Lesson 69 · 3 min
AI's electricity and water
One AI answer uses a small amount of energy, but billions of answers, plus training, add up to a real load on power grids and water.
AI runs on chips called GPUs in large data centers. They draw a lot of electricity, and the heat they make is often removed with cooling systems that use water.
Training a big model is a one-time job that can use as much electricity as many homes use in a year. Answering questions, called inference, uses far less each time, but it happens billions of times a day.
Estimates for a single chatbot reply vary widely, because they depend on the model, the length of the answer, the hardware and the data center. Most put a typical text reply in the range of a few seconds to a few minutes of running a light bulb. Images, video and long reasoning cost more.
Companies are building more efficient chips and cooling, and some use cleaner power. You can help a little too: pick a fast model for easy jobs, skip needless regenerations, and do not make videos just to throw them away.
Do it now
Use the right size of tool
- 1. Look at how you used AI this week.
- 2. Mark the tasks a fast or small model could have done, or that a quick search would have answered.
- 3. Next week, save the large reasoning or video tools for jobs that really need them.
Quick check
Keep going
Next →Where AI actually runsWords from this lesson
- GPU
- A chip that does huge numbers of small calculations at once. First built for video games, now the main engine for AI.
- Inference
- The moment a trained model is actually used to produce an answer, as opposed to the earlier learning phase.
- Training
- The process of feeding a model many examples and adjusting its internal numbers until its outputs improve.