Each chatbot reply uses real electricity, real cooling water, and real grid capacity. A single image generation can use as much energy as charging your phone. None of this means you should never use AI — but knowing the cost helps you choose when it's worth it and push companies to do better.
Some examples
A long chat session can use 10x the energy of a Google search.
Image and video generation are dramatically more expensive than text.
Smaller, on-device models exist for many tasks and use far less energy.
Some providers publish per-query carbon estimates — look for them.
Try it!
Pick one task you usually ask AI for that a smaller tool could handle. Switch for a week and notice if you missed anything.
End-of-lesson check
15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-ai-environmental-cost-teens-final2-teen
Each chatbot reply uses what real-world resources?
Electricity, cooling water, and grid capacity
None
Paper and ink
Disk space only
A single image generation can use roughly as much energy as what?
Charging your phone
Lighting a city
One LED for a year
Nothing
How does a long chat session compare to a single Google search?
Roughly 10x the energy
Roughly equal
Much less
One million times more
Which modality is dramatically more expensive than text?
Image and video generation
Plain text replies
Markdown formatting
Code highlighting
What is one energy-saving alternative for many tasks?
Smaller, on-device models
Always use the largest cloud model
Use multiple cloud models in parallel
Generate ten variations and pick one
What can some providers publish to help users understand impact?
Per-query carbon estimates
The CEO's calendar
The model's birthday
The number of pixels
Which heuristic captures 'use the smallest tool that works'?
If a calculator does it, don't ask GPT-5
If you can ask GPT-5, do
Always pick the biggest model
Avoid all tools
Why do data centers also use real water?
For cooling the hardware
To wash the servers
For decoration
To make the AI human
What does 'training emissions' refer to?
Greenhouse gases produced while training a model
Daily inference emissions
Emissions from the user's laptop
Emissions from data labelers' coffee
What does 'inference cost' refer to?
Compute used per request after the model is trained
Cost to design the chip
Cost to draw the logo
The user's electricity bill
Which decision lowers your AI footprint with no real cost to you?
Switching from a model to a calculator for arithmetic
Asking the model to think out loud forever
Generating ten images instead of one
Repeatedly regenerating identical text
Why does Google search use less than a long chatbot session?
Search is heavily optimized; chat runs a large generative model per turn
Search is older
Search uses no servers
Chat uses no servers
Why isn't the lesson telling you to never use AI?
Because mindful use plus pressure on companies is more realistic than abstinence
Because AI has zero impact
Because abstinence is illegal
Because AI cleans the planet
What's a useful one-week experiment from the lesson?
Replace one AI task with a smaller tool and notice if anything was missed
Stop drinking water
Disable the internet
Generate twice as many images
Why is collective pressure on companies part of the answer?
Providers respond to demand for greener compute and transparency
