🔆 Stay hydrated: AI's thirst for water

🗞️ Issue 33 // ⏱️ Read Time: 5 min

Hello 👋

Just as plastic revolutionized industries and daily life in the 20th century, Artificial Intelligence (AI) is reshaping our world today. However, the rapid adoption of plastic led to unforeseen environmental consequences, such as microplastics in vital human organs 🧠. As we embrace AI, we have the opportunity to learn from history to avoid repeating the same ecological mistakes. By 2030, energy requirements for AI are projected to surge by 160%, highlighting the urgent need for sustainable practices in AI development and deployment. This is part one of three on AI’s environmental impact, and we are starting with the resource we all need to survive: Water. 

Water Consumption: The Hidden Cost of Cool Computing

The Issue

AI relies heavily on data centres, which require massive amounts of water for cooling. Reports show that an average data centre consumes around 300,000 gallons (approximately 1.14 million litres) of water daily, which is comparable to the daily usage of 1,000 U.S. households. 

AI's water footprint includes both water withdrawal (freshwater taken from sources) and water consumption (water not returned to the environment). While withdrawal indicates competition for water resources, consumption reflects the long-term impact on water availability, both crucial for understanding AI's full environmental impact.

Water covers approximately 70% of the planet - so it seems we shouldn’t worry too much about running out of it, right? The thing is, only 3% of that water is freshwater: The kind of water that we can drink to quench our thirst, and use to irrigate our farms. Only one-third of that freshwater is available to us, as the rest is safely frozen in glaciers - meaning that we cannot consume it.

Why does AI need water?

AI computations generate significant heat. Data centres use two main types of cooling systems: cooling towers and outside air cooling. Both use water, either through evaporation in cooling towers or for humidity control in air cooling systems. The more powerful the AI systems, the more heat generated, and the more water needed for cooling.

Water usage in AI can be categorized into three scopes, similar to carbon emissions:

• Scope-1: On-site water use for cooling

• Scope-2: Off-site water use for electricity generation

• Scope-3: Water used in the supply chain (e.g., for manufacturing chips)

Major Contributors

Some of the biggest consumers of water in this context are the big tech companies - not very surprising, right? 

• Google: In 2023, Google’s data centres consumed 6.1 billion gallons of water (27 million cubic meters)—17% more than the previous year. And up from 4.3 billion gallons in 2021.

• Microsoft: In 2023, Microsoft's water consumption increased by 23% from the previous year - from 6.4 to 7.9 million cubic meters. This was largely due to the expansion of its AI and cloud services. 

• OpenAI: “GPT-3 needs to ‘drink’ (consume) a 500ml bottle of water for roughly 10-50 responses”, depending on when and where it is deployed.

• Meta: In 2023, Meta’s data centre water withdrawal and consumption totalled 6.8 million cubic meters, an 11% increase from 2022.

Collectively, the combined on-site and off-site global water withdrawal of Google, Microsoft, and Meta reached an estimated 2.2 billion cubic meters in 2022. This is equivalent to the total annual water withdrawal of two Denmark-sized countries.

The Plastic Parallel

Like the invisible microplastics now permeating our oceans, the water consumption of AI often goes unnoticed. Just as plastic production strained water resources, AI's thirst poses similar challenges. About two-thirds of people worldwide face serious water shortages for at least one month yearly. By 2030, nearly half the global population could struggle with severe water problems. To prevent this, we need to find new ways to use less water without hindering economic growth and productivity.

Future Projections 

By 2027, the global AI demand may be accountable for up to 6.6 billion cubic meters of water withdrawal. This is more than the total annual water withdrawal of half the United Kingdom.

Microsoft, Google, and AWS all committed to being water-positive by 2030, pledging to replenish more water than they consume globally. However, critics fear that these companies are pushing a plan to allow companies to report emissions numbers that bear little relation to their real-world pollution and not fully compensate for those emissions.

While carbon emissions are often included in AI model cards, water usage information is typically missing. This lack of transparency makes it difficult to fully assess and address the environmental impact of AI.

The public should maintain a critical perspective on the environmental programs and net-zero promises made by tech companies. These initiatives are not always designed to genuinely reduce environmental impact. In some cases, they may be driven more by reputation management, marketing strategies, or public relations efforts rather than a true commitment to sustainability.

Addressing the Challenge

• Increase transparency in reporting water usage data for AI models and data centres.

• Develop technologies to reduce or maintain water consumption, without compromising performance. 

• Develop policies that support the innovation of these technologies.

• Schedule AI workloads during cooler hours of the day or in water-efficient locations.

• Implement air-based or recycled water cooling systems in data centres.

Until next time.
On behalf of Team Lumiera

Emma - Business Strategist
Sarah - Policy Specialist
Allegra - Data Specialist

Lumiera has gathered the brightest people from the technology and policy sectors to give you top-quality advice so you can navigate the new AI Era.

Follow the carefully curated Lumiera podcast playlist to stay informed and challenged on all things AI.