The smallest billable unit of machine cognition, and the economy being built on it
This collection is about the economics of AI tokens — the sub-word units that language models read, generate, and bill by — not the tokens of crypto and blockchain. Every prompt sent to an AI model runs a meter: text is broken into tokens, tokens consume compute, compute is billed, and the output either creates value or wastes capacity. Across eight modules and 41 sessions, it follows that meter from what a token literally is, through the cost structure of inference and the gap between token cost and token value, to workflow design, the AI industry stack, labor and productivity measurement, and the physical and sovereign limits on compute supply. Because prices and market figures change quickly, the focus stays on durable mechanisms — pricing structures, cost drivers, and value capture — rather than on today's price list. You finish able to reason about AI as an economic system rather than a magic interface, and to aim tokens at work worth paying for.
What a token literally is, and why every AI bill starts there.
Where the bill really comes from — and why the same price list produces very different invoices.
Token cost is mechanical; token value depends on the question, the workflow, and the operator.
Turning token spend into a managed budget with measurable outcomes.
Who collects rent on every token — and who gets squeezed.
What metered cognition does to jobs, output, and the statistics meant to track them.
Every token has a material footprint — and a flag.
Access, direction, capture — the operator's playbook for turning tokens into kept value.
“AI economics taught from its smallest unit — a sober, practical account of what every prompt costs and earns.”
Most AI writing covers capability; this collection is about the meter. It traces the token from tokenizer to power grid, separates durable pricing structures from fleeting price lists, and treats phantom GDP and labor repricing as open debates. Practical tools — budgets, model routing, an efficiency review, an ROI canvas — keep it grounded.
A token is not a word; it is the smallest billable unit of machine cognition, and every prompt runs a meter.
Language models break text into sub-word pieces and process them as sequences of integers, so a common word may be one token while an unusual name splits into several. Providers price usage per token, typically with separate rates for input and output. Once cognition is counted this way, it can be budgeted, audited, and optimized like any other metered input.
Econ Beyond publishes collections on the forces that shape our economic lives, from prices and work to money, cities and nations. Big ideas and real evidence, told plainly, so readers can see the economics in their own decisions at work, at home and at the ballot box.
Ranked by semantic similarity
What an agent costs to run, how it might earn, and who could capture the value
The portable toolkit for seeing what a choice really costs and what happens next
How AI, trade fragmentation, aging, and debt are rewiring the world economy
