File 03 — The Jevons Machine
The Jevons Machine
The price collapses. The bill grows. Jevons called it in 1865.
Token prices have been falling tenfold a year — and every AI budget on Earth went up anyway. Set how fast the price collapses and how your appetite answers, and watch two straight lines fight over your invoice on log paper. The whole paradox turns out to be one tick on a dial: below ×10, efficiency saves you money; above it, efficiency is what's raising the bill. Coal did this after Watt. Light did it for seven centuries. The dials are calibrated to what tokens just did — prices ÷10 a year, usage ×50. Every figure sourced.
Every year the price of a token collapses, and every year the bills grow. Set how fast the price falls and how your appetite answers — and watch which of the two straight lines wins the fight over your invoice. The same experiment has run before, on coal and on light.
History’s dial position: about ÷10 a year at constant capability for three years running; the fastest tiers fell ÷50.
×10 is the Jevons line.Below it, efficiency saves you money. Above it, efficiency is what’s raising the bill. The whole paradox is this one tick.
Whose appetite?
Your own dial position — drag until it feels honest.
Log paper: every gridline is another factor of ten. The falling line is the price of a token. The rising line is how many you run. The red line is their product — the only number your bank account ever sees.
A token in year 4
÷10k
of today's price
The bill everyone predicts
$0.002/mo
today's usage at tomorrow's prices
The bill you actually get
$796/mo
running ×398k as much
This is the paradox, mid-flight. Every single thing you run got cheaper — by year 4 a token costs ÷10kof today’s price, and no individual ask ever feels expensive again. But at your appetite, usage grows ×25.1 for every ÷10 of price, so the bill compounds ×2.51 a year anyway. Nobody decided to spend more. The price cut decided for them.
The tell: a bill is a multiplication, price × appetite — and the forecast everyone makes only watches the price. Whether cheaper means less was never decided by the price line. It is decided by the appetite dial, and for coal, light, compute and now tokens, that dial has historically sat above the line.
Deterministic arithmetic, not a simulation of chance: usage is modeled as a constant elasticity to price, which is the textbook form of the rebound effect. The calibration — prices ÷10 a year at constant capability, the market’s usage ×50 in a year — is sourced in the essay. Illustrative, not a forecast of any real invoice.
The essay
Cheaper Never Meant Lessread →