ventilation to regulate temperatures. While it reduces water usage, it is often less effective for cooling high-performance AI workloads.

   
   

AI Data Centers' Water Consumption Compared to Other Industries

AI data centers require massive amounts of water for cooling, with some facilities consuming between 11 and 19 million gallons daily—comparable to the water usage of a city with 30,000 to 50,000 people. However, when placed alongside other industries, AI’s water consumption is relatively smaller. Agriculture remains the most water-intensive sector, particularly in livestock farming, where large quantities of water are needed to produce animal feed. Similarly, semiconductor manufacturing relies heavily on water, contributing to concerns about resource depletion. While AI-driven technologies add to water demand, they are just one part of a broader discussion on industrial water use.

AI Data Centers in Water-Stressed Regions

Another growing concern is the establishment of AI data centers in regions already facing water scarcity. For example, a group of data centers in Iowa used 6% of the local water supply during the training of GPT-4. Similarly, major tech companies are setting up data centers in areas like Arizona in the U.S. and Maharashtra in India, which are already grappling with severe water shortages. This expansion could worsen existing water crises, negatively impacting communities and ecosystems that rely on these scarce resources. This underscores the urgent need for better planning, regulatory policies, and sustainable water management approaches.

AI Applications in Water Conservation

Ironically, AI has the potential to enhance water conservation efforts. In agriculture, AI-powered systems can optimize irrigation by analyzing factors like soil moisture, weather patterns, and crop needs, resulting in more efficient water use. Additionally, AI models are being developed to predict droughts, aiding in better management of water resources and preparation for scarcity; MIT researchers created an AI model that analyzes satellite-based temperature and humidity data to enhance the accuracy of drought forecasts. For example, the ITIKI initiative in Africa blends artificial intelligence with traditional knowledge and real-time sensor data to help rural communities anticipate and prepare for drought conditions. When applied responsibly, AI could serve not only as a driver of technological progress but also as a tool for sustainability, addressing some of the environmental challenges it creates.

Water Consumption in Streaming Services

In addition to AI, other digital technologies like streaming services (such as Netflix and YouTube) contribute significantly to water consumption. The data centers that power these platforms generate considerable heat, requiring continuous cooling, often through water-heavy systems. While detailed water usage statistics for streaming services are less frequently disclosed compared to AI, the environmental impact of these data centers remains considerable. As demand for high-definition streaming and cloud storage increases, the water footprint of digital media consumption is expected to rise as well.

The Way Forward: Finding a Balance for Sustainable AI

As AI and cloud computing expand, balancing technological progress with sustainability is crucial. Tech companies must improve transparency, adopt innovative cooling solutions, and implement stricter conservation strategies. Given global water scarcity, optimizing data center water use is essential for a sustainable digital future.

It is also important for AI companies to engage with local communities where data centers are expanding. By understanding the unique water needs of these areas and working together on sustainable solutions, tech companies can ensure that their growth doesn’t come at the cost of essential resources. By combining innovation with thoughtful policies and community collaboration, we can reduce AI's environmental footprint while still pushing forward with technological advancements.