A meter ticks one notch at a time. AI makes each word fresh — never copied. That one notch is a Token.
🎲 "Artificial intelligence changes the world" = how many Tokens?
ANo fixed answer
BOne per letter
COne per word
DAlways 16
No fixed answer! Different AIs cut it into 3, 7, or a dozen pieces — and charge by the piece.
A Token is made fresh, one notch at a time — with no standard ruler.
Chapter 1 · Did you know?
Same sentence, different "meters"
Not a product, not a file, not electricity — a brand-new category.
Chapter 2 · Value — Not all Tokens are equal
Two fates of one bucket of water
Splashed on the road, or saving a tree in the desert — same water, value worlds apart. Same for Tokens.
🎲 Same Tokens: review a huge contract vs chit-chat — value gap?
AAbout 2×
BAbout 10×
CAbout 100×
D1000× or more
1000× or more! Price records what it cost to make — never what it earned you.
A Token's value comes from what you do with it, not what it is.
Chapter 2 · Did you know?
Same Tokens in — wildly different value out
Cost is linear (by Token); value is non-linear (by outcome).
Chapter 3 · Cost — What does one Token cost?
Not a photo album — a restaurant
Every word means real fire, ingredients, a chef's time — a power plant, chips, cooling. Heavy industry, not a few lines of code.
🎲 Building a $2B AI factory to make Tokens. What's the most expensive thing?
ABuilding & land
BElectricity
CChips
DSoftware
Chips!
HBM (high-bandwidth memory) alone = 40–55% of the cost.
Only 3 companies in the world make them — shortage may last till 2027. Chips are the engine of every Token.
Bonus: software iterates by the day, but the physical world is slow — transformers take 2.5 years, grid queues 5–12 years. The book calls this "two clocks".
AI looks like software, but it's heavy industry.
Chapter 3 · Did you know?
Cheaper per word — yet the bills explode
The faster software runs, the more the slow physical world becomes the limit.
Chapter 4 · Demand — Why Tokens are never enough
From "giving directions" to "driving itself"
Give it a destination and it drives there itself — detours, U-turns, replanning. That's an Agent. The fuel gauge drops fast.
🎲 Ask AI to "throw a birthday party" — how many steps?
AJust 1
B3–5
CDozens, on its own
DIt can't
Dozens — by itself! A simple Q&A = a few hundred Tokens; an Agent task = tens of thousands.
AI learned to drive itself — from "answering" to "getting things done".
Chapter 4 · Did you know?
When AI drives itself, usage explodes
It's not that people ask more — machines now work non-stop.
Chapter 5 · Competition — efficiency reshapes the industry
Not the biggest tank — the best mileage
The race is shifting from "whose model is smartest" to "who makes the same result with fewer Tokens".
🎲 In the cake, which layer makes the most money?
AApps (closest)
BThe model
CThe chips
DThe power plant
The chipmaker! NVIDIA keeps ~$75 of every $100 — more than all model companies combined.
The winner isn't the biggest tank — it's the best mileage.
Chapter 5 · Did you know?
The race: raw power vs efficiency
Likely outcome: two tracks coexist — top power AND low cost & speed.
Chapter 6 · Business — from selling Tokens to selling intelligence
Nobody yells "give me 3 kWh!"
We buy electricity for cool summers, bright nights — not the kWh. Companies want the job done, not "a million Tokens".
🎲 You're the AI boss — the most advanced way to charge?
APer use
BMonthly pass
CHire a "digital employee"
DFree
From "how much you used" to "did it get the job done" — selling cool air, not electricity.
Smart businesses sell the "cool air", not the "electricity".
Chapter 6 · Did you know?
Six ways to sell intelligence
Manage it well, or the AI bill becomes a cost black hole.
Chapter 7 · Rules — what new rules does the Token economy need?
Cars just hit the road — no traffic lights yet
Lights, licenses, insurance all came later. Tokens are at that stage — customs can't tell if it's a good or a service.
🎲 A US kid uses a Chinese AI for an essay — how to tax it?
AAs goods
BAs software
CAs a phone call
DNobody knows
Nobody has figured it out! Part good, part software, part service — old rules can't hold it.
AI is a new species — old rules can't hold it. You'll write the new ones.
Chapter 7 · Did you know?
Four things packed into one cross-border call
First it crosses borders — then it must enter statistics, tax, and law.
Wrap-up
The future belongs to those who use AI well 🌱
It sees differently
but you can learn to talk to it
Asking > answering
good questions come from you
Humans set direction
what we build is up to you
"The end of the token economy is still — people."
🙋 What's the one thing you'd most want AI to do for you?
← → navigate · click an option to reveal · F fullscreen
