Archive · 6 Jun 2025
AI's Unseen Cost: Energy and Water Consumption Rising
AI, like ChatGPT, needs massive data centers to work.
These centers use a lot of electricity, almost as much as a whole country.
They also need vast amounts of water for cooling, comparable to a small town.
AI’s growth is making this problem worse, even if companies try to use less water and more renewable energy.
This creates a challenge: we need to find a way to enjoy AI without harming the planet, which requires significantly increased clean energy production.
AI relies on data centers that consume vast amounts of energy and water.
Global data centers used approximately 460 TWh of electricity in 2024.
Data center electricity use could double by 2026.
Cooling data centers consumes significant amounts of water.
Efforts to create "green AI" are underway, but face challenges.
- Who
- AI developers and users, along with data center operators.
- What
- Data centers powering AI are consuming significant amounts of electricity and water.
- When
- Ongoing, with increasing impact.
- Where
- Globally, with a focus on the US and areas with large data centers.
- Why
- AI models and processes require huge energy and water resources for operation and cooling.
AI's Environmental Cost
Green AI Initiatives
Environmental Impact of AI
AI's Environmental Cost
The gap between AI's energy needs and clean energy supply is still wide.
Green AI Initiatives
Strive to reduce environmental impact by using reclaimed water, alternative cooling, and renewables.
- Global Data Center Electricity Use (2024)
- 460 TWh, equivalent to Sweden's annual power consumption
- Projected Data Center Electricity Use (2026)
- 1,000 TWh, almost matching Germany's yearly demand
- Energy to Train GPT-3
- 1.3 GWh, enough for 120 US homes for a year
- Energy per GPT-4 Prompt
- 0.0003 kWh, like a 60-watt bulb for 20 seconds
- Google Data Center Water Consumption (2024)
- Nearly 6 billion gallons
- Water per kWh of Electricity (Data Centers)
- 1.8 liters
