Token Prices Fell 98%. Bills Tripled.
Cheaper tokens should mean cheaper AI. They didn’t.
Goldman Sachs’s numbers: token prices down 98%, enterprise bills up 3x.
Both are true, and the gap between them is the whole story.
Cheaper didn’t mean less
When something gets cheap, people use far more of it. Economists call this the Jevons paradox.
Agents are the clearest example. A chatbot answers once. An agent loops: plan, call a tool, read the result, try again. Every step re-sends the context.
The price per token fell. The tokens per task exploded.
Goldman’s outlook is 24x more tokens by 2030.
Where the extra tokens go
Not into new thinking. Into repetition:
- Re-reading the same codebase every session.
- Re-sending the same instructions every call.
- Re-learning decisions the team made last week.
At a 98% discount, that waste feels free. At 24x volume, it isn’t.
You can’t discount your way out
Waiting for the next price cut won’t fix this. Price falls 98%, usage grows 24x, and the bill still climbs.
The only lever left is buying fewer tokens for the same work.
The fix
Stop paying for the same context twice:
- Index once.
- Share across every agent.
- Refresh only what changed.
- Keep the decisions, not just the code.
Cheaper tokens are a discount on waste.
Cut the waste.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article