Sell-side broker for AI training data

We help you sell your data to AI labs.

We find the buyers your data fits, put it before all of them at once and negotiate the terms. No sale, no fee.

  • No data upload
  • About two minutes
  • No email needed
  • Free

What is your data worth?

Kind of data
People at its largest
Years of records
Base case

Comes with your exact valuation

Pessimistic

$67,000

Optimistic

$580,000

Pessimistic $67,000, optimistic $580,000. The base case is marked on the line; its figure comes with your exact valuation.How it is worked out

On two answers. An indication, not an offer.

1 of the 83 figures behind it is a sale, carrying 0% of the weight; the rest are the price calculators and offers of buyers and marketplaces. Each is scaled to a company like yours; point at one to read it.
  • Replay
  • Ooak
  • Grepped
  • Handshake AI
  • Nyne
  • micro1
  • Polyshares
  • Mode (BizBuyData)
$10K$100K$1M$10M

Your range, pessimistic to optimisticBaseSoldOffered or estimatedThicker counts more

Your exact figure is waiting. Answer a few more questions and your full valuation opens with it, no email needed.

Get my exact valuation
  • No data upload
  • About two minutes
  • No email needed

What buyers pay for each kind of data

Every public price, per unit, one row per kind. Each row opens its kind. Updated 31 Oct 2026.

One public priceMedianMiddle half697 of the 1,434 public prices: those stated in each kind's main unit. How it is computed

Why sell through us

With AIData.Broker, you are more likely to sell, for more, and faster than you would alone.

More likely to sell

Every buyer wants different data and sets its own rules. We know them, and find the ones yours fits.

  1. Source code17 buyers
  2. Images11 buyers

How many of the 44 buyers on our board take each kind of data.

For more

We know what buyers pay for each kind of data, and every buyer that fits values yours, so they compete.

  1. Opening bid (Google)$5M
  2. Backup bid (Mercor)$7.5M
  3. Winning bid (Google)$10M
  4. Late offer (micro1)$12.5M

Spirit Airlines' records at auction: the last bid was 2.5x the first. Docket

Faster

Every buyer that fits hears of your data at the same time, without your name, not one after another.

Alone: one buyer after another

Through us: every buyer at once

Every buyer answers to one deadline, the dashed line, so none waits on another.

More likely to sell

Finding a buyer at all is the hard part. Every buyer wants different data (code, tickets, email, audio, a certain size, years or industry) and sets its own rules on where you are based, what rights you hold and how personal data is removed. Writing to buyers one by one mostly reaches the wrong ones, and a buyer drops data that cannot answer its questions on rights.

We follow the buyers of this market and the requirements each one publishes, so your data goes only to the buyers it fits. We settle the rights questions with you before any buyer learns who you are or sees your datasheet, and your records are de-identified on your own computer to a published 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.

We know the market: what buyers actually pay and offer for each kind of data, at what size and on what terms, from every public price and published clause we can source, so we know what your data is worth before a buyer names a number. Every buyer that fits gets the same datasheet and one deadline, you see every offer with its terms and what you would net, and we decline the fee some buyers pay whoever introduces a company, asking them to put it into your price.

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.

Because we already know who buys what and how each one buys, nothing waits on finding out. 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. We answer you within one business day, and in a sale we set one deadline for every buyer, normally 7 days to say it is interested and 14 days to make its offer.

You lose nothing. No sale, no fee. You decide every offer, and a valuation commits you to nothing. Our fee

How a sale works

How a sale through us works, from a free valuation to payment.

  1. You, free

    Value it

    A range from a few answers, with no data upload and no account.

  2. You, then us

    Hand it to us

    Accept the mandate in one step from your valuation. We confirm who you are and that you may sell it.

  3. Us, with you

    Get it ready

    We settle the rights and write the datasheet a buyer's lawyers review. You approve it.

  4. Buyers, all at once

    Buyers compete

    They hear at once, without your name, on one deadline: 7 days to say they are interested, 14 to offer.

  5. You decide

    Choose and get paid

    You pick the offer and its terms. The buyer pays you, and our fee comes only from that payment.

  • No data upload
  • About two minutes
  • No email needed

1,434

public prices, each sourced

44

buyers and what they publish

130

disclosed licensing deals

$12.5M

highest public offer for one archive

Sales on record

What company data has sold for

Recent sales of companies' own records and code that state a figure. Every figure opens its source.

All 1,434 prices
This week in the market, 26 Oct 2026

micro1 raised its published prices for company records, and Handshake's calculator now shows less

The two buyers that publish the most about company records moved their prices in opposite directions this week. A public company filed what one patient record sells for, and Getty Images' bankruptcy papers showed how small and uneven AI licensing still is for the largest image library.

Optional: so we write about what matters to you

Issues go out on Mondays: what moved in AI data prices, deals and rulings. Free, and every issue has a link to leave. Subscribing means you accept our terms and privacy policy.

  • What buyers publish now

    micro1 rewrote its data partnerships page: its tiers now read $100K to $500K, $500K to $2M and $2M+ (on 9 October they read $100k+, $500k+ and $1M+). It says it pays in full within 30 days of acceptance, typically $150K to over $2M, and it now licenses the trajectories of a company's own AI agents for $1M to $2M a year.

    micro1, data partnerships page

  • What it sold for

    Tempus AI's 8-K prices a licence of de-identified neurology patient records to Pathos AI at $100 per delivered record: 500,000 records over five years, up to $55 million, exclusive only for a window after each delivery. With Bristol Myers Squibb's oncology licence ($137.5 million fixed to 2031) the two total $192.5 million.

    Tempus AI, SEC Form 8-K, 23 October

  • Prices

    Ten published offers per code repository now give a median of about $2,060 (middle half $1,350 to $5,870).

    AIData.Broker market rates, from the buyers' own pages

If you hold data

Find out what your data is worth.

Company records, code, footage, speech, archives and datasets, for holders based in the United States, the United Kingdom and Canada. You pay us only out of a sale.

Value your data
  • No data upload
  • About two minutes
  • No email needed
  • Free

How a sale works, from the rights file to payment

If you need data

Buying 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.

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

Questions

Questions people ask first

License it to a buyer that pays for company records (micro1, Handshake AI, Mercor, Turing and Scale AI publish programmes), either directly or through an agent such as AIData.Broker, which puts several buyers in competition. First settle what you may license (customer contracts, the terms of the tools you export from, what you told employees), remove personal data, and describe the data without naming the company; then compare offers on price and on terms. You keep ownership and license the data yourself; the few public sales so far have a median of $72.5K per company.

Yes. AIData.Broker is a sell-side broker for company data and acts for the company selling alone in a sale, for holders based in the United States, the United Kingdom and Canada: it values the data against sourced public prices, settles the rights first, and describes the data to every buyer that fits at once, without naming the company until one is interested and has signed an undertaking. No buyer pays it in a sale it runs. The company keeps ownership and licenses the data itself, and pays a success fee only if a sale closes: 12% of the first $250,000, less above it. The fee and every term of its agreements are published, clause by clause, before anyone signs.

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.

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.

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 do not store: we delete it once it is measured and our judgment of the lot has read it), 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.

The holder pays us a fee fixed in the mandate, and only if a sale closes. Besides that, a buyer pays us only through its own published referral programme when we introduce you to it, out of its own money: you see the most it may pay us before we introduce you, and which buyer it is once we have. The fee is 12% of the first $250,000 you are paid, less above that. Several buyers pay whoever introduces a company to them. Where a buyer that fits your data runs such a programme, we introduce you through it: it pays us its published payout on top of our fee, out of its own money. You see the most any of them may pay us before we introduce you, and which buyer each is once we have, and you can ask us to stop at any time. Any other payment a buyer offers for introducing you we decline, tell you about, and ask for in your price instead.

As listed on