The hidden cost of AI: electricity and water
Big idea: AI runs on real electricity and real water, so we should use it wisely.
The story in short
In the hottest summer anyone could remember, the society borewell dried up and Prakash uncle made a water-tanker rota. Riya read that a chatbot may use about half a litre of water for every few dozen answers. On a family day at a data centre in Whitefield, the children saw ID checks, phone lockers and a two-door “airlock”; cold aisles in front of the computer racks and hot air behind them; and white clouds of evaporating water on the rooftop cooling towers. Back home, Happy asked a chatbot “what is 2 + 2?”, and Riya just held up her calculator.
The idea
- Data centre: a building full of computers (servers, often with GPU chips) that run apps and AI.
- Waste heat: nearly all the electricity used ends up as heat.
- Evaporative cooling: evaporating water to carry the heat away.
- Training: like building a stadium once: a huge, one-time energy cost.
- Inference: like switching on the floodlights for every match: a little energy each time someone asks, repeated billions of times.
- Physical security: protecting the machines themselves (no tailgating through the airlock!).
Use AI wisely
- Ask: is this use worth this cost?
- Use a calculator, a search or your own brain for simple things. Your brain runs on about 20 watts, “and idli”, says Mitthu.
- AI can also save resources: forecasting floods, finding water leaks, helping farmers.
The book’s numbers are estimates, and companies’ own figures are still debated.
Remember: nothing in the cloud is free. Make every drop and every watt count.
Check yourself
Which uses more energy in total over time for a popular chatbot?
Where does most of the electricity used by a data centre end up?
You need to add 37 + 48. What is the most sensible tool?
More for you: 🎒 Students · 🧑🏫 Parents and teachers