Acquire AI training data, sourced to your spec

What can be sourced, and what it sells for

Public prices by kind of data, and which way demand is moving. Each kind has its own page with sources and rights.

Kinds of data that can be sourced, with public prices and demand
Kind of data	Public price	The rights question
Company records	$174K to $693K per company	Owning the files is not the right to license them. Customer contracts, software terms, employee notices and third-party material inside the archive each need checking before a buyer sees anything.
Source code	$934 to $24.2K per repository	Code written for a client usually belongs to the client. Open-source components keep their own licences, and a contractor's work is yours only if the contract says so.
Video and footage	$15 to $99 per hour of video	You must own the footage, and the people in it must have agreed to this use. Licensed music and third-party content in the frame stay someone else's.
Audio and speech	$65 to $266 per hour of audio	Every voice is biometric data in some jurisdictions. Each speaker must have agreed to this use; a recording made for quality and training does not cover selling it.
Images	$0.27 to $5.27 per image	You must own the copyright, and identifiable people and private property need releases for this use.
Text and archives	$881 to $3,082 per book or title	A publisher often holds print rights without holding AI training rights; authors may need to agree title by title. User-written content belongs to its writers unless the terms say otherwise.
Datasets, tasks and environments	$1,200 to $2,500 per environment	You must own the environment and everything inside it. Tasks derived from someone else's proprietary material carry that material's restrictions.
Physical-world data	$4.48 to $25 per hour of video	Every worker filmed must consent, and so must anyone else in view. Footage of a client's premises needs the client's agreement.
Expert and human work	$55 to $101 per hour of work	Work made to a buyer's brief belongs to the buyer under the contract. Past work written for an employer or a client is theirs, not its author's, and cannot be sold by the person who wrote it.

Company records
Public price
$174K to $693K
per company
The rights question
Owning the files is not the right to license them. Customer contracts, software terms, employee notices and third-party material inside the archive each need checking before a buyer sees anything.
Source code
Public price
$934 to $24.2K
per repository
The rights question
Code written for a client usually belongs to the client. Open-source components keep their own licences, and a contractor's work is yours only if the contract says so.
Video and footage
Public price
$15 to $99
per hour of video
The rights question
You must own the footage, and the people in it must have agreed to this use. Licensed music and third-party content in the frame stay someone else's.
Audio and speech
Public price
$65 to $266
per hour of audio
The rights question
Every voice is biometric data in some jurisdictions. Each speaker must have agreed to this use; a recording made for quality and training does not cover selling it.
Images
Public price
$0.27 to $5.27
per image
The rights question
You must own the copyright, and identifiable people and private property need releases for this use.
Text and archives
Public price
$881 to $3,082
per book or title
The rights question
A publisher often holds print rights without holding AI training rights; authors may need to agree title by title. User-written content belongs to its writers unless the terms say otherwise.
Datasets, tasks and environments
Public price
$1,200 to $2,500
per environment
The rights question
You must own the environment and everything inside it. Tasks derived from someone else's proprietary material carry that material's restrictions.
Physical-world data
Public price
$4.48 to $25
per hour of video
The rights question
Every worker filmed must consent, and so must anyone else in view. Footage of a client's premises needs the client's agreement.
Expert and human work
Public price
$55 to $101
per hour of work
The rights question
Work made to a buyer's brief belongs to the buyer under the contract. Past work written for an employer or a client is theirs, not its author's, and cannot be sold by the person who wrote it.

Where it comes from

Company records under mandate

Operating companies licensing the record of how their work gets done, rights file first.

Closed companies and estates

Archives of companies that stopped trading, sold by the people who control them.

Made to your spec by a partner

Recordings, demonstrations and expert tasks made by people through AITraining.jobs, which you contract and pay directly.

Made-to-order work goes through AITraining.jobs, a job board for AI training work with the same owner as this site. It publishes what each vendor advertises for the same expertise, the cost side of any commissioned request.

What you receive first: a datasheet

Disclosure rules for training data now ask developers where it came from, whose it was and whether it holds personal information. The datasheet of every source we bring you answers those questions in writing, in the same order, as the holder's own statement.

We do not buy a holder's data, take ownership of it, or license or sell it in our own name. The holder, or a provider it chooses, delivers the data to the buyer, and the buyer pays the holder. The licence is between you and the holder.

Source: Who made the records, in what business, over which years.
Contents: The kinds of records, the systems they came from and how they link to each other.
Measured size: Counts measured by the holder's own valuation after duplicates, generated files and third-party material are removed, not raw volume.
Rights: Who owns each part, which contracts were checked, what was excluded and why.
Personal data: What was present, what was removed, by whom and to which standard.
Format and delivery: Export formats, schemas, whether delivery is by connector or by file.
Licence offered: Field of use, term, exclusivity, what happens to the copy when the term ends.
Known limits: Gaps, languages, quality problems and anything a research team would otherwise find later.

Tell us what you need. We source it to your spec.