Illustrative clock based on a published July 2026 daily estimate.
Constant assumed rate, not a live provider feed.
Baseline from Tokens Per Day, verified July 16, 2026.
How many imaginary tanks could that energy warm?
We cannot count goldfish killed by AI. Instead, use this pretend experiment to see how changing the assumptions changes the result. Real fish are not involved.
A 10-gallon tank needs about 44 watt-hours of heat to rise 1°C, assuming no heat escapes. This is physics, not a fish-health threshold.
These are deliberately adjustable hypothetical values. The energy-per-token assumption is not calibrated to a particular provider or workload, and token types are not interchangeable.
Where does the energy really go?
AI servers turn most of the electricity they consume into heat. Cooling systems move that heat to air or water, and the power grid must supply the electricity. Real environmental effects depend on the energy source, local water conditions, equipment production, and what the AI is being used for.
Show the calculation and limitations
Daily energy = assumed daily tokens ÷ 1,000 × assumed Wh per 1,000 tokens. Imaginary transferred energy = daily energy × selected transfer fraction. A tank's energy to warm by 1°C = gallons × 3.78541 kg/gallon × 4,186 J/(kg·°C) ÷ 3,600 J/Wh. Tank equivalents = transferred energy ÷ energy per tank. This model does not estimate fish mortality, actual cooling-water consumption, or measured global AI energy.
There is no complete public real-time global token feed. The published 360.4 trillion/day figure is a third-party model from July 2026. The watchboard spreads that daily baseline evenly across a UTC day for illustration, not because usage is uniform.
Sources: Tokens Per Day global estimate and Joule (2026) study of AI inference energy. Actual global throughput and energy per token remain uncertain.
Follow the Energy →Play Don't Kill Your Goldfish →