The Token Paradox
Published
AI got 98% cheaper. Enterprise AI bills went up. Token volumes are exploding, agents multiply consumption, and most organisations still cannot say who owns the AI bill. What technology leaders must decide.
Blended research: grounded in surveys, powered by AI (read our methodology)Key statistics
3.2 quadrillion
tokens a month processed across Google services in May 2026, 7× a year earlier
Evidence: CitedGoogle I/O 2026
15×
more tokens used by multi-agent AI systems than by a chat interaction
Evidence: CitedAnthropic engineering
98%
of FinOps teams now manage AI spend, up from 31% in 2024
Evidence: CitedFinOps Foundation, State of FinOps 2026
Download this Provocation free
Summary
The price of AI has collapsed: Google says the cost of a token has fallen 98% since 2024. Yet enterprises are spending more on AI, not less, and many cannot say where the money goes.
This second issue of 22decisions Provocations examines the token paradox: exploding volumes, agents that consume many times more tokens than chat, and an enterprise response that is building brakes before deciding who holds the steering wheel.
Our provocation
Spend caps stop runaway agents, but they cannot tell whether an agent was worth running. The organisations that win the token era will not be those that spend least on AI, but those that know what every agent is worth.
What's inside
- The token explosion
- What is driving the tokens
- The agent multiplier
- The paradox: cheaper tokens, bigger bills
- How enterprises are responding
- The 22decisions view
- What technology leaders should decide
8 pages, PDF. Free to share with attribution.
Written byVarun Shourie, Founder, 22decisions
Varun Shourie founded 22decisions to enable technology leaders to prepare for, and get ahead in, the AI era.
About 22decisionsDownload this Provocation free
- Topics:
- AI costs, AI agents, FinOps, AI leadership
- Sectors:
- Cross-industry