If I ask 10 to 15 questions to ChatGPT, it will consume more than 1L of water to answer them. This consumption would be similar to other AI models. The primary reason behind this consumption is linked to the energy usage and cooling of their data centres. In this article, we will explore how the tech boom is draining our water and what we can do about it.

The calculation of water consumption is even provided as an answer by ChatGPT as you can see above. Every query takes around 100-500 Wh of energy, and for 10 queries it can take around 1000-5000kWh [1Kwh-5kWh] of energy. For each kWh, the water consumption is around 0.79 litres. This consumption is not limited to AI-based data centres; the data centres have been around for the last two decades.

What are Data Centres and Why do they need water consumption?

You might be using various websites to read articles, YouTube to watch videos and social media to interact with your friends. Every data that you receive [i.e. blog, video and social media content in the form of images, posts and reels] has to be stored somewhere. The storage unit is called as server. Now these servers need additional components to operate like routers and switches to interact with the internet, firewalls for cyber security, backup storage drives, etc. All these once placed together at a physical location, is known as Data Centre.

Example: A data centre of Imgix. You can explore the above data centre here.

Modern data centres have high-density equipment that generates substantial heat, requiring significant power not only for computing but also for cooling systems. In fact, cooling can consume up to 40% of a data centre’s total energy usage. Maintaining optimal temperatures—ideally between 68°F and 71°F (20°C to 21.6°C)—is crucial to prevent hardware failures and ensure the longevity of servers. Even minor deviations can lead to catastrophic failures, resulting in costly downtime and equipment replacement.

There are two primary methods of cooling in data centres:

1. Air Cooling: This traditional method uses fans and heat sinks to dissipate heat by moving air over the components. While widely used, air cooling is becoming less efficient as servers produce more heat than they did a decade ago.
2. Liquid Cooling: This method is gaining traction due to its efficiency in heat removal.

• Water-Cooled Racks: Water circulates near hot components to absorb heat. While effective, there’s a risk of leaks, and mixing water with electrical equipment poses safety concerns.
• Liquid Immersion Cooling: Servers are submerged in a thermally conductive but electrically non-conductive liquid (dielectric coolant). This method provides excellent cooling performance without the risk of electrical hazards from water.

A simple architecture diagram of a Water-Cooled Racks Data centre. Key components involved in data centre cooling systems include:

• Chillers: Devices that cool liquid (usually water) via refrigeration cycles, supplying chilled water to cool air in data centres.
• Cooling Towers: Structures that expel waste heat by cooling water through evaporation, aiding chillers in heat dissipation.
• Computer Room Air Conditioning (CRAC) Units: Air conditioning units using refrigeration cycles with internal compressors to cool air in server rooms.
• Computer Room Air Handler (CRAH) Units: Units using fans and cooling coils with chilled water from chillers to cool air in data centres.

The demand for freshwater-based cooling systems is high, as it has more cooling effects compared to air. The water needs to be fresh drinkable water [even more cleaner than that] so that the minerals inside the water would not deteriorate the equipment.

How the tech boom is draining our water in the USA?

Major companies related to technology and Artificial intelligence have headquarters and data centres in the USA. Below is the list of AI companies in the USA.

The United States has the highest number of data centres in the world. Below is a list of countries with the most data centres in the world.

More data centres mean more cooling requirements. And more cooling requirements means more water consumption. The computing power required by AI doubles every 100 days and will be 1 million times over the next 5 years. It is estimated that by 2027, the global demand for water consumption by AI will be 6.6 billion cubic meters of water [Source: Making AI Less “Thirsty”]. This would be deeply concerning at a time of growing water scarcity around the world.

What can be done to reduce water consumption due to AI?

We have already discussed how the tech boom is draining our water. There are some generic ways to solve this issue that is applicable to all industries, like:

• Improve Energy Efficiency: By using the latest technology and advanced systems to improve efficiency and thus reduce water consumption.
• Shift to Renewable Energy: Shifting the energy usage to renewable energy, would result in less pollution, thus indirectly decreasing water consumption.
• Recycle and Reuse Water: Reusing water that is used and not evaporated by cooling towers, and also the water from rain can reduce the water consumption from natural water resources.

Apart from the above-mentioned ways, there are other methods that data centres can implement:

• Optimize Data Center Locations: Move Data Centers to Coastal Regions, in cooler climates or near renewable energy sources.
• Use AI to Manage Data Center Efficiency: AI can be utilized to investigate, research, and find optimized ways to use water for cooling, save water for reuse and predict cooling demands and prevent overuse of water.
• Use better AI models that are energy efficient: By reducing model complexity, and by using model distillation techniques to create smaller and more efficient versions of AI models.
• Reducing Data Centers by Sharing: If the same operations can be performed by a data centre, then other companies can utilize them rather than building their own data centres.

How Leading AI Companies Are Reducing Water Consumption in Data Centers?

We have already discussed how the tech boom is draining our water and how to solve this issue. The good part is that the giant industries are already working on it:

• Google has already committed to running data centres on 100% renewable energy by 2030.
• AWS has committed to achieving 100% renewable energy by 2025. AWS has implemented sustainable water management practices, particularly in regions where water is scarce, using rainwater harvesting and water reuse systems in some of its facilities.
• OpenAI uses Microsoft Azure’s cloud infrastructure, which enables resource sharing across organizations, reducing the overall water and energy footprint.
• IBM data centers, particularly in locations like New York, have implemented water recycling systems for cooling purposes, significantly reducing the use of freshwater
• Oracle has invested in water-efficient cooling systems, including water-side economizers, which reduce water consumption by utilizing outside air and water for cooling during cooler months.
• Apple strategically locates some of its data centres in regions where natural cooling methods can be used, reducing water consumption.
• Meta uses customized cooling systems that adjust airflow and cooling power based on real-time demand, reducing both energy and water use.
• Through its Green Cloud initiative, SAP provides cloud services that are designed to be energy- and resource-efficient. SAP’s data centres follow strict sustainability certifications like LEED (Leadership in Energy and Environmental Design), which include guidelines for efficient water use.

Also Read: The Impact Of Artificial Intelligence On Electrical Energy Grids