Skip to main content

COMPEL Glossary / token-economics-vdt

Token economics (VDT)

A decomposition of generative-AI inference cost across input tokens, output tokens, context overhead, retrieval tokens, and tool-call tokens — each priced and tracked separately.

What this means in practice

Enables targeted optimisation (prompt caching for context, reranking for retrieval, structured output for generation).

Synonyms

token-cost decomposition , AI inference token economics

See also

  • Unit economics — Cost and revenue per atomic unit — per transaction, per successful decision, per hour saved — for an AI feature.
  • Prompt caching — An inference optimisation that caches the attention key-value state for a prompt prefix so that subsequent requests sharing the same prefix skip re-processing.
  • Compute budget (VDT) — A pre-agreed ceiling on training or inference compute per AI feature per period — expressed in tokens, FLOPs, or dollars.
  • FinOps for AI — The FinOps Foundation's Inform-Optimize-Operate lifecycle applied to AI workloads — visibility into AI spend, optimisation of token and compute costs, and operational discipline.