AI Training Data Price Index: Disclosed Deals (2026)

The index: disclosed deals with a price

Compiled by Origin Data Partners research. Newest first. Prices are as reported; "up to" and "more than" figures are ceilings or floors, not exact amounts. The CSV holds the same rows.

Spirit Airlines: the sale is contested and has not been approved. A docket tracker lists the 17 November 2026 hearing on the contested proposal (ElevenFlo docket tracker); unions withdrew their objections on 10 November 2026 (Law360). micro1 proposed $12.5M after the auction; that proposal was not the selected bid (Fortune).

Company claims (not part of the index)

These figures come from the seller, the buyer or a data vendor describing its own deals, offers or spending. They are useful as signals, but nobody independent has confirmed them, so they are kept out of the index and labelled.

Methodology: what counts and how each row is graded

A row enters the index only when a price for a specific deal appears in a regulatory filing (SEC, ASX), a court record, or reporting by an independent news outlet. A figure published by one of the parties, a vendor or a sponsored article goes in the company-claims table instead. Programme ranges, labour rates, asking prices and company revenue are not deals and are left out.

Evidence grades: Filing; Court record; Independent press; Company claim. Where a filing or a news report was read through another outlet, the grade says so ("read through independent press", "WSJ, relayed"). Each row links to the page we read.

Every source was re-opened on 13 November 2026 and a verbatim sentence containing the price was recorded; a row whose source could not be opened, or no longer showed the figure, was dropped. The editorial and sourcing policy explains how Origin checks and corrects figures.

What is not known

Most deals disclose a total or a yearly fee, not what was paid per record, per hour or per gigabyte, so rows cannot be compared on one unit. Exclusivity and term are disclosed for very few. "Up to" figures may never be reached.

Whether the Spirit Airlines sale will be approved, and on what terms, will not be known until the court rules after the 17 November 2026 hearing.

For several record types that AI developers ask for, no closed price has been published at all: insurance claims files, semiconductor design databases, cybersecurity incident data, logistics records, loan files, law-firm data and aviation maintenance records sold on their own. For these, only buyers' published ranges and expert labour rates exist.

Private deals are the large majority. This index shows the public part of the market only, which leans toward publishers, platforms and listed companies that must disclose.

Rows considered and left out

Checked on 13 November 2026 and left out of the index for the reason given.

Use and cite this data

Free to reuse under CC BY 4.0 with a link to this page. Download the CSV. For deals without a public price and the wider history since 2023, see the AI data licensing deals tracker; for what buyers publish they will pay companies, see the directory of AI data companies; for where models still fail by sector, see where AI still fails at professional work.

Agent traces from production: no disclosed price yet

No closed price for production agent traces has been disclosed, so none is in the index. The published figures are company offers and ranges, listed with their grades in the guide to licensing agent traces from production.

Valuable data types with no disclosed price yet

Several record types that buyers ask for have no disclosed deal price, so none is in the index. Their guides list the published offers and ranges, graded:

• Security incident and SOC case files: cybersecurity incident data for MSSPs and IR firms
• Evaluation sets, corrections logs and refresh licences: evaluations, tasks and corrections a company can license beyond its archive
• Insurance claims files: insurance claims file holders
• Chip design and verification data: chip design firm data
• Loan files with performance: non-bank lender loan files
• Pathology slides and outcome-linked imaging: patient records and imaging data for practices

Changelog

13 November 2026: first edition. 17 disclosed deals and 12 company claims, built from Origin's deals tracker (updated 3 November 2026) and November 2026 price research; every source re-opened; 10 groups of figures left out (see above). Next check after the Spirit Airlines hearing on 17 November 2026.

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.