Updated 17 Oct 2026
How prices for AI training data are actually negotiated.
There is no price list, so every price is negotiated. These are the levers that moved real prices, each measured from a court record, a filing or a buyer's own terms, and what to do with each, whether you are selling data or buying it.
For a holder
Run a competitive process, not a bilateral talk
Strong evidence
- What the record shows
- In the Spirit Airlines records auction three qualified bidders took Google's $5.0M opening bid to a $10.0M winning bid in about two and a half hours, and a fourth buyer (micro1) later filed $12.5M, a 25 percent premium over the winner.
- What to do with it
- Never answer the first unsolicited offer with a price: set a bid deadline, qualify at least three funded buyers, give them one asset schedule so bids compare, and keep the right to accept a topping bid until signature.
3 sources, with the words read
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“The auction commenced by remote teleconference, was moderated by Davis Polk and PJT Partners, and lasted approximately two and a half hours ... Google's opening bid was $5 million in cash”
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“Google opening bid $5.0M Mercor.io first overbid (third-party deidentification) $5.2M Mercor.io Alternate Bid $7.5M Google Successful Bid $10.0M micro1 Inc. noticed bid (September 3) $12.5M”
- Fortune: micro1 tries to beat Google's bid to buy bankrupt Spirit Airlines data for $12.5 million(2026-09-14)
“Micro1, which was founded in 2022 and is based in Palo Alto, has proposed paying $12.5 million for the trove of records, beating Google’s previous top bid of $10 million.”
Processing costs belong to the buyer
Strong evidence
- What the record shows
- Google and micro1 each bore all de-identification costs in the Spirit sale with no price reduction, and Mercor's company-data program states it covers extraction, anonymization and transfer with no fees or deductions.
- What to do with it
- Strike any clause that nets extraction, cleaning or de-identification costs out of the price; the funded buyers already absorb them, so a deduction is a price cut in disguise.
2 sources, with the words read
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“It states that it will bear all deidentification costs under section 1(a)(iii) of the Sale Agreement with no price reduction.”
- Mercor: Data monetization(2026-10-11)
“Mercor covers all costs associated with data extraction, anonymization, and transfer. You receive compensation with no upfront costs, setup fees, or deductions from the payment.”
Get one written offer, then shop it
Strong evidence
- What the record shows
- Nyne commits to beat a qualifying written offer for the same records, rights, payment conditions and work, for US operating companies with at least 20 full-time W-2 employees and three years in business; calculator estimates and 'up to' amounts do not qualify and no fixed uplift is promised.
- What to do with it
- Convert interest into one dated, scoped, written offer as early as possible; it is the only document a competing buyer will match or beat, while estimates carry no leverage.
2 sources, with the words read
- Nyne: offer qualification terms(2026-10-11)
“Nyne will beat a qualifying written offer for your company’s operational data. The comparison covers the same eligible dataset, licensing rights, payment conditions and required work.”
- Nyne: offer qualification terms(2026-10-11)
“Calculator estimates, promotional “up to” amounts, expired or unverifiable offers, speculative future revenue and separate compensation for new employee work do not qualify as the competing price.”
Non-exclusive data can be sold again
Indicative
- What the record shows
- Troveo advertises an average of 4 sales per dataset (labelled illustrative), SuperAnnotate pays on delivery and again on each relicense, and Kled signed a $12M non-exclusive licence over two years.
- What to do with it
- Before granting exclusivity, compare one exclusive price against the number of buyers who would take a non-exclusive copy; keep the right to relicense and to block named competitors instead.
3 sources, with the words read
- Troveo: business data(2026-10-11)
“$20M+ Paid to Licensors 40+ Active Buyers 4× AVG Sales Per Dataset ILLUSTRATIVE EXAMPLE”
- SuperAnnotate: Data Partnerships(2026-10-11)
“We prepare the data and license it to buyers you approve. Payment on delivery, and on each relicense.”
- Solana Compass: Kled AI signs $12M data deal(2026-06-26)
“Kled AI signed a $12M non-exclusive data licensing deal on June 26, making the marketplace officially profitable.”
Know the market split before an intermediary takes a share
Strong evidence
- What the record shows
- Created by Humans pays authors 80 percent; the Authors Guild calls 75 to 85 percent to the author fair and objects to academic presses offering 25 percent; HarperCollins' Microsoft deal paid $5,000 per title split evenly; Perplexity sends 80 percent of Comet Plus revenue to publishers.
- What to do with it
- A pure agent or marketplace should leave you 70 to 80 percent or more; an intermediary that also holds rights may justify 50/50; refuse splits near 25 percent for data you own outright.
4 sources, with the words read
- Jane Friedman: Created by Humans officially launches(2025-01-15)
“It pays 80 percent of proceeds to the author”
- Authors Guild: AI licensing for authors, who owns the rights and what is a fair split(2024-12-12)
“a fair split for most deals will give 75–85 percent to the author, depending on the publisher’s role. Note that some academic presses are presently offering authors a mere 25 percent”
- eWeek: HarperCollins books to train Microsoft AI models(2024-12-07)
“the terms of the deal offered $5,000 per nonfiction book, split evenly between the author and HarperCollins.”
- Shelly Palmer: Perplexity will pay publishers(2025-08-26)
“says 80% of Comet Plus revenue will go to participating publishers”
Pick the marketplace fee tier by expected volume
Strong evidence
- What the record shows
- Datarade charges sellers $0 a year plus 30 percent commission, $6,000 a year plus 20 percent, or $12,000 a year plus 15 percent; FileYield takes 8 percent from sellers and 6 percent from buyers.
- What to do with it
- At $60,000 of annual sales Datarade's $6,000 tier costs the same as commission-only, and above it the fixed fee is cheaper; use the 8 to 30 percent band as the yardstick for any broker's fee, including ours.
3 sources, with the words read
- Datarade: provider pricing(2026-10-11)
“$0 / year + 30% commission ... $6,000 / year + 20% commission”
- Datarade: provider pricing(2026-10-11)
“$12,000 / year + 15% commission”
- FileYield: sell data(2026-10-11)
“Current agreement rates are 8% for sellers and 6% for buyers.”
Claim the referral budget when you come direct
Strong evidence
- What the record shows
- Buyers budget 5 to 10 percent of a deal for whoever introduces a seller: micro1's default referral payout is 7.5 percent of the licence price, Mercor pays 5 percent up to $100,000, Grepped 10 percent of the owner's payout up to $100K, Scale up to $50,000 and micro1 $50,000 per company.
- What to do with it
- A seller who arrives without an introducer saves the buyer this budget: ask for it in the price, and check that any introducer's fee is disclosed and not taken from your side.
4 sources, with the words read
- micro1: referral program terms(2026-10-08)
“the Referral Payout will equal seven and one-half percent (7.5%) of the total license or purchase price paid to Company in connection with such Qualified Referral.”
- Grepped: referrals portal(2026-10-11)
“Refer data. Earn 10% of the deal. Company records, datasets and RL environments. Up to $100K per referral, paid when the owner is paid.”
- Scale AI: data partnership(2026-10-11)
“Refer a company and earn up to $50,000 bonus.”
- micro1: refer a company(2026-10-11)
“$50,000 referral reward Company compensation: $100K-$2M+ for approved data packages”
Resellers keep about 40 percent: that is your room
Strong evidence
- What the record shows
- Innodata, a data vendor to large technology companies and AI labs, spent 60 percent of 2025 revenue on direct operating costs (61 percent in 2024), so it kept about 40 percent as gross margin; its Digital Data Solutions segment spent about 61 percent.
- What to do with it
- A vendor that buys your data to resell can pay up to about 60 percent of its resale price and still earn its usual margin; selling or licensing directly to the lab captures that spread, so ask an intermediary what the end buyer pays.
2 sources, with the words read
- Innodata Inc., Form 10-K for 2025 (SEC)(2026-02-26)
“Direct operating costs as a percentage of total revenues were 60% and 61% for the years ended December 31, 2025 and 2024, respectively.”
- Innodata Inc., Form 10-K for 2025 (SEC)(2026-02-26)
“Direct operating costs for the DDS segment as a percentage of DDS segment revenues were approximately 61% and 63% for the years ended December 31, 2025 and 2024, respectively.”
Sell access to refreshing data, not a one-off file
Strong evidence
- What the record shows
- Reddit's January 2024 licences carried $203.0M of aggregate contract value over two to three years with at least $66.4M in the first year; Meta pays News Corp up to $50M a year for at least three years; Microsoft paid Informa an initial access fee of about $10M plus recurring payments over three years.
- What to do with it
- If the data keeps growing, price an upfront access fee plus annual payments over two to three years, and tie renewal to fresh deliveries.
3 sources, with the words read
- Reddit, Inc., prospectus (SEC Form 424B4)(2024-03-21)
“In January 2024, we entered into certain data licensing arrangements with an aggregate contract value of $203.0 million and terms ranging from two to three years. We expect a minimum of $66.4 million of revenue to be recognized during the year ending December 31, 2024”
- Editor and Publisher: News Corp, Meta in AI content licensing deal worth up to $50 million a year(2026-03-04)
“will pay The Wall Street Journal owner up to $50 million a year. The deal will run for at least three years”
- interactive investor: Informa does AI data deal with Microsoft(2024-05-08)
“Microsoft will pay Informa both an initial data access fee of around $10 million along with recurring payments over the next three years.”
Ask for a floor plus a usage-linked variable
Strong evidence
- What the record shows
- OpenAI pays Dotdash Meredith a minimum of about $16M a year, the fixed component only, with a variable component on top; Adweek reports most of OpenAI's publisher deals use fixed and variable components.
- What to do with it
- Never accept pure revenue share or usage pricing without a guaranteed minimum; ask for the fixed floor first, then negotiate the variable rate.
2 sources, with the words read
- Adweek: OpenAI is paying Dotdash Meredith at least $16 million to license its content(2024-11-18)
“OpenAI is paying the digital media company Dotdash Meredith a minimum of around $16 million per year to license its content”
- Adweek: OpenAI is paying Dotdash Meredith at least $16 million to license its content(2024-11-18)
“Most of the deals that OpenAI has struck with publishers are structured around two components, according to a leaked pitch deck: fixed and variable”
Payment terms: days to weeks, with a signature tranche
Strong evidence
- What the record shows
- Grepped pays within 7 days of data passing review; Mercor wires on delivery, typically 2 to 4 weeks after signing; Ooak pays in 2 to 4 weeks; Corpus Bridge's example is half on signature and half on delivery.
- What to do with it
- Ask for a share on signature and the rest on delivery, with a short written acceptance window, and refuse open-ended acceptance tests or payment only after the buyer's own resale.
3 sources, with the words read
- Grepped for Business(2026-10-11)
“Get paid Within 7 days of your data passing review.”
- Mercor: Data monetization(2026-10-11)
“Payment is issued via wire transfer upon delivery of the anonymized dataset to the buyer. This typically occurs within 2-4 weeks of the signed agreement”
- Corpus Bridge(2026-10-11)
“The buyer pays on the milestones in the agreement, for example half on signature and half on delivery.”
Bring every connected system, especially documents and email
Strong evidence
- What the record shows
- Ooak's estimator values a 51 to 100 person operating company with 4+ years at $85k to $121k with six tools (Google Drive, Gmail, Slack, Jira, GitHub, HubSpot), but $16k to $22k with only Jira, GitHub and HubSpot; Handshake says the highest payouts include data from 8 or more named applications.
- What to do with it
- Inventory every system the company ever ran, retired ones included, and scope the package so documents, email or chat and the work trackers link to each other; dropping the document store or communications cuts value about five-fold.
4 sources, with the words read
- Ooak Data: partner page estimator(2026-10-11)
“Estimated value of your data $85k to $121k This is a preliminary range.”
- Ooak Data: partner page estimator(2026-10-11)
“Estimated value of your data $16k to $22k This is a preliminary range.”
- Handshake AI: data acquisition form (served page code)(2026-10-11)
“The highest payouts typically include data from 8 or more named applications”
- Mercor: Data monetization(2026-10-11)
“Compensation is based on data volume, the tools you connect, and the depth of your data. Connecting more tools generally results in a higher payout.”
An operating company's records are worth more than a closed one's
Strong evidence
- What the record shows
- Ooak's estimator gives the same 11 to 50 person company with the same six tools $51k to $71k while operating and $33k to $47k once closed, about 1.5 times more; SimpleClosure notes operating companies license and keep using the data while winding-down companies sell outright.
- What to do with it
- License while the company runs (and keep the data), and if a shutdown is coming start the sale before SaaS subscriptions lapse, while the company can still answer diligence questions.
3 sources, with the words read
- Ooak Data: partner page estimator(2026-10-11)
“Estimated value of your data $51k to $71k This is a preliminary range.”
- Ooak Data: partner page estimator(2026-10-11)
“Estimated value of your data $33k to $47k This is a preliminary range.”
- SimpleClosure: What your assets are worth(2026-09-03)
“If you’re still operating: licensing can create a new revenue stream while allowing you to continue using the underlying assets. If you’re winding down: an outright sale can help capture additional value”
Lead with what is rare, not with how much there is
Indicative
- What the record shows
- SimpleClosure reports that less common languages such as Ruby or C++ fetch top dollar, especially with linked Jira, Linear or Figma context; AIxBlock puts low-resource languages and regulated domains at the top of its $5 to $25 per hour range for call-centre audio.
- What to do with it
- Describe the archive by its rare parts (uncommon languages, regulated domains, linked tickets and outcomes) and price those separately rather than averaging them into the bulk.
2 sources, with the words read
- Upstarts Media: SimpleClosure data on what AI labs buy(2026-09-03)
“programming that’s less common – say, Ruby or C++ – can fetch top dollar from the AI labs and data labeling shops. Even better if you’ve got context from Jira, Linear or Figma”
- AIxBlock: off-the-shelf call center audio vs custom(2026-10-11)
“English call-center audio sits at the bottom of the range. Low-resource languages and regulated domains (medical, banking) sit toward the top.”
Sell company records into rising demand
Strong evidence
- What the record shows
- micro1 committed to spend $1B over 12 months acquiring and licensing enterprise operational data (9 Oct 2026) and told the Spirit court it had closed more than fifty data transactions in 45 days; SimpleClosure counts 9 times more AI buyers than in 2025.
- What to do with it
- Run processes now while several funded buyers compete, keep deadlines short, and do not lock in long exclusivity at today's prices.
3 sources, with the words read
- micro1: commitment to spend $1B on enterprise data acquisition(2026-10-09)
“Today micro1 is announcing a commitment to spend $1 billion over the next 12 months acquiring and licensing enterprise operational data through our Company Data Partnerships program.”
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“"has closed more than fifty data transactions in the last forty-five days,"”
- SimpleClosure press release (BusinessWire, via Webull)(2026-09-03)
“9X buyer growth: The number of AI buyers has grown 9X compared with 2025”
Not needing to sell is leverage
Indicative
- What the record shows
- After Warmly agreed to be acquired by HubSpot it received four offers of up to $300,000 for its codebase and communication records, and its CEO turned them all down.
- What to do with it
- Unsolicited offers arrive quickly after a public event; answer with a process and a reservation price, and be ready to decline rather than accept the first number.
1 source, with the words read
- The Information, via FNNews: AI firms buying old company data(2026-08-14)
“it received multiple offers to buy its codebase and employee communication records for as much as $300,000. ... said he received four such offers but turned them all down.”
Benchmark success fees against court-approved scales
Indicative
- What the record shows
- A Chapter 7 trustee asked the court to approve an auctioneer at no more than 20 percent of the first $20,000, 10 percent of the next $50,000 and 8 percent of the balance, expenses included.
- What to do with it
- In an estate or wind-down sale, a sliding scale that falls with size is what a court accepts; propose fees in that shape rather than a flat high percentage.
1 source, with the words read
- US Bankruptcy Court, E.D.N.C., Case 25-05055, Doc 36(2026-03-18)
“PERSONAL PROPERTY 20% of first $20,000.00 10% of next $50,000.00 8% of balance”
Sell the clean records first and the personal data separately
Indicative
- What the record shows
- In the Spirit sale the competitive bidders agreed an asset schedule that excluded personal information, and the personal data of 97.5 million passengers was set aside for a later, separate auction.
- What to do with it
- Split an archive into operational records that can be de-identified and personal or customer data that needs consent or a separate process; the clean set clears faster and draws more bidders.
2 sources, with the words read
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“by the first round of the auction the most competitive bidders had agreed to bid on an asset schedule that expressly excluded personally identifiable information.”
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“a subsequent auction to sell the personal data of the 97.5 million passengers will be conducted at a later date”
Read the grant: a perpetual, sublicensable licence is a resale right
Strong evidence
- What the record shows
- micro1's published data licence master terms take a worldwide, fully paid-up, irrevocable, perpetual licence with the right to sublicense through multiple tiers, for the one fee the holder agreed. The buyer then has 30 business days after delivery to reject the data, and silence counts as acceptance.
- What to do with it
- Treat a sublicensable grant as permission to resell to every lab: strike it, or price it as a per-sublicensee fee or a revenue share, and ask for a fixed term instead of perpetual. Tie payment to delivery, not to the end of a six-week acceptance window.
2 sources, with the words read
- micro1, Data License Master Terms(2026-10-07)
“Licensor grants Licensee and its affiliates a worldwide, royalty-free, fully paid-up, nonexclusive, irrevocable, perpetual license, with the right to sublicense through multiple tiers”
- micro1, Data License Master Terms(2026-10-07)
“Licensee has thirty (30) business days after delivery of the Data to reject it by written notice identifying the affected Data and describing the deficiency in reasonable detail; if Licensee does not give timely notice, the Data is deemed accepted.”
Export a codebase with its tickets and links, before the archive expires
Indicative
- What the record shows
- Buyers ask for ticket history and pull-request context with a codebase, yet the standard tools lose it: a GitLab project export leaves out the links between issues and to related merge requests, and a GitHub organisation migration archive is deleted after seven days.
- What to do with it
- Before pricing, export repositories together with the issue tracker and the links between tickets and changes, keep the archive, and show the buyer that the links survive. Code without its tickets is the cheapest kind; a wind-down that loses the links loses value it cannot recover.
3 sources, with the words read
- DataVendor docs, What you can sell(2026-10-14)
“issue or ticket history (Linear, Jira, and so on) and pull-request context”
- GitLab docs, Migrate projects by using file exports(2026-10-14)
“Links between issues and linked items Links to related merge requests”
- GitHub REST API docs, Organization migrations(2026-10-14)
“Migration archives are automatically deleted after seven days.”
Compare offers net of the platform's cut and payout rules
Strong evidence
- What the record shows
- Two code marketplaces quote on opposite bases. HUD's DataVendor takes 20% of earnings at withdrawal, holds earnings 14 days after payment and pays out from $1,000. Pangea's partner quotes the owner's side with nothing deducted and pays any introducer 15% on top.
- What to do with it
- Before comparing two figures, ask whether each is gross or net, when money settles and what the minimum payout is. A $20,000 ceiling on a 20% platform is $16,000 to the holder.
3 sources, with the words read
- DataVendor docs, Fees and limits(2026-10-14)
“Platform fee | 20% of earnings | Only at withdrawal, and never on referral fees.”
- DataVendor docs, Fees and limits(2026-10-14)
“Withdrawals | $1,000 minimum before fees; one open withdrawal per team at a time”
- Pangea.ai, Code Licensing referral partner guide(2026-10-14)
“nothing is deducted from what the owner takes home. Your 15% is paid on top.”
Introducer fees run from 5 to 20 percent: ask for that share when you come direct
Indicative
- What the record shows
- Company-data buyers publish what they pay whoever introduces a seller: Red Deer Data 5% of the deal, Pangea's code programme 15% of what owners are paid, and Telegraph Lab 20% of what the business receives on licences of $100,000 to $4 million.
- What to do with it
- A seller who arrives without an introducer saves the buyer that budget. Ask for part of it in the price, and treat a buyer who pays introducers 20% as having at least that much room.
3 sources, with the words read
- Telegraph Lab, Referral Partner Program(2026-10-14)
“We pay businesses $100,000–$4 million to license data for AI training, and you earn 20% of what they receive.”
- Pangea.ai, Code Licensing referral partner guide(2026-10-14)
“15% of whatever the people you refer are paid”
- Red Deer Data, Referral partners(2026-10-14)
“Earn 5% of the deal when a company you refer licenses its business data through Red Deer Data.”
Drop consumer transaction systems to clear privacy review; the price held
Indicative
- What the record shows
- After the auction, Spirit and Google agreed that passenger transaction (PNR), refund, inflight purchase and WiFi sales systems will not be transferred in any form, de-identified or not, leaving corporate email in scope under Tonic.ai's de-identification; the ombudsman then backed the sale, which is still described as a $10 million sale. The excluded list is final for those systems, though the revised schedule is a draft.
- What to do with it
- Offer the internal-operations records (email, chat, documents, wikis, workflow systems) and leave structured customer databases out of a training-data sale: they carry most of the privacy risk and, on this evidence, little of the price. Sell a customer list, if at all, separately to its natural buyers.
2 sources, with the words read
- In re Spirit Aviation Holdings, Dkt. 1684 (Consumer Privacy Ombudsman supplemental report)(2026-10-05)
“The following datasets will not be part of the sale transaction and will not be provided to Google in any form (deidentified or otherwise): ... B. Consumer Transaction Systems – i.e., Transactions, Passenger Name Record (PNR) C. Refund systems D. Inflight/Onboard Purchases E. WiFi sales F. Others TBD”
- Chapter11Cases: Spirit Airlines case update, the data sale slips to November(2026-10-14)
“The Consumer Privacy Ombudsman now supports the $10 million sale of deidentified data to Google, but the hearing has moved to November 6.”
Find and exclude vendors' confidential data before the sale
Strong evidence
- What the record shows
- Five third parties have objected to the Spirit records sale over their own data inside it: Springshot says its data in email, Teams and SharePoint is not estate property (supplemental objection 9 October 2026, with sample payloads to help exclude it), and International Aero Engines, Navitaire and Amadeus each cross-moved to enforce confidentiality agreements.
- What to do with it
- Inventory partner and vendor material held under NDA or licence (engine and maintenance data, booking-system data, SaaS exports) and either exclude it, get consent, or price an indemnity; buyers will ask for a warranty that the records contain no third-party confidential information.
2 sources, with the words read
- In re Spirit Aviation Holdings, Dkt. 1703 (Amended agenda, responses received)(2026-10-12)
“F. Limited Objection of International Aero Engines LLC and IAE International Aero Engines AG to the Debtors’ Proposed Sale of the Deidentified Data and Cross- Motion for Enforcement of Confidentiality Agreements [ECF No. 1538]”
- Chapter11Cases: Spirit Airlines case update, the data sale slips to November(2026-10-14)
“Springshot, Inc. filed a supplemental limited objection on October 9 (Dkt. 1693). It says it has no wish to derail the sale, but contends the debtors negotiated the transaction without accounting for third-party vendor data that is not estate property”
Write the standard carve-outs into the records sale agreement
Strong evidence
- What the record shows
- The Spirit sale agreement excludes attorney-client privileged material (an inadvertent transfer waives nothing and the buyer must return or destroy it), lets the seller keep copies for wind-down and legal compliance, bars the seller from selling the same records to anyone else, and still lets it sell its customer data list (traveller spend aggregated by year) to hospitality or travel buyers. The buyer may pass the data on only to recipients bound by the same no-re-identification commitment.
- What to do with it
- Ask for the same four clauses: privilege excluded with a clawback, a retention right for legal and wind-down needs, exclusivity limited to the defined records (not every dataset the company holds), and onward transfer only on flow-down terms. Exclusivity is something the buyer is paying for: price it.
2 sources, with the words read
- In re Spirit Aviation Holdings, Dkt. 1463, Exhibit A (Sale Agreement, Google LLC)(2026-08-14)
“The Assets shall not include any communications, documents, or other information protected by any attorney-client privilege, work product or similar protection”
- In re Spirit Aviation Holdings, Dkt. 1463, Exhibit A (Sale Agreement, Google LLC)(2026-08-14)
“the Seller shall not sell, assign or transfer the Assets or any interest in the Assets to any purchaser other than Buyer or to any other third- party ... provided that the Seller may sell its customer data list, which includes individual traveler spend aggregated by year, to third-parties in hospitality or travel industries.”
A training-data licence must be disclosable when the company itself is sold
Indicative
- What the record shows
- In Ooma's purchase of FluentStream ($45,000,000 enterprise value, agreement dated 31 October 2025) the seller represented that, except as scheduled, the company had not let any company or customer data, de-identified data included, be used as training data by another party.
- What to do with it
- Living companies: license records on terms you can schedule at your own exit (defined dataset, defined use, non-exclusive or time-limited, return or deletion on request, no transfer of customer data without consent), keep the agreement on file, and expect an acquirer to ask about it and price it.
2 sources, with the words read
- Ooma, Inc. Form 10-Q exhibit 10.1: Stock Purchase Agreement between Ooma, Inc. and FluentStream Holdings, LP(2025-10-31)
“(D) permit any data associated with the Company or any of its Subsidiaries or its or their customers (including deidentified data) to be used as Training Data by any other Person.”
- Ooma, Inc. Form 10-Q exhibit 10.1: Stock Purchase Agreement between Ooma, Inc. and FluentStream Holdings, LP(2025-10-31)
““ Enterprise Value ” means $45,000,000.”
Check the buyer's completed deals and funding before sharing records
Indicative
- What the record shows
- SimpleClosure, whose AssetHub has handled assets from more than 350 companies, warned founders on 1 October 2026 that new buyers advertise $1 million to $5 million offers with no completed purchase behind them, and that a buyer who takes raw files and promises to scrub them later already holds what was meant to be removed.
- What to do with it
- Before any data room: ask for a completed, paid purchase from another company, who signs, who pays and where the money comes from, and whether the counterparty is the end user or a broker; keep scrubbing on your side (or with an independent agent) before files move, and require the agreement to state the price, the conditions for payment and the payment date.
3 sources, with the words read
- SimpleClosure: How real are those million-dollar data offers? (Dori Yona)(2026-10-01)
“now there are ads from companies like SHFA.ai promising founders $1 million, $3 million, even $5 million. Those numbers can make a new company look like an established buyer.”
- SimpleClosure: How real are those million-dollar data offers? (Dori Yona)(2026-10-01)
“If the buyer receives the raw files first and promises to clean them up later, you have already given that buyer access to the material you meant to remove.”
- SimpleClosure: How real are those million-dollar data offers? (Dori Yona)(2026-10-01)
“The agreement should state the actual price, what must happen before payment is due, and when it will arrive.”
A private agent task is priced from the human time it takes to solve
Strong evidence
- What the record shows
- Evaluators that buy private tasks directly size the payment by the professional time the task takes: METR paid a base of 3x the human cost to complete it plus up to 50% bonus (2023), then $300 per professional-hour (2024); the UK AI Safety Institute indicated GBP 100 to 200 per development hour, GBP 3,000 to 15,000 per task (2024). Both pay in full only once the task passes their QA, and METR requires the task to stay off the public web.
- What to do with it
- When pricing an environment or task for a buyer, state how long a professional needs to solve it and anchor at a multiple of that cost; offer confidentiality (never published, never trained on) as part of the price.
3 sources, with the words read
- METR, Bounty: Diverse hard tasks for LLM agents(2023-12-16)
“we'll pay a base price of 3x the estimated $ cost of getting a human to complete the task”
- METR Task Development Guide, Introduction(2024-11)
“we'll award a cash bounty of $300 per hour it would take a professional to complete the task”
- UK AI Security Institute, evals bounty(2024-11-05)
“we expect to reward a successful task with £100-200 per development hour”
Compute credits are offered at twice the cash value
Indicative
- What the record shows
- Prime Intellect's environment bounty sheet paid each bounty twice over when taken as GPU credits instead of cash, so a lab with spare compute values its credits at about half their face value to the payee.
- What to do with it
- When a buyer with its own compute offers credits instead of cash, ask for at least twice the cash figure in credits, and take credits only if you will actually spend them.
1 source, with the words read
- Prime Intellect Environments Program bounty sheet, legacy tab(2025-10)
“Bounty (x2 if GPU credits)”
A documented source is what holds the price of robot data
Indicative
- What the record shows
- In China, raw monocular robot video with a known source trades at 15 to 20 yuan an hour while data of unknown origin, resold again and again, goes for 3 to 5 yuan (Caijing, October 2026); the same market quotes organised real-robot teleoperation at 500 to 1,000 yuan an hour (Xinhua Finance, September 2026).
- What to do with it
- Sell physical-world data with its chain of custody (who collected it, where, under what consent, sold to whom before) and refuse to compete with resold data on price; a buyer should demand that record before paying a premium.
2 sources, with the words read
- ifeng Tech, citing Caijing: 机器人训练数据被曝反复转卖(2026-10-10)
“普通单目原始数据实际成交价已降至每小时15元至20元,来源不明的数据甚至低至每小时3元至5元。”
- Xinhua Finance (cnfin.com), 从遥操作到无本体 具身智能数据迎有效性大考(2026-09-24)
“目前,真机遥操作数据的报价约为每小时500-1000元”
Off-the-shelf mocap is cheap because it carries no training rights
Strong evidence
- What the record shows
- MoCap Online sells cleaned optical motion capture at $0.15 to $1 per animation in packs, but its licence requires prior written permission for any AI use and its FAQ answers 'No' to AI or machine learning; studio capture costs $50 to $200 or more per clip. The gap is the price of training rights plus custom capture.
- What to do with it
- A holder of motion-capture libraries should sell AI-training rights as a separate, priced licence and never quote against game-pack prices; a buyer must not assume a purchased animation pack can be trained on.
2 sources, with the words read
- MoCap Online, License(2026-10-14)
“Can I use MoCap Online assets in AI or machine learning? No.”
- MoCap Online, How Much Does Motion Capture Cost? Complete Pricing Guide 2026(2026-03-17)
“Professional Optical Studio | $1,500–$5,000/day session | $50–$200+ after cleanup”
A holder who sells through a stock or footage platform gives up about half of every AI data deal
Indicative
- What the record shows
- Wirestock moved its dataset-deal commission to 50 percent in May 2023 (against 15 percent on ordinary licences); Newsflare's worked example pays footage owners a 50 percent share of the hourly rate an AI client pays; Shutterstock's Contributor Fund paid a median of 0.69 cents per portfolio image for one half-yearly payout.
- What to do with it
- A holder with a large, rights-clean archive should price a direct licence first and treat a platform route as worth about half the gross; when comparing a platform's per-hour or per-image headline, halve it before comparing with a direct offer.
3 sources, with the words read
- Alltag eines Fotoproduzenten: Wirestock und die fehlenden Datendeals-Auszahlungen(2023-08-17)
“Mit der Einführung des Dataset Deals-Programms haben wir einen Provisionssatz von 50 % für diese speziellen Geschäfte eingeführt.”
- Newsflare: License Your Archive to AI(2026-10-14)
“if a client pays $40/hour and they used 400 hours of your footage, your 50% share would be: 400 × $40 × 0.5 = $8,000”
- Alltag eines Fotoproduzenten: Analyse: Wie viel zahlt Shutterstock für KI-Trainingsdaten an Anbieter?(2023-05-08)
“Der Median lag bei 0,0069 USD ... Bild”
A larger publisher can carry your content into an AI licence, for a royalty
Indicative
- What the record shows
- Wiley's Research Solutions business sells AI licences that include content Wiley licensed from other publishers, and pays those publishers royalties out of it: about 55 percent of that business's quarterly revenue came from contracts that include AI licence revenue, and the year before carried higher royalty costs for content from other publishers.
- What to do with it
- A small press or society too small to reach a lab can sell through a larger publisher's licence, but should get the royalty rate in writing, keep the licence non-exclusive and compare it with selling direct or through an agent before signing.
2 sources, with the words read
- John Wiley & Sons, Form 10-Q for the quarter ended 31 July 2026(2026-09-04)
“Research Solutions products and services generated approximately 55 % of their revenue in the three months ended July 31, 2026 from contracts with customers that include artificial intelligence (AI) license revenue that includes content which Wiley licensed from other publishers”
- John Wiley & Sons, Form 10-Q for the quarter ended 31 July 2026(2026-09-04)
“The prior year included higher royalty costs related to AI license revenue from content licensed from other publishers.”
Removing names does not make records anonymous: set the de-identification standard in the contract
Strong evidence
- What the record shows
- A 2019 study in Nature Communications estimated that 99.98% of Americans would be correctly re-identified in any dataset using 15 demographic attributes, so a scrub that only removes names and direct identifiers leaves people identifiable from what remains.
- What to do with it
- Before any data leaves, write into the contract which de-identification standard applies, who runs it, who pays, and who certifies it independently. Records whose value depends on who the people are cannot be sold lawfully however they are scrubbed.
1 source, with the words read
- Rocher, Hendrickx and de Montjoye, Estimating the success of re-identifications in incomplete datasets using generative models, Nature Communications 10, 3069 (2019)(2019-07-23)
“Using our model, we find that 99.98% of Americans would be correctly re-identified in any dataset using 15 demographic attributes.”
Benchmark a seller's agent's fee against agents in other markets
Indicative
- What the record shows
- Agents who sell a one-off asset for its owner are paid by the owner on success: US literary agents take 15% of an author's gross, and business brokers selling companies worth under $1 million charge 10 to 12% flat with a minimum of $15,000 to $25,000.
- What to do with it
- Expect a seller's agent on a small one-off sale to charge in the 10 to 15 percent range, falling as the deal grows. Ask for a sliding scale, nothing payable if nothing sells, and in writing who else pays the agent.
2 sources, with the words read
- Promise Legal, Literary agent agreements: what authors are actually signing(2026-06-17)
“The number you'll see in nearly every US literary agency agreement is 15%.”
- CT Acquisitions, M&A advisor fee structure(2026-07-04)
“Under $1M Business broker 10-12% flat, minimum $15K to $25K”
Refuse a 48-hour deadline and collect several quotes, because buyers price on different theses
Indicative
- What the record shows
- Business Insider (13 Oct 2026) reports six-figure offers to small businesses that expire in 48 hours, and a broker describes buyers saying 'I'm offering you double market. I need an answer by Friday.' A buyer, Surge AI's Nick Heiner, told MIT Technology Review the same day that buyers quote wildly different numbers because each values an archive on its own research thesis of how many useful tasks it yields; SimpleClosure's Miskell sale shows code and workspace data even close at different times and prices.
- What to do with it
- Sellers: treat a deadline as a reason to slow down; send the same inventory summary to three or more buyers in the same week and ask each to state its basis (tasks per 100,000 lines of code, per team, per year of history). Buyers: a short deadline buys speed but invites a re-trade once the seller learns the spread.
2 sources, with the words read
- Business Insider, AI labs are fueling a wild market for your emails and Slack DMs(2026-10-13)
“"The offers being made look almost identical to those early oil offers," he said from his home base in Dallas. "They'll say, 'I'm offering you double market. I need an answer by Friday.'"”
- MIT Technology Review, AI labs are buying data from dead startups to train their models(2026-10-13)
“Pricing reflects this uncertainty: Buyers quote wildly different numbers, Heiner says, because valuations are based on “research theses” estimating how many useful tasks come from, say, 100,000 lines of code.”
Deliver records with their links intact: a ticket tied to its code commit is worth more than a loose document
Indicative
- What the record shows
- Sunset's CEO told Forbes (April 2026) that it prices a closing company's data on size, age and 'data richness', meaning how traceable the records are inside and how well they connect across platforms; a Jira ticket tied to a specific code commit is worth more than a standalone document, and healthcare and finance carry a premium. Surge AI and micro1 said the same in October 2026 (one company's deep Slack can beat another's whole archive; without Slack, CRM or email the rest is worth less).
- What to do with it
- Sellers: export with the identifiers that join systems (ticket keys in commit messages, thread ids, CRM record ids in email) and keep them through de-identification, then show a buyer three or four traced examples. Buyers: ask whether cross-system links survive the scrub before pricing on volume.
1 source, with the words read
- Forbes JAPAN (licensed translation of Forbes, Anna Tong): AI's new training data, your old work Slacks and emails(2026-04-16)
“「データの豊かさ」とは、データ内部の追跡可能性や、プラットフォーム間の結びつきの度合いを示す指標である。特定のコードコミットに紐づいたJiraチケットは、単独のドキュメントより価値が高いとマホニーは言う。医療や金融など特定の業界ではプレミアムがつくとも付け加えた。”
A holder who keeps ownership and licenses non-exclusively sells the same dataset again and again; a made-to-order sale ends with the first buyer
Indicative
- What the record shows
- Haitian Ruisheng (SpeechOcean) sold usage licences on 395 of its own off-the-shelf datasets in 2020, 576 licences in all, from a catalogue of 762; its made-to-order data passes to the client and cannot be resold. In 2020 those owned datasets earned 45.48 percent of its revenue. A Brazilian data supplier to Defined.ai told Reuters he passes 20 to 30 percent of deal amounts to the owners of the photos, podcasts and medical data he sources.
- What to do with it
- Before granting exclusivity, price it against the repeat licences a non-exclusive catalogue earns; and when a sourcing intermediary offers a share, expect 20 to 30 percent of the deal unless the holder negotiates directly.
3 sources, with the words read
- Haitian Ruisheng IPO prospectus (招股说明书), training data products(2021-08-09)
“销售量(个数) 395 293 183 销售量(授权次数) 576 414 265”
- Haitian Ruisheng IPO prospectus (招股说明书), business model(2021-08-09)
“此类训练数据集一经开 发完成,可多次销售并获取授权许可收入”
- Reuters, Inside Big Tech's underground race to buy AI training data (Katie Paul and Anna Tong)(2024-04-05)
“he pays owners of the photos, podcasts and medical data he sources about 20% to 30% of total deal amounts”
Anchor a per-title AI price on the book's own list price
Indicative
- What the record shows
- In Korea the price of a book copied into a national training corpus was set as a multiple of its e-book list price: the contractor's 2022 settlement paid 70% of list price times 3 copies (about 2.1 times list), and the Seoul appeal court in July 2026 set damages at 70% of list price times 5 copies times each author's share (about 3.5 times list), rejecting fair use.
- What to do with it
- Seller of a book catalogue: when a buyer offers a flat sum per title, test it against 2 to 4 times the title's list price; a flat $100 to $355 is below that for most professional and academic titles. Buyer: a list-price multiple is a defensible formula that courts have already used.
3 sources, with the words read
- Kyunghyang Shinmun, opinion column, 13 September 2022(2022-09-13)
“해당 금액을 전자책 정가의 70%의 3copy로 책정했다고 알린 후”
- Bizhankook, Woongjin loses AI copyright lawsuit(2026-08-05)
“손해액은 전자책 정가의 70%에 인쇄 부수 5부, 저작물별 원고 지분율을 곱하는 방식으로 산정됐다.”
- Bizhankook, Woongjin loses AI copyright lawsuit (the court on the licensing market)(2026-08-05)
“재판부는 AI 개발을 위한 데이터베이스 구축 수요가 늘면서 저작물 이용허락 시장이 이미 형성돼 있다는 점도 짚었다.”
Tagging and metadata, not the footage alone, decide where an hour of video prices in a thousand-fold range
Indicative
- What the record shows
- Veritone's CEO says the broker can charge $1 to $1,000 per hour of video depending on the application and the available tagging and metadata; a MIPCOM 2025 panel cited AI platforms paying $60 to $300 per hour for premium, metadata-rich content; Troveo's own pricing guide says no universal per-hour rate exists and readiness is one of the four things that make a catalogue worth licensing.
- What to do with it
- Before pricing an archive, inventory what descriptive metadata, shot logs, rights records and transcripts exist per hour; quote the tagged share separately from untagged bulk, and treat a buyer's per-hour figure as meaningless until the metadata level it assumes is stated.
3 sources, with the words read
- The Media Leader: TV rights owners promised new revenue stream by licensing content for AI training(2026-01-29)
“Veritone can charge $1-1,000 per hour of video, depending on the application and the available tagging and metadata.”
- Fantrust: Turning archives into assets, AI licensing at MIPCOM 2025(2025-10)
“AI platforms pay $60–$300 per hour for premium, metadata-rich content”
- Troveo: How Much Does AI Training Data Cost in 2026?(2026-08-27)
“Scarcity, exclusivity, clean rights, and readiness are what make your catalog worth licensing, and there is no universal price per message or per hour”
Per-task coding pay looks wildly different ($30 to $2,200 a task) but converges on $45 to $100 an hour once the platform's own time per task is applied
Indicative
- What the record shows
- Vetto pays $30 for a task it says takes about 40 minutes (its own figure: about $45 an hour); Mercor pays $2.2K for a database engineering task of about 25 hours (about $88 an hour); micro1 pays $50 to $100 an hour for coding RL environments of about 15 hours each ($750 to $1,500 a task). Contractor-reported ceilings sit far above (Snorkel for Anthropic, up to $280 for about an hour, reported by Business Insider).
- What to do with it
- Compare a per-task offer by its stated hours, never by the headline: divide the rate by the platform's own time per task, and treat anything under about $45 an hour for professional code work as below the market these postings set.
3 sources, with the words read
- AITraining.jobs (same owner as AIData.Broker), Vetto: Coding, Data and Automations Experts(2026-10-15)
“Payment will be US$ 30 per approved task, converted and paid in your local currency. Each task takes approximately 40 minutes, which corresponds to an effective rate of about US$ 45 per hour.”
- Mercor job board, Database Systems Evaluation Expert(2026-10-06)
“Database Systems Evaluation Expert $2.2K / task ... Selected experts will complete one task, with approximately 25 hours of work”
- micro1 job board, Software Engineer (GitHub Contributor)(2026-10-15)
“Pay: $50–100/hour Time per task: ~15 hours per run”
For a buyer
Buy off the shelf when it fits
Indicative
- What the record shows
- AIxBlock prices off-the-shelf call-centre audio at $5 to $25 per hour against $50 to $300 per delivered hour for custom collection, a 5 to 60 times gap.
- What to do with it
- Buyer: use catalogue data when time-to-data is short, the catalogue matches most of your domain and non-exclusive is acceptable; commission custom only for the gap. Seller of an existing archive: you compete against custom collection cost, which is the ceiling to quote against.
1 source, with the words read
- AIxBlock: off-the-shelf call center audio vs custom(2026-10-11)
“Off-the-shelf call center audio licenses at $5 to $25 per hour of audio depending on language, domain, and licensing tier. Custom collection runs $50 to $300 per hour delivered”
Check membership and reduced-licence tiers before paying list
Strong evidence
- What the record shows
- The Linguistic Data Consortium lists Switchboard-1 Release 2 at $3,000 for non-members, $1,500 under a reduced licence and $0 for members of the 1993 and 1997 years.
- What to do with it
- Before buying from a catalogue, ask for membership, reduced-licence or volume tiers; the same corpus can cost half or nothing.
1 source, with the words read
- Linguistic Data Consortium: Switchboard-1 Release 2 (LDC97S62)(2026-10-11)
“"mm":{"name":"1993, 1997 Members","price":"$0.00"},"nm":{"name":"Non-Member","price":"$3,000.00"},"rl":{"name":"Reduced-License","price":"$1,500.00"}”
An auction result is not final until the sale order
Strong evidence
- What the record shows
- Twenty days after Google won the Spirit auction at $10.0M, micro1 noticed a $12.5M all-cash bid with no financing contingency and proof of funds, a $2.5M (25 percent) premium over the winner and $5.0M over the alternate.
- What to do with it
- Buyer: a clean, fully funded topping bid with better privacy terms can reopen a closed process before the sale hearing. Seller: ask for bid protections and a signed alternate so a late bid raises price rather than delaying the sale.
2 sources, with the words read
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“It describes the price as a $2,500,000 or 25 percent premium over the Successful Bid and a $5,000,000 premium over the Alternate Bid.”
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“states a purchase price of $12,500,000 payable entirely in cash from cash on hand, "no financing contingency, no board or investment-committee consent condition,"”
Pay per accepted episode, and budget for re-collection
Indicative
- What the record shows
- Robot data vendors price per accepted episode (Particula) or per usable demonstration net of a 20 to 30% discard (Robotics Center), and Robotics Center warns that re-collection after the first policy evaluation adds 30 to 50% of the original budget. Quotes below $15 an hour usually exclude rejection and replacement.
- What to do with it
- Compare quotes per accepted or usable unit, never per raw unit; write the acceptance rule and who pays for re-collection into the order before collection starts.
2 sources, with the words read
- Particula Tech, Robotics data collection(2026-10-14)
“Quotes below $15 an hour usually exclude QA rejection and replacement.”
- Robotics Center (RCSV) blog, How Much Does Robot Data Collection Cost in 2026?(2026-03-31)
“Re-collection after first policy evaluation (30-50% of original budget).”
Buy a made-to-order environment as a perpetual internal licence and price redistribution separately
Indicative
- What the record shows
- Robotics Center's custom RL environments ($5,000 to $50,000+) ship by default as a private repo licensed to the buyer for perpetual internal use, open-sourcing is a free add-on, and commercial redistribution rights are negotiated on top. The builder keeps the right to resell unless the buyer pays for more, the same structure Epoch reports for software environments where exclusivity costs 4 to 5 times the non-exclusive price.
- What to do with it
- On either side of an environment deal, split the price into three rights: internal use (the list price), exclusivity, and redistribution. A holder selling an environment or task set should quote the non-exclusive internal licence first and price exclusivity as a multiple; a buyer should not pay for rights it will not use.
2 sources, with the words read
- Robotics Center (RCSV), Custom Reinforcement Learning Environments for Robotics Research(2026-10-16)
“Default delivery is a private Git repo licensed to your team with perpetual internal use. We offer an open-source add-on (MIT or Apache-2.0) at no extra cost when the upstream components allow it. Commercial redistribution rights are negotiable.”
- Epoch AI, An FAQ on reinforcement learning environments(2026-01-12)
“Two RL environment founders independently agreed that exclusive deals are roughly 4-5x more expensive ... non-exclusive ones.”
For either side
Who controls de-identification is worth money
Strong evidence
- What the record shows
- In the same Spirit auction round Mercor offered $5.2M with third-party de-identification or $7.0M if it could de-identify with its own tools, and later $10.0M on in-house terms; the debtors chose the lower third-party bids, saying privacy factors may be outcome-determinative.
- What to do with it
- Seller: decide before pricing whether the buyer may de-identify the raw data itself, and quote both versions, since buyer-controlled scrubbing was worth about 35 percent more here but can fail a regulator or ombudsman. Buyer: a premium for in-house processing only helps where the seller can legally accept it.
2 sources, with the words read
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“Mercor.io submitted the first overbid, offering either $5.2 million using Google's form bill of sale or $7 million if the terms allowed it to complete the deidentification process with its own tools. The Debtors determined that $5.2 million "was the highest bid for that round despite the lower price."”
- Chapter11Cases: The Spirit Airlines deidentified data sale, auction results, objections and the September 30 hearing(2026-09-13)
“the Debtors considered "not only economic factors, but also non-economic factors, such as whether the sale would provide for a documented and acceptable deidentification process," and that "given various privacy laws and other process considerations, these noneconomic factors may be outcome-determinative"”
The same company gets estimates twenty times apart
Strong evidence
- What the record shows
- For a US company of about 100 people and about 10 years, public estimators show Grepped $37.5K to $54K, Ooak $99k to $139k, micro1 $379,290 to $888,088 and Troveo $800K to $1M (16 Oct 2026): the low ends differ about 21 times. Nyne computes the same figure as micro1 and Handshake but displays a random 10 to 15 percent of it, about $38K to $57K at the low end, different on each visit.
- What to do with it
- Seller: run every public estimator before the first call and anchor on the highest credible buyer figure, never on a broker's low owner estimate. Buyer: an estimator figure is not a price; ask what the comparable written offer is.
4 sources, with the words read
- Grepped for Business: payout estimate(2026-10-11)
“Estimated payout range $37.5K to $54K”
- Ooak Data: partner page estimator(2026-10-11)
“Estimated value of your data $99k to $139k This is a preliminary range.”
- Nyne: data partnership estimator (page code)(2026-10-16)
“D.format(Math.round(m*s));if(i){const m=b?1:p?.75:.3,T=14e4+2333*l+11667*a ... function J(){return .1+Math.random()*.05}”
- Troveo: business data estimator(2026-10-16)
“ESTIMATED RANGE $ 800 K $ 0 K – $ 1 M $ 0 M LOW HIGH”
Discount numbers that arrive through an introducer
Strong evidence
- What the record shows
- Grepped's own site configuration calls its referrer calculator 'a deliberately generous range a referrer can show an owner', calibrated so 200 people over 10 years in semiconductors shows $800K to $2.4M, while its owner estimator showed $41.5K to $59.5K for 200 employees and 10 years with revenue undisclosed (read 11 October 2026; the page now opens on its default company, 24 employees and 10 years, at $29K to $41.5K).
- What to do with it
- Seller: ask which calculator produced any figure an introducer quotes, and treat referrer ranges as marketing. Buyer: a seller anchored by a referrer's range needs realised deal evidence early, before the gap becomes a broken negotiation.
3 sources, with the words read
- Grepped: site configuration (config.js)(2026-10-11)
“a deliberately generous range a ... referrer can show an owner ... Calibrated so 200 people × 10 years in semiconductors ≈ $800K–$2.4M.”
- Grepped for Business: payout estimate (the page's default view)(2026-10-17)
“Estimated payout range $29K to $41.5K”
- Grepped for Business: payout estimate for 200 employees and 10 years, revenue undisclosed (shows once the form is filled in)(2026-10-11)
“Estimated payout range $41.5K to $59.5K”
Price exclusivity as a multiple, and never give it by default
Strong evidence
- What the record shows
- Two RL-environment founders told Epoch AI that exclusive deals cost roughly 4 to 5 times non-exclusive ones; Troveo repeats the multiple, and 95 percent of its 7,000+ licensors signed exclusively.
- What to do with it
- Seller: quote non-exclusive first and treat exclusivity as a separately priced, time-limited option (a marketplace contract may take it from you by default). Buyer: pay the multiple only where the data differentiates your model; otherwise buy non-exclusive.
3 sources, with the words read
- Epoch AI: The state of RL environments(2026-01-12)
“Two RL environment founders independently agreed that exclusive deals are roughly 4-5x more expensive than non-exclusive ones.”
- Troveo: AI data licensing(2026-07-23)
“more than 7,000 licensors have signed licensing agreements with Troveo that explicitly cover AI training, 95 percent of them exclusively, with over $20 million paid out to rights holders so far.”
- Troveo: How to sell RL environments(2026-09-24)
“The same environment sold exclusively Roughly 4 to 5 times the non-exclusive price Exclusivity is the biggest lever on price”
A handful of labs set the price
Strong evidence
- What the record shows
- One customer accounted for about 58 percent of Innodata's 2025 revenue (48 percent in 2024); no other customer reached 10 percent.
- What to do with it
- Seller: a vendor's offer is a proxy for one or two labs' budgets, so for unique data try to reach the lab or put several vendors in competition. Buyer (vendor): diversify supply so one lab's pause does not strand your inventory.
1 source, with the words read
- Innodata Inc., Form 10-K for 2025 (SEC)(2026-02-26)
“For the fiscal year ended December 31, 2025, one customer in our Digital Data Solutions (“DDS”) segment accounted for approximately 58% of the Company’s total revenues.”
Most of a multi-year licence can be collected at the start
Strong evidence
- What the record shows
- Getty recognised about $40M of two multiyear deals worth $65M (about 62 percent) in the quarter they were signed; Wiley realised $17M of a $21M content-rights project in its first quarter.
- What to do with it
- Seller: ask for 60 to 80 percent on signature or first delivery even on a multi-year licence; labs have accepted it. Buyer: front-loaded cash is worth a lower total price, so trade one for the other explicitly.
2 sources, with the words read
- Getty Images Q4 2025 earnings call transcript (The Motley Fool)(2026-03-16)
“Two new multiyear deals contributed approximately $40 million of Q4 revenue, out of $65 million total deal value, and will provide $15 million in future revenue during 2026.”
- John Wiley & Sons, Q1 FY2025 earnings presentation (SEC)(2024-09-05)
“Executed a second content rights project valued at $21 million; $17 million realized in Q1”
Headline ceilings are not what deals pay
Strong evidence
- What the record shows
- SimpleClosure's test marketplace recovered about $1M across nearly 100 deals (about $10K each) against a quoted $10K to $100K; Ooak headlines 'up to $300K' while its smallest closed deal was $20K; Turing paid an average in the tens of thousands per failed-startup codebase against an 'up to $1M' headline.
- What to do with it
- Seller: set expectations from realised averages for your size, and treat ceilings as the price of the rare, large, linked archive. Buyer: cite realised averages when a seller anchors on a ceiling.
3 sources, with the words read
- Upstarts Media: SimpleClosure data on what AI labs buy(2026-09-03)
“Yona told Forbes that the test version of his market had recovered $1 million for founders over the previous year, across nearly 100 deals.”
- Ooak Data: partner page(2026-10-11)
“Deals to date range from $20K, the smallest closed, to several hundred thousand dollars.”
- Fortune: micro1 tries to beat Google's bid for Spirit Airlines data(2026-09-14)
“his company had bought five to ten failed-startup codebases, paying an average in the tens of thousands of dollars for each.”
Ask for the per-unit build-up behind a lump sum
Indicative
- What the record shows
- Turing pays $50 per org artifact (tickets, specs, docs) only when bundled with a codebase, against a single payout of up to $1M for a whole operating record; Corpus Bridge's illustration prices 2.1M emails at $41,000 (about 2 cents each) and 96k tickets at $27,500 (about 29 cents each) for a $152,700 total.
- What to do with it
- Seller: count units by system and ask the buyer to price them, since tickets and structured records price at ten times emails per item. Buyer: a per-unit schedule caps what you pay if the delivered volume shrinks after cleaning.
2 sources, with the words read
- Turing: AGI Advance newsletter 57 (Project Lazarus)(2026-03-24)
“Org artifacts (tickets, specs, docs, support threads): $50 per artifact when bundled with a codebase”
- Corpus Bridge: valuation illustration(2026-10-11)
“Gmail 2.1M emails $41,000 Slack 1.4M messages $38,000 Zendesk 96k tickets $27,500 Google Drive 86k files $24,000 QuickBooks 64k invoices $12,400 Excel 12k workbooks $9,800 Estimated value Illustrative figures $152,700”
Headquarters moves the estimate by up to 70 percent
Strong evidence
- What the record shows
- The micro1, Handshake and Nyne estimators weight the US at 1, Canada and Europe at 0.75 and elsewhere at 0.3: for 100 employees and 10 years micro1 shows $379,290 to $888,088 in the US, $284,468 to $666,066 in Europe and $113,787 to $266,426 elsewhere (Handshake about 1 percent more).
- What to do with it
- Seller outside the US: lead with any US customers or operations, since these buyers price geography first. Buyer: in these models a company headquartered in Canada or Europe is priced at 75 percent of the US figure, and one elsewhere at 30 percent.
3 sources, with the words read
- micro1: data payout estimator(2026-10-11)
“Estimated payout range: $113,787 to $266,426 Indicative estimate only.”
- micro1: data payout estimator(2026-10-11)
“Estimated payout range: $284,468 to $666,066 Indicative estimate only.”
- Nyne: data partnership estimator (page code)(2026-10-16)
“if(i){const m=b?1:p?.75:.3,T=14e4+2333*l+11667*a”
Count net, proprietary volume before pricing
Strong evidence
- What the record shows
- SimpleClosure found that about 37 percent of raw code disappeared after removing third-party or generated code and internal duplication, and its CEO says a codebase a quarter of the size can be worth double.
- What to do with it
- Seller: deduplicate and measure before quoting, and hand over a datasheet with net counts, so the buyer's diligence does not cut the price after signature. Buyer: price on post-deduplication counts and write a volume adjustment into the contract.
2 sources, with the words read
- SimpleClosure: What your assets are worth(2026-09-03)
“Across the codebases we analyzed, roughly 37% of raw code disappeared after removing third-party/generated code and internal duplication.”
- Upstarts Media: SimpleClosure data on what AI labs buy(2026-09-03)
““There could be a codebase that’s a quarter of the size, but double the value.””
Archive deals are often one-off: price refreshes, not repeats
Indicative
- What the record shows
- Getty guided 2026 revenue to $948M to $988M (down 3.4 percent to up 0.6 percent), attributing the decline exclusively to one-time revenue recognition from 2025 licensing deals.
- What to do with it
- Seller: assume a static archive sells once per buyer and price it accordingly, or add refresh deliveries that justify renewal. Buyer: a buyer that already holds the archive will pay only for new material.
1 source, with the words read
- Getty Images Q4 2025 earnings call transcript (The Motley Fool)(2026-03-16)
“Revenue guidance: $948 million to $988 million (down 3.4% to up 0.6%), adjusted EBITDA guidance: $279 million to $295 million (down 12.9% to 8.1%), declines attributed exclusively to one-time revenue recognition from licensing deals in 2025.”
Litigation outcomes set a floor for licensing
Indicative
- What the record shows
- The court-approved Anthropic settlement pays about $3,000 per work, four times the $750 statutory minimum, against a statutory ceiling of $150,000 per work.
- What to do with it
- Seller of copyrighted works: a per-work licence well below what unlicensed use has cost a lab is a weak opening. Buyer: licensing with clean warranties is cheaper than the exposure; insist on title and provenance warranties in return.
1 source, with the words read
- Authors Guild: Court grants final approval of $1.5 billion Anthropic copyright settlement(2026-07-21)
“the approximately $3,000 per-work award is “four times the statutory minimum for ordinary infringement” of $750”
Price the term and the life of the trained model separately
Indicative
- What the record shows
- Disclosed AI licences run two to three years (Reddit), at least three years (News Corp and Meta) or three years (HarperCollins and Microsoft); licensing counsel recommends a perpetual right to models already trained plus a finite licence to the dataset, limited to a single model with extension for an additional fee.
- What to do with it
- Seller: grant a fixed dataset term, name the model, and charge for extension to successor models. Buyer: secure perpetual rights in models already trained before agreeing a short dataset term.
3 sources, with the words read
- Terms.Law: An AI training data licence that actually works(2026-10-11)
“For most enterprise data licensors, the answer is a single model with the right to extend on additional fee.”
- Terms.Law: An AI training data licence that actually works(2026-10-11)
“The cleaner structure: a perpetual, worldwide license for training and use of models trained, with a finite license for ongoing access to the dataset itself.”
- eWeek: HarperCollins books to train Microsoft AI models(2024-12-07)
“Microsoft signed a three-year deal with Harper Collins to train an as-yet-unnamed AI model on the major publisher’s catalog.”
Check the buyer's home country before a US data sale
Strong evidence
- What the record shows
- The US Justice Department's Data Security Program forbids a US person from knowingly engaging in data brokerage with a country of concern or a covered person, and a covered person includes any entity with its principal place of business in a country of concern such as China.
- What to do with it
- Screen every buyer and its parent for headquarters and ownership before sharing a sample. Records holding US personal data, including names tied to usernames, cannot go to a China-headquartered lab at any price, so a high offer from one is not a usable anchor.
2 sources, with the words read
- eCFR, 28 CFR 202.301 Prohibited data-brokerage transactions(2026-10-14)
“no U.S. person, on or after the effective date, may knowingly engage in a covered data transaction involving data brokerage with a country of concern or covered person”
- eCFR, 28 CFR 202.211 Covered person(2026-10-14)
“or that is organized or chartered under the laws of, or has its principal place of business in, a country of concern”
A listed crowd vendor keeps 37 to 40 percent after paying its workers
Strong evidence
- What the record shows
- Appen, which defines gross margin as revenue less crowd expenses, kept 40.3 percent in 2025 (39.3 percent in 2024) and 36.7 percent in the first half of 2026, so a client paid about $1.58 to $1.68 for each $1 its crowd earned. Sacra reports that Scale marks up the cost of labour to a gross margin above 50 percent (at least $2 per $1). Expert marketplaces sit lower: Mercor 27 to 33 percent, Handshake about 30 percent (already on the board).
- What to do with it
- Take 30 to 50 percent of a vendor's price as the room between what the lab pays and what the person making the data receives. When an intermediary offers to pay you for work or data, ask what the end buyer pays per unit; a direct deal, or a broker taking 10 to 20 percent, leaves most of that room with you.
3 sources, with the words read
- Appen Limited, ASX announcement: FY25 results and FY26 outlook(2026-02-25)
“Gross margin improved 100 basis points to 40.3% due to greater mix of generative AI projects ... Gross margin refers to revenue less crowd expenses.”
- Appen Limited, ASX release: H1 FY26 investor presentation(2026-08-27)
“36.7% gross margin for H1 FY26, down 30 bps on H1 FY25 ... Small decrease reflects a change in customer and project mix”
- Sacra: Scale AI company page(2026-10-14)
“Scale monetizes per task, marking up the cost of labor giving the company 50%+ gross margin.”
Settle the employee-data question before going to market
Strong evidence
- What the record shows
- The Spirit records sale hearing, first noticed for 19 August 2026, has been adjourned five times and now sits at 6 November; the 9 October notice says the debtors are still negotiating with objectors over personally identifiable or sensitive employee information, which the privacy ombudsman's consumer-only mandate does not cover. Four unions and EPIC (as amicus) object, and EPIC says no employer has previously tried to sell its employees' data as an asset in bankruptcy.
- What to do with it
- Sellers: before a buyer sees an asset schedule, decide how HR files, payroll, timecards, personnel records and union matters are excluded or de-identified, and get workforce or union comfort where it exists; a consumer-privacy clearance alone does not close. Buyers: expect months of delay on any archive with employee data and price the time, or ask for HR systems to be carved out up front.
3 sources, with the words read
- In re Spirit Aviation Holdings, No. 25-11897, Dkt. 1703 (Amended agenda for the 14 October 2026 hearing)(2026-10-12)
“Status: This matter has been adjourned to 11:00 a.m. on November 6, 2026.”
- Chapter11Cases: Spirit Airlines case update, the data sale slips to November(2026-10-14)
“The hearing was originally noticed for August 19 and has now been adjourned five times (ECF Nos. 1488, 1557, 1601, 1663 and 1695). The September 28 notice moved it from September 30 to October 14 (Dkt. 1663). The notice states that the debtors continue to work with objecting parties toward a consensual resolution, including on the treatment of personally identifiable or sensitive employee information.”
- In re Spirit Aviation Holdings, Dkt. 1684 (Consumer Privacy Ombudsman supplemental report, 5 October 2026)(2026-10-05)
“The Ombudsman did not investigate or reach any conclusion about whether the data offered for sale or the deidentification process to be followed presents any risks to affected employees or third parties.”
Certified de-identification is the closing condition, so payment waits for it
Strong evidence
- What the record shows
- In the Spirit sale agreement the $10M is paid only on delivery of de-identified data, and delivery requires an agent's certification to the CCPA standard (HIPAA for any health data) while preserving referential integrity; the debtors hired Tonic.ai as independent agent, which must file an affidavit with the court before the first and every later transfer, and the ombudsman (5 October 2026) asked each certification to state how many passengers were detected and de-identified. On that structure she recommended approval.
- What to do with it
- Sellers: put an independent de-identifier and a per-transfer certificate in the offer package, since that is the structure a court-appointed privacy reviewer accepted; because the price is paid on certified delivery, negotiate a deposit or a payment at signing and a deadline for the buyer's de-identification work. Buyers: expect to name or accept the agent and to wait for the certificate before data moves.
3 sources, with the words read
- In re Spirit Aviation Holdings, Dkt. 1463, Exhibit A (Sale Agreement, Google LLC)(2026-08-14)
“Contemporaneously with the delivery of the Assets to Buyer in accordance with Section 1, Buyer shall deliver the Purchase Price in immediately available funds”
- In re Spirit Aviation Holdings, Dkt. 1684 (Consumer Privacy Ombudsman supplemental report)(2026-10-05)
“Prior to any transfer, and for every transfer thereafter, the third party expert will file an affidavit with the Court certifying that the data transferred has been deidentified.”
- In re Spirit Aviation Holdings, Dkt. 1684 (Consumer Privacy Ombudsman supplemental report)(2026-10-05)
“In conclusion, the Ombudsman recommends to the Court that from a privacy perspective, approval of the sale of the deidentified data would be appropriate.”
In a privacy-sensitive sale, structure beats a higher late price
Indicative
- What the record shows
- micro1's $12.5M proposal (25% above Google's $10M) has not displaced Google two months after the auction: it came with micro1 de-identifying in house, plus offers to exclude sensitive employment material, keep data in the US, destroy raw employment records and give an independent reviewer a 1% sample; the ombudsman asked for further review of micro1 because it would not use an independent party, and the 14 October agenda lists only the Google transaction for approval.
- What to do with it
- Sellers: when a late higher bid arrives, compare it on the terms the court or the seller's own privacy review relies on (who de-identifies, onward transfer, deletion), not price alone; ask the higher bidder to match the structure before using its number. Buyers: a topping bid must copy the winning structure to be taken seriously.
2 sources, with the words read
- In re Spirit Aviation Holdings, Dkt. 1684 (Consumer Privacy Ombudsman supplemental report)(2026-10-05)
“micro1 proposed to conduct the deidentification themselves and not to use an independent party. If the Court considers the company as the successful bidder, the Ombudsman requests an opportunity to conduct further research into their proposal”
- Fortune: micro1 tries to beat Google's bid to buy bankrupt Spirit Airlines data for $12.5 million(2026-09-14)
“Micro1 also proposed excluding sensitive employment material, storing records in the U.S., destroying raw employment records after processing and giving an independent reviewer a 1% sample of the de-identified assets before onward transfer.”
Robot data made to order is exclusive by default: non-exclusive is the discount
Indicative
- What the record shows
- Robotics Center licenses all custom-collected teleoperation data to the buyer exclusively by default and offers non-exclusive licensing at a discount; Particula hands the buyer the corpus outright and does not resell it. Exclusivity is the starting point in robot collection, the reverse of RL environments, where exclusive deals cost 4 to 5 times more.
- What to do with it
- A holder of robot or teleoperation data selling to order should price exclusivity in, not give it away as standard; a buyer should ask for the non-exclusive discount when it does not need sole use.
2 sources, with the words read
- Robotics Center (RCSV), Custom Robot Teleoperation Data Collection(2026-10-14)
“All custom-collected data is licensed to you exclusively by default.”
- Particula Tech, Robotics data collection(2026-10-16)
“Data rights are written into the contract per program, and by default you own the delivered corpus outright. We don't resell collected episodes and we don't fold them into anything we license.”
Keep the communications systems in the package, offer exclusivity, and lead with white-collar headcount
Indicative
- What the record shows
- Buyers named the price drivers on the record in October 2026: micro1's CRO says that without Slack, CRM or email records the rest of a corpus is worth less; Mode.Inc's CEO says value rises with the number of white-collar employees; a codebase a rival lab already holds draws less demand; law and accounting see the strongest demand. Surge AI adds that one company's Slack with varied, deep conversations can be worth more than another's Slack, email and calendars combined.
- What to do with it
- Sellers: include the communication systems (de-identified) rather than selling code or files alone, state white-collar headcount and years of history first, and price exclusivity to one lab as a separate term. Buyers: ask for a conversation-depth sample (threads with disagreement and decisions) before pricing on volume.
3 sources, with the words read
- Business Insider, AI labs are fueling a wild market for your emails and Slack DMs(2026-10-13)
“If a company won't sell its Slack, customer relationship management, or email records, the rest of the corpus is less valuable.”
- Business Insider, AI labs are fueling a wild market for your emails and Slack DMs(2026-10-13)
“Other factors, like exclusivity, matter ... there's lower demand for a codebase that a rival lab already has its hands on. Fields that AI giants are trying the hardest to automate, like law and accounting, see especially high demand.”
- MIT Technology Review, AI labs are buying data from dead startups to train their models(2026-10-13)
“Heiner explains that one company’s Slack history, if it captures conversations with variety and depth, can be worth more than another’s Slack, email, and calendars combined.”
Show clear title to the records, because buyers have been offered stolen archives
Indicative
- What the record shows
- Surge AI's head of environments told Business Insider (13 Oct 2026) that a group offered him a large company's codebase for millions of dollars, which turned out to be an illicit copy shopped to every data vendor; Mercor faces a lawsuit alleging it encouraged contractors to upload employers' work materials without consent, which it denies.
- What to do with it
- Sellers: bring proof of authority up front (a board or officer resolution, the system owner's admin export, the list of excluded third-party and client material) so a careful buyer can clear title quickly. Buyers: ask who holds the admin rights to each source system and whether the signer can bind the company before any price discussion.
1 source, with the words read
- Business Insider, AI labs are fueling a wild market for your emails and Slack DMs(2026-10-13)
“"It became apparent that this group did not have the rights to sell this data," Heiner told Business Insider. "They had illicitly obtained a copy of the codebase and were just going around to all the data vendors saying, 'Can I get a few million bucks for this?'"”
The price is one line. The terms are the rest.
A lever is worth something only against a market rate and a set of terms. The market rates say where a price sits; the licence terms say what the price buys.
We negotiate for holders of data, with every lever above. 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.
Cite this
AIData.Broker, “How prices for AI training data are actually negotiated”, reviewed 17 Oct 2026. https://aidata.markets/negotiate
Free to reuse under CC BY 4.0 if you credit AIData.Broker and link to this page. Each row names the publication that stated it, and words quoted from a source keep that source's own terms.