The holder's broker for AI training data

We find a buyer for your company's data.

AI labs pay for the records a company makes as it works: tickets, email, chat, code, documents. We value yours. We sell your data for you: faster, more likely to sell, and for more than you would get alone. No sale, no fee.

Value your dataI need data
  • No data upload
  • No account
  • Free
  • About two minutes

What is your data worth?

Kind of data
People at its largest
Years of records

$74,000to$590,000

On two answers. An indication, not an offer.

Base case $220,000. Against the market records that carry nine tenths of its weight, by who stated them, each moved to a company like yours. Point at one to read it.
  • Replay
  • Ooak
  • Grepped
  • micro1
  • Nyne
  • Mode (BizBuyData)
  • Handshake AI
  • Troveo
$10K$100K$1M$10M

Your range, pessimistic to optimisticBaseSoldOffered or estimatedThicker counts more

Full valuation
  • No data upload
  • No account
  • Free
The market on one scale

What your data is worth rests on what buyers have paid: every public price per unit, one row per kind of data, the most valuable first. Each row opens its kind. Updated 17 Oct 2026.

  1. Company recordsper company

    $316K median of 97

  2. Datasets, tasks and environmentsper environment

    $2,000 median of 33

  3. Source codeper repository

    $1,579 median of 9

  4. Text and archivesper book or title

    $515 median of 18

  5. Audio and speechper hour of audio

    $118 median of 132

  6. Expert and human workper hour of work

    $77 median of 114

  7. Video and footageper hour of video

    $37 median of 45

  8. Physical-world dataper hour of video

    $16 median of 56

  9. Imagesper image

    $0.53 median of 46

One public priceMedianMiddle half550 of the 1,022 public prices: those stated in each kind's main unit. How it is computed

Why sell through us

Faster, more likely to sell, and for more than you would get alone.

Data that took years to make can be worth a great deal to a lab. Getting it bought means reaching the right buyers, proving what it is, protecting the people in it and signing terms you can live with. We do that work for you.

Faster

Alone, a company writes to one buyer, waits for an answer and then tries the next. An unknown company's email rarely gets a reply, and buyers rarely review data without a datasheet.

We already know who buys what and each buyer's published rules. We describe your data to every buyer whose rules it meets at the same time, without naming you, and each one that is interested gets your datasheet after its written undertaking.

Who buys, and what each asks for

More likely to sell

Most sales of company data stop on rights and personal data: who owns each part, which contracts limit it, and whether the people in it are protected. A buyer drops data that cannot answer.

We settle the rights questions with you before any buyer sees anything, your records are de-identified on your own computer to a published standard, and we approach only buyers whose rules your data meets.

Our de-identification standard

For more

One buyer means one offer, and the first number is often the lowest. Alone you have nothing to compare it with, and terms such as exclusivity, onward sharing and when you are paid change what an offer is really worth.

Competition sets the price: the same datasheet goes to every interested buyer on one deadline, you see every offer with what you would net, and we decline the fee some buyers pay whoever introduces a company, asking them to put it into your price.

In the one public auction of a company's records, Spirit Airlines', the opening bid was 50% of the price competition reached.

Direct, or through us

You lose nothing

Only you pay us, and only out of a sale: no sale, no fee. In a sale we run no buyer pays us, so we earn more only when you are paid more. You decide every offer and can leave any buyer out, the buyers you already talk to stay yours, and a valuation commits you to nothing.

1,022

public prices, each sourced

37

buyers and what they publish

111

disclosed licensing deals

$12.5M

highest public offer for one archive

One sale where every bid is public

The same archive drew four figures from three buyers.

Spirit Airlines' corporate archive, auctioned in its Chapter 11 case in August 2026. The court has not yet approved the sale, and the late offer came after the auction closed. Most sales are private and show the holder one price.

Docket summary

2.5x

the last figure against the first

$7.5M

between the first figure and the last

  1. 01Opening bid (Google)

    $5M

  2. 02Backup bid (Mercor)

    $7.5M

  3. 03Winning bid (Google)

    $10M

  4. 04Late offer (micro1)

    $12.5M
If you hold data

Find out what it is worth, then put it before every buyer that fits, at once.

Company records, code, footage, speech, archives, datasets. You describe it and answer the rights file; we put it before every buyer that fits, without your name until one is interested, and are paid only if a sale closes. We sell for holders based in the United States, the United Kingdom and Canada.

Value your dataCheck an offer you have
  • No data upload
  • No account
  • Free

How a sale works, from the rights file to payment

If you need data

Give us the spec and we source it.

From holders under mandate, from the archives of closed companies, or made to order through AITraining.jobs, a job board for AI training work with the same owner as this site.

Post a request

Know someone sitting on data? Introduce them and earn one third of our fee on their sales that close within 12 months.

Sales on record

What data has sold for

The largest sales on record, from court records, company filings and named press. There is no price list in this market; these are the figures it has. Pick a kind of data to see its own.

Kind of data

73 sales on record across all kinds of data

  • Bartz v. Anthropic class settlement: the $1.5 billion fund divided over the 482,460 works on the final Works List (books downloaded from pirate libraries)

    Gross settlement value per listed work, from the final approval order; the court itself rounds it to 'approximately $3,000'. It pays for past acquisition of pirated copies (the court found training itself fair use), so it prices liability, not a training licence going forward.

    Text and archivesSales that happenedJul 2026Court record

    Bartz v. Anthropic, Order Granting Final Approval of Class Action Settlement (N.D. Cal., Dkt. 680, 20 July 2026)

    $1.5B

    per settlement covering 482,460 works

    $3,109 per book or title

  • News Corp's current and archived journalism licensed to OpenAI for five years (2024 deal)

    The largest known publisher deal, reported and not confirmed by the parties; it covers training and display in ChatGPT. The per-year figure is the total divided evenly by the term.

    Text and archivesSales that happenedMay 2024Press report

    LLM Pulse, AI content licensing deals 2023 to 2026

    $250M

    over five years

    $50M per contract year

  • Reddit's data licences signed in January 2024 (access to search and analyse Reddit data), aggregate of several contracts

    The seller's own filing, the firmest figure behind the reported Google and OpenAI deals. It sums several contracts; Reddit expected at least $66.4 million of it as revenue in 2024.

    Text and archivesSales that happenedJan 2024The payer's own page

    Reddit, Form S-1 registration statement (SEC)

    $203M

    aggregate contract value of several licences, two to three years each

  • Two multiyear licences of Getty Images content (a major social platform and a large AI company)

    Disclosed by management; buyers unnamed, and the social-platform deal includes display rights, so neither is a pure training-data price.

    ImagesSales that happenedMar 2026The payer's own page

    Getty Images Q4 2025 earnings call (transcript, The Motley Fool)

    About $65M

    for two deals combined

  • Troveo's cumulative payouts to video content owners for AI-training licences, against the hours of video it has licensed (creator, studio and archive footage)

    What reached rights holders on average per licensed hour across Troveo's whole library, the realised floor of the creator-footage market; up from $5 an hour in its first year ($5 million for 1 million hours, already in the store). Payouts are the owners' share, after Troveo's cut.

    Video and footageSales that happenedOct 2026The payer's own page

    Troveo, For Content Owners page

    $50M

    cumulative payouts across all licensed hours

    $7.14 per hour of video

  • Shutterstock image-library licences to large tech buyers

    Initial contract sizes; Reuters reported most were later expanded.

    ImagesSales that happenedApr 2024Press report

    The Decoder (summarising Reuters)

    $25M to $50M

    per buyer deal

  • News Corp US and UK news content, including archives, licensed to Meta for AI

    Terms from people familiar; News Corp confirmed the deal but not the money.

    Text and archivesSales that happenedMar 2026Press report

    Mediaweek (citing The Wall Street Journal)

    Up to $50M

    per year

  • Wiley scholarly and learning content licensed for AI (all licensees, fiscal year to 30 April 2026)

    Aggregate revenue, not one deal; $19.4M of it came from other publishers' content Wiley resold for a royalty.

    Text and archivesSales that happenedApr 2026The payer's own page

    John Wiley & Sons Form 10-K (SEC)

    $49.1M

    per year (all AI licensees)

All 1022 prices
The 2026 price guide: what each kind of data sells for, and why

Kinds of data

The market, by kind of data

The public price of each kind (the middle half of its prices, in its main unit), which way demand is moving, and who buys it.

Every rate and how it is computed

Records

The rest of the data

Download all of it as CSV or JSON

Tools

Tools built on it

All tools

Questions

Questions people ask first

Where do these figures come from?
Every price is something somebody else published: a court record, a buyer's own page or named press. Each row links to where it was read and carries the date of the sale or the publication. None is our estimate.
How is the valuation worked out?
It is arithmetic on those public prices: a unit price range for the kind of data, times the size you enter, adjusted for what the evidence says moves a price. It shows its sources under the result. It is an indication, not an offer.
Do you hold or see the data?
Of a holder's data, we hold only what it chooses to share with us to value and sell it: its answers and counts, a de-identified sample (whose text we delete once it is measured), the rights file, the datasheet and the facts of the documents it adds for verification. We use these for that holder, under the agreements it accepts, and its counts also, without its name and combined with other holders' figures, in the market statistics every valuation is priced from. We also keep what public records and our own checks say about a holder, as the Privacy Policy describes. 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.
Who pays you?
Only the holder pays us, a fee fixed in the mandate, and only if a sale closes. In a sale we run no buyer pays us, so our fee depends on what you are paid, not on which buyer pays it. A buyer pays us only if you choose to apply to it through its own referral programme, and you see that amount before you choose. The fee is 15% of the first $250,000 you are paid, less above that. Several buyers pay whoever introduces a holder. In a sale we run we decline that money, tell you beside each offer what that buyer pays introducers where it publishes it or tells us, and ask it to put that amount into your price instead. If you choose instead to apply to one buyer through its own referral programme, that buyer pays us its published payout as well as our fee, and you see the amount in writing before you choose.
AIData.Broker: find a buyer for your company's data among AI labs