This calculator shows thought experiments: What if ALL the electricity for your digital life came from pure coal? Or pure nuclear? Plus the very real water footprint of the data centers and power plants behind your digital life. The energy scenarios are worst-case / single-source — real grids are a mix. The water numbers reflect actual industry averages.
Device usage assumes an average laptop drawing ~40 W. Streaming adds the data center energy (~0.04 kWh/hour) on top of your device. AI usage estimates ~20 queries per hour at ~0.5 Wh each (active use with longer responses), based on published energy estimates for large language models.
The coal scenario uses 820 g CO2/kWh (IPCC lifecycle median for coal power). The nuclear scenario uses 2.8 mg spent fuel/kWh (total spent nuclear fuel, from whatisnuclear.com). Spent fuel contains ~1% plutonium-239, a weapons-grade material with a lethal inhaled dose of ~20 μg (ICRP). This waste remains dangerously radioactive for over 100,000 years.
Water waste comes from two sources: data center cooling (evaporative cooling towers use ~1.8 L/kWh, industry average WUE) and power plant cooling (thermal plants use ~2 L/kWh to cool steam turbines). AI training water is amortized: the industry trains ~10 frontier models per year at ~30 GWh each, consuming ~540 million liters, shared across ~500 million daily AI users (Li et al. 2023, “Making AI Less Thirsty”, UC Riverside).
The hidden cost: hardware waste. Data centers replace GPUs every 2–3 years to stay competitive. Nvidia ships ~5 million data center GPUs annually, each containing rare earth metals and toxic materials. That’s ~15,000 tonnes of high-tech e-waste per replacement cycle — and the pace is accelerating as the AI race intensifies. This e-waste is not included in the numbers above because per-person allocation is tiny (~30 g/year for an average AI user), but at industry scale it’s a growing environmental crisis.
Sources:
- Nuclear waste per kWh — whatisnuclear.com
- Streaming video carbon footprint — IEA
- AI energy usage estimates — Epoch AI
- CO2 emissions from electricity — World Nuclear Association
- Global electricity review — Ember
- Making AI Less “Thirsty” — Li et al. 2023, UC Riverside
- Carbon Emissions and Large Neural Network Training — Patterson et al. 2021