Energy demand
AI Energy Usage
Individual prompts are a rounding error. The buildout behind them is a grid, water, siting, and transmission problem.
Electricity used by common AI tasks, set against everyday activities.
Your AI prompts are not boiling the oceans. You can relax.
The headlines will make you feel like a criminal for looking up a baseball player's stats. But if you are concerned, you likely aren't the problem.
A single text prompt uses roughly 0.3 Wh of energy. For context, that's about:
- One Google search
- Two minutes of light from a 9W LED bulb
- Nine seconds of watching TV
Even at 100 prompts per day, you are still using less energy than microwaving a burrito for 2 minutes.
So why the fuss? Because the real cost isn't the casual user. It's scale and complexity.
- Image generation: roughly 17x worse than text
- Long reasoning models: roughly 100x or worse
- Coding agent sessions: roughly 140x worse
- A 5-second AI video clip: up to roughly 1,700x worse
Then there are the true power users. The AI one-percenters. In one viral Reddit post a user boasted they used 1,156,308,524 tokens in a single month. That is as much electricity as a US home uses in roughly four months.
So is the average person the problem? No.
The first problem is scale: millions of images, millions of videos, always-on agents, enterprise automation.
But the bigger problem is the buildout behind it. It's a grid problem, a water problem, a siting problem, a transmission problem, a cooling problem, and a clean energy problem.
After a decade working in energy, I believe individual usage is a distraction. The real focus should be on the systems we are creating.
So should you punish yourself for casual AI use? No. Should you be paying attention to AI's real impact on the grid? Absolutely.
Next insight
The Data Center Buildout