AI Ledger
Hepburn Advisory
Hepburn Advisory
AI Ledger
Hepburn Advisory
Hepburn Advisory
The AI application layer, company by company — who is at scale, on what evidence, at what price structure, in which regions.
What AI companies actually earn, who funded them, what they charge and where they operate. Every number is dated, graded for how good the evidence is, and linked to where it came from. The typical figure here is — days old, and we say so rather than round it off.
Swipe sideways for the full table. The company column stays put.
Click any row for the full detail: dates, sources, funding history, pricing and regional evidence. Default order ranks disclosed and reported figures above estimates regardless of size. Estimates render as ranges and never enter the totals above.
Projections are directional arithmetic, not forecasts. Benchmarks are dated and sourced. Every derived figure on this page (current revenue read, growth, capital raised, implied multiple, repricing exposure) is computed at build time from dated inputs; nothing is stored as a standalone "current" number.
| Badge | Meaning | Renders as |
|---|---|---|
| Disclosed | Filing, company statement, or named-executive quote | Point value + as-of date |
| Reported | Named outlet or named analyst citing a figure | Point value + as-of date + outlet |
| Estimate | Documented method over cited inputs | Range only — the midpoint never renders, and estimates never enter totals |
| Not published | No credible observation | Stated as such — never a placeholder value |
The current read is the latest non-superseded observation in a company's chain. Where two observations fall within ninety days, the higher evidence tier wins; at equal tier, corroborated beats indicative; still tied, more recent wins. Two credible figures for the same period (within thirty days of each other) disagreeing by more than twenty per cent render side by side and the row enters the review queue. A chain sourced entirely from one outlet carries a single-source marker. Growth renders as the percentage change between two credible observations at least ninety days apart, and only then. It is the change observed across that window, never annualised from a single point and never inferred from funding activity, which prices expectations rather than revenue. The window, the two figures and their sources are in the row expand.
A large share of rows currently carry an observation inherited from this site's earlier audit layer: dated and tier-labelled, but pending re-verification against a primary source. Those rows are marked "source pending" in the row expand and counted openly under Open questions below; the monthly review loop works through them in revenue order.
Run-rate figures annualise a current period; booked revenue is what customers paid across twelve months. They are not interchangeable, and every observation here is a run-rate unless its source states otherwise. Where basis (gross versus net of reseller margin) is disclosed, it is noted on the observation.
Capital raised sums primary rounds and extensions only. Secondary sales, tender offers and debt are recorded in the timeline (they date valuations) but excluded from the total. An implied revenue multiple renders only when a closed round's post-money valuation and a credible revenue observation sit within sixty days of each other; announced-but-unclosed rounds never ground a multiple.
| Level | Meaning | Minimum evidence |
|---|---|---|
| 0 | No presence | — |
| 1 | Sells into region | Regional pricing, case study or partner announcement |
| 2 | Established | Local legal entity (registry), office, or sustained local hiring |
| 3 | Operational | Data-residency option, regional infrastructure or regional support, from vendor documents |
Levels are granted only against stored evidence items, each with a URL and retrieval date, visible in the row expand. Hiring alone never grants level 3. Cells not yet assessed show a dash rather than a zero.
A computed classification of whether today's price is structurally likely to hold, from five stored inputs: inference intensity of the category, dependence on frontier model APIs, how directly pricing meters cost, months since the last funding round, and any price rise, restructure or credit-value reduction in the trailing eighteen months. The rules, published verbatim: elevated = high inference intensity AND flat-seat pricing AND (more than eighteen months since funding OR a recent price event); watch = any two of {high intensity, flat-seat pricing, funding gap, recent price event, frontier-API dependence}; otherwise lower. The chip is a computed fact, not a verdict, and ships as beta; inputs are listed and dated in each row's expand. Context: at the frontier layer, customer revenue covers roughly a third to a half of modelled full cost, which is the structural reason application-layer pricing can move under buyers. Precedents: the Cursor Pro restructure (17/06/2025) and the GPT-5 API price rise (23/04/2026).
Each row's dot reflects the age of its current observation: green within thirty days, amber beyond that, red past twelve months (which also queues the row for review), grey where nothing is published. This page deliberately shows its staleness rather than hiding it.
The Usage Ledger's consumer band and this index read the same underlying company records. The headline total here is narrower by construction: it counts only disclosed and reported figures, only the AI application layer, and excludes gateway pass-through revenue and the digital-native comparison class. The Usage Ledger's "known ARR" tally sums every tracked figure regardless of rung, so it will always read equal or higher.
Data as of — · rebuilt from the company record store on every site build · method detail on the Methodology page.