“which vendors and exchanges of ai training data are growing or shrinking the fastest over the last 3 years”
A best-evidence market screen of 45 candidate rows covering AI training-data vendors and exchanges over the three-year window ending 2026-09-24 — not exhaustive market coverage, since most leading firms are private and disclose inconsistent metrics. Figures mix audited revenue, ARR/run-rate, marketplace adoption, web traffic and closure events; the visual keeps those bases apart and never puts unlike metrics on one axis. The headline chart shows the 14 defensible entities; all 45 screened rows are in the table and appendix.
Bar length = change multiple on a log scale, computed only where a numeric start and end share the same metric basis. Green = growing, red = shrinking; a grey status block marks a closure or retirement, where no numeric rate exists. Metric bases differ between panels (revenue vs run-rate vs adoption vs traffic) — compare within a panel, not across. Hover or tap a bar for the reported values, period and source.
Click a header to sort. Confidence is a screen of source quality: disclosed filings and first-party posts rate higher than secondary research templates. Rows outside the headline chart include misclassified entities and weakly supported claims — e.g. the “Invisible Technologies shrinking” row covers only a generic transcription sub-segment while the company overall grew, and Scale AI's alleged $870M→$850M dip conflicts with growth estimates from stronger sources.
| Type | Entity | Direction | Metric basis | Start → end | Period | Multiple | Confidence | Source |
|---|
Method: screen of 45 candidate rows on AI training-data vendors and exchanges gathered from independent web sources for the three-year window ending 2026-09-24; retrieved Sep 2026. Values are as reported — a mix of audited/reported annual revenue, ARR/run-rate, marketplace adoption counts, web-traffic estimates and closure announcements; multiples are computed only where start and end share a metric basis. Scale AI's $250M start (2022) predates the strict window and is shown as three-fiscal-year context. Only the 14 defensible entities appear in the headline chart; the remaining rows, including misclassified or weakly sourced claims, appear in the table only. Stated-change text truncated for space.