Microsoft
Use closed-loop cooling, expand reclaimed-water systems, and replenish water in the watersheds where facilities operate.
How AI Uses Water
AI depends on physical infrastructure. Follow water through chip manufacturing, electricity generation, and data-center cooling—then see what companies can measure, reduce, and disclose.
Play the water quiz
AI Uses Water in Three Places
Water enters the system before a server answers a question: manufacturing creates chips, power plants generate electricity, and data centers manage heat.
- Chip manufacturing
- Electricity generation
- Data-center cooling
AI Server Electricity Grew More Than 20×
Annual electricity used by AI servers grew from under 2 TWh in 2017 to over 40 TWh in 2023.
| Category | Value |
|---|---|
| 2017 | Under 2 TWh |
| 2023 | Over 40 TWh |
Global Data-Center Water Use Is Rising
Global data-center water use was about 560 billion liters in 2023 and is projected to reach about 1.2 trillion liters in 2030.
| Category | Value |
|---|---|
| 2023 | 560 billion liters |
| 2030 | 1.2 trillion liters |
Water Use Happens in Three Places
Roughly two-thirds of data-center water use is connected to electricity and energy, one-quarter happens through on-site cooling, and the remainder comes from semiconductor manufacturing.
| Category | Value |
|---|---|
| Electricity and energy | Roughly two-thirds |
| On-site cooling | About one-quarter |
| Semiconductor manufacturing | The remainder |
Most U.S. Data-Center Water Use Happens Off Site
In 2023, U.S. data centers used 66 billion liters on site and nearly 800 billion liters through the electricity that powered them.
| Category | Value |
|---|---|
| On site | 66 billion liters |
| Electricity generation | Nearly 800 billion liters |
Companies Do Not Publish AI’s Exact Water Share
Companies report facility or fleet totals, but they generally do not separate the water used for AI workloads from other computing. Without workload-level reporting, an exact AI share cannot be verified.
Companies Can Reduce Water Use
Better cooling, cleaner water inputs, careful siting, and watershed work can reduce pressure. Each action still needs a clear place, timeframe, and measured result.
Improve cooling efficiency, use alternatives to drinking water where appropriate, and support watershed replenishment projects.
AWS
Track water-use efficiency, expand reclaimed-water cooling, and invest in community water and watershed projects.
Meta
Design efficient cooling systems, reduce operational water intensity, and restore water through local watershed projects.
Could AI Compute Move Into Orbit?
Elon Musk is betting on orbital AI data centers powered by near-constant solar energy. In space, servers would reject heat through radiators instead of terrestrial cooling systems.
Companies Should Publish Their Water Use
A useful public record should make each facility understandable on its own and connect efficiency claims to local water conditions.
- Facility-level withdrawal
- Facility-level consumption
- Water source
- Watershed stress
- Cooling efficiency
- AI workload share
