What does an AI prompt cost the planet?

An AI prompt doesn't feel physical. But every message runs on servers in a data center, and those servers draw electricity and need cooling. Multiply that by billions of prompts a day and it adds up.

From tokens to electricity

Language models process text as tokens, and the work a model does grows with the number of tokens it handles. More tokens in your prompt means more computation, which means more energy. Published estimates for a single prompt vary widely, from a fraction of a watt-hour for a short question to far more for long, reasoning-heavy requests, because the answer depends on the model, the hardware, and the length of the conversation.

From electricity to CO₂ and water

The carbon cost depends on the grid powering the data center: the same kilowatt-hour can be close to zero carbon on renewables or much higher on fossil fuels. Water comes in mostly through cooling, as many data centers use evaporative cooling to keep servers at safe temperatures.

How TerraWatch estimates it

To give you a consistent yardstick, TerraWatch uses fixed, published-research-based figures:

• 3.5 Wh of energy per 1,000 tokens
• 380 g of CO₂ per kWh, roughly an average grid
• 1.8 L of water per kWh for cooling

By that math, a 500-token prompt works out to about 1.75 Wh, 0.67 g of CO₂, and 3 mL of water. Treat these as estimates for comparing prompts, not a utility bill: real numbers depend on the provider, model, and location.

What actually makes a difference

One prompt is tiny. Your habits across thousands of prompts are not. The levers you control:

• Send fewer tokens.Cut filler, repetition, and pasted context the model doesn't need. Our guide to reducing token usage covers the habits.
• Avoid retries. A clear prompt that works the first time beats three vague ones.
• Start fresh chats. Long threads make every new message heavier to process.
• Ask for what you need. Shorter answers cost less to generate too.

TerraWatch shows the estimated energy, CO₂, and water behind each prompt as you write it, and compresses it in one click, so the savings add up without you having to think about it.