AI Ledger Hepburn Advisory Hepburn Advisory

Usage Ledger

Token volume, pricing, model mix, and demand quality across the AI inference economy.

Revenue reconciliation: This page shows combined provider ARR (trailing-quarter run-rate × 4, includes inference from ad-platform workloads). The home page shows $22B customer revenue (CY23–25 cumulative, collected). Audited fiscal-year collected revenue surfaces on /timeline when available. See Methodology for the reconciliation. ARR and collected revenue measure different things and are never interchangeable.

Token Demand Quality Matrix — who buys × how they buy. Actual 2025 collected revenue. Not all "$1 of AI revenue" is equal.

Token Volume & ARR by Provider — click bars to filter

Demand Composition

Model Pricing & Estimated Volume

SaaS AI Transition Map — Twelve trad-SaaS AI products scored on five posture axes

Last editorial review: Cadence: per earnings cycle ARR cells refresh automatically between reviews

A larger filled area = a more aggressive transition posture (which can be the cost of doing the work, not a verdict). Two axes read counter-intuitively: GM rewards gross-margin compression attributable to AI (proof of real workload) — stable margins on a heavily-marketed AI product score low. PRICE reserves a 5 only for vendors where new commercial models (per-action, per-agent, outcome-based) are the only way they sell — none of this cohort qualifies. Private vendors render four-axis radars (no public GM disclosure).

Scoring rubric (5 axes)
Product / Parent
Posture
Reported AI ARR → Estimated Real → Share
AI ARR
Source

Watch notes & sources

Tier 1b — directly disclosed in earnings
Tier 2a — narrative + reconciled
Tier 2b — editorial estimate
Tier 3a — modelled / press

Top AI Natives

What would have to be true

Data points that would confirm or challenge the usage growth trajectory.
Token volume growth compounding ≥ 80% YoY
~120% YoY
Target: ≥80%
Ahead
Paid vs. free token share ≥ 35%
~28%
Target: ≥35%
Lagging
Price-per-token decline slowing (cost floor approaching)
~10x/yr decline
Threshold: ≤3x/yr
Lagging
Sources: OpenRouter API, provider earnings calls, AI Ledger multi-model consensus. See methodology →