AI-Specific Water & Cooling Crisis

AI data centers require far more cooling than traditional computing facilities because AI chips generate extreme heat. To prevent equipment failure, companies are shifting from air cooling to water-intensive liquid and immersion cooling systems. This has led to rapid increases in freshwater use, chemical pollution risks, and strain on local water supplies — often without updated regulations or public reporting requirements.

Key Facts

• AI data centers use about 34% more water than traditional data centers.
• Training and running AI models require advanced liquid cooling, not basic air cooling.
• A single AI interaction can consume up to one bottle of water for cooling.
• 80% or more of the cooling water can be lost to evaporation during operation.
• AI inference (everyday use, not just training) accounts for 40–60% of AI’s total water footprint.
• Major tech companies have reported sharp increases in water use directly tied to AI expansion.
• Current regulations do not limit or monitor AI data centers’ water withdrawals or chemical discharges.

Frequently Asked Questions (FAQs)

• Q: Why do AI data centers use so much water?

  A: AI chips generate extreme heat while training and running models. To prevent overheating, data centers use water-based liquid cooling systems, which are much more water-intensive than traditional air cooling.

• Q: Is the water only used when AI models are being trained?

  A: No. While training uses large amounts of water, everyday use of AI systems (called “inference”) accounts for 40–60% of AI’s total water footprint.

• Q: Does each AI prompt really use water?

  A: Yes. When accounting for cooling and electricity generation, a single AI interaction can require up to one standard bottle of water to keep servers from overheating.

• Q: Can AI data centers use recycled or non-drinkable water?

  A: Usually not. Most AI cooling systems require potable (drinkable) water to prevent corrosion and equipment damage, limiting reuse options.

• Q: What happens to the water after it’s used for cooling?

  A: Much of it is lost to evaporation. In some systems, 80% or more of cooling water can be released into the atmosphere, meaning it cannot be recovered.

• Q: Are chemicals used in AI cooling systems?

  A: Yes. Cooling systems often rely on glycol-based coolants and other chemical additives. Some operations are also linked to PFAS (“forever chemicals”), which persist in the environment and pose health risks.

• Q: Are AI companies required to report their water and chemical use?

  A: In many regions, no. Current regulations often do not require public reporting of water withdrawals, chemical discharges, or long-term environmental impacts from AI data centers.

• Q: Why is this an issue for local communities?

  A: Large water withdrawals can strain local supplies, especially during droughts, raise utility costs, and increase pollution risks — while providing limited long-term jobs or public benefits.

Resources/ Sources

• Bloomberg — AI Impacts: Data Centers and Water. AI data centers are increasingly being built in water‑stressed areas and are consuming massive amounts of water to cool servers and support AI workloads, adding pressure on already scarce local water supplies.
• University of Illinois — AI's Challenging Waters. AI data centers are rapidly increasing water demand because they use large amounts of freshwater for cooling high‑heat computing equipment, which contributes to water stress and environmental challenges in many regions.
• Energy & Environmental Study Institute — Data Centers and Water Consumption. Data centers are increasingly tapping freshwater sources for cooling, consuming millions to billions of gallons per year and threatening local water supplies as AI and server density grow.
• Inside Climate News — Illinois Data Centers and Water Use. Data centers in Illinois are drawing large amounts of water for cooling, prompting local communities and advocates to push for better planning and management of these growing demands on freshwater supplies.
• Water Education Foundation — Microsoft's Water Consumption Jumps Amid AI Boom. Microsoft's water consumption increased by about 34% in 2022, largely due to AI‑related data center operations, consuming 6.4 million cubic meters of water.
• Manufacturing Water Footprint
• Water Consumption Crisis
• Miguel Yanez-Barnuevo: Data Centers and Water Consumption. Data centers are increasingly tapping freshwater supplies to cool servers, consuming millions to billions of gallons annually — much of it evaporated or otherwise removed from local use — and this demand rises alongside energy use and AI workloads, putting pressure on water resources and nearby communities.
• Ana Pinheiro Privette: AI's Challenging Waters. AI data centers are growing rapidly and use vast amounts of freshwater for cooling servers, significantly contributing to water stress in many regions.
• Ganesh Hegde: Closed-Loop Cooling: Water Saver or Chemical Time Bomb? Closed‑loop liquid cooling systems, while saving water, rely on chemical additives like glycols and corrosion inhibitors that can build up and create toxic wastewater risks when discharged or leaked from AI data center cooling loops.
• Clean Water Action: Data Centers: A Threat to Minnesota's Water Fact Sheet. AI data centers' massive water use for cooling and the chemical treatment of that water can deplete and contaminate local freshwater supplies, threatening drinking water and ecosystems in Minnesota and similar regions.
• Science and Environmental Health Network: Water Use in Data Centers: What do we need to know? AI data centers consume huge amounts of fresh water for cooling and electricity generation—often in areas already facing water stress—which can drain local supplies and worsen regional water crises.
• Karen Hao: You're Thinking About AI and Water All Wrong. AI water use debates are often based on misleading estimates, and while data centers do use water for cooling, the actual impact varies widely by location and technology and is not always as extreme as some published figures suggest.
• Move Past Plastic: Data Center Toxins Infographic by Move Past Plastic. A 1-gigawatt AI data center can use 5 million gallons of water daily, relies on hazardous cooling chemicals, and strains local utilities and ecosystems while generating few permanent jobs.
• Bad Data Centers- Water
• Bad Data Centers- Cooling