If AI decisions got radically cheaper, would AI use less energy — or more?
A scenario study on the Jevons paradox, applied to AI decision inference. Move the sliders below — this is not a forecast.
TypeSafe AI says it named Jev after William Stanley Jevons, expecting intelligence to follow coal's path: demand grows as cost falls (TypeSafe blog, FAQ). This page asks whether that is what happens to energy.
What this is (and isn't).
This is a scenario-analysis tool, not a forecast. Every number below is a cited figure (with a source you can open), a value derived from cited figures by a stated formula, an explicitly labeled scenario assumption you can change, or a live output of the model. No figure for JEV's actual energy use in watt-hours has been published by TypeSafe AI or measured independently, as far as this project could find — the "energy ratio" parameter below is reconstructed from three indirect proxies (price, latency, token count), not a measurement. No number on this page describing JEV's energy consumption is a direct measurement.
Presets
Each preset sets every slider to specific, checked values (see DECISIONS.md in the repository) — they are illustrations of the model's own math, not independent predictions.
Scenario parameters
measured / vendor claim / estimate badges mean the default value traces to a specific source (click the source link to open it). scenario means there is no measured value — you are exploring an assumption. derived means it's computed from other cited numbers (see the formula in METHODOLOGY.md).
Result
Where's the threshold? (E1/E0 vs. elasticity)
At the current r, q, s, c, this line shows how the outcome changes as elasticity (eps) varies. The dashed red line marks the breakeven elasticity, where energy returns exactly to E0: … When r = q the breakeven is exactly eps = 1, whatever c is (see METHODOLOGY.md §4). The marker shows your current eps.
Monte Carlo: how uncertain is this?
Loading…
What we know vs. what we assume
Sources
| Description | Value | Publisher | Kind | Status | Accessed |
|---|
Share this scenario
The link encodes your current slider positions in the URL query string; opening it restores this exact scenario.