An offer for company data is a buyer's written proposal to license or buy a defined set of data at a stated price, and it is worth only what its payment trigger, acceptance step and license grant let the company keep. Four things decide that: price, payment trigger, review window and grant of rights. No program cited here publishes its pricing formula, and every term is the program's own statement, checked 2026-11-09 unless another date is given.
An offer is where AI training data licensing turns from a program page into terms.
DataPacts Editorial Team dated every term to its page or filing; the page on how we source, date and check every figure sets out the rules.
What an Offer for Company Data Contains
A written offer names the data, the price and how it is calculated, and points to the terms under which the buyer may use the data. micro1 states that the sole consideration is the License Price in the Order Form.
- Price and unit: a fixed sum, a fee per record or a share of use.
- Payment trigger: the event that makes the money due.
- Acceptance or review step: the period in which the buyer may reject the data.
- License grant: scope, term, exclusivity and who owns the models.
- Conditions: due diligence, eligibility and excluded data.
Patronus AI states that a written offer follows a sample, and Mode states that the seller reviews and approves the offer and its scope.
How AI Data Buyers Calculate Offers
Buyers price an offer from the inputs they publish, a sample of the data and the terms they want, and none of the programs cited here publishes the formula.
The Inputs Buyers Say They Price
The inputs buyers name are company size, years of history, the systems the data comes from, the country and, before a firm offer, a sample. Polyshares states: "We do not pay for a record we have not seen".
- Company size and history: micro1 states 30+ full-time employees, and Handshake AI states 20+ full-time employees and 3+ years.
- Systems connected: Mercor states that "Connecting more tools generally results in a higher payout".
- Country: an input of the Handshake AI estimator, with employees, tools and years.
- Record count: the Tempus and Pathos order form of 2026-10-22 prices USD 100 per delivered record.
Estimates, Illustrations and Written Offers Are Different Documents
A calculator range or illustration on a buyer's or broker's page is marketing, and only a written offer tied to an order form states a price the buyer proposes to pay. Handshake AI calls its estimator an indicative estimate, not an offer, and shows "Earn $100K to $4M+" beside an estimator heading of "up to $100,000" (checked 2026-11-09). Mode marks its calculator "Indicative only", and Regulr, a broker, labels its examples "illustrations, not offers". The range on Turing's page comes from a third-party calculator and is not a Turing offer.
Set a range against what company data is worth to AI companies, where such ranges sit beside disclosed deal values.
Four Ways an Offer Prices the Data
Offers for company data price the data in one of four ways: a fixed sum, a fee per delivered record, an upfront fee plus a revenue share, or a revenue share alone. The table compares them with each party's own example, as of 2026-11-09.
| Structure | How the price is set | Example | Who states it | As of | Source |
|---|---|---|---|---|---|
| Fixed sum | one license price in an order form, or agreed before data is shared | micro1: the License Price in the Order Form is the sole consideration; Endless (aggregator): fixed payment agreed before anything is shared | micro1; Endless | checked 2026-11-09 | micro1.ai master terms; endless-data.com |
| Fee per delivered record | unit price times records delivered, plus fees for updates | Tempus / Pathos Neuro Order Form: USD 100 per delivered record, up to USD 55.0 million including USD 5.0 million for updates | Tempus AI 10-Q | 2026-10-22 | sec.gov |
| Upfront fee plus revenue share | fee at signing, then a share of downstream use | Labelbox Prism: upfront licensing fee at signing plus revenue share | Labelbox | checked 2026-11-09 | prism.labelbox.com/faq |
| Revenue share only | share of each resale or use, nothing upfront | Turing Project Lazarus: "Nothing upfront", paid every time the record sells; Protege: revenue share with each use | Turing; Protege | checked 2026-11-09 | lazarus.turing.com; withprotege.ai |
The revenue-share percentage is not published by Labelbox, Turing or Protege. See when and how AI data buyers pay for the mechanics.
Comparing Offers Side by Side
Two offers can be compared only after each is reduced to the same seven points: price and unit, costs the company still bears, payment trigger, review window, grant, deletion and models, and open conditions. The table gives the published contrast for each, as of 2026-11-09.
| Point to compare | What to read in the offer | Published contrast | Source |
|---|---|---|---|
| Price and unit | fixed sum, per record or share; quantity assumed | fixed License Price vs USD 100 per delivered record | micro1.ai; sec.gov |
| Costs the company still bears | who extracts, de-identifies, packages; fees | Mercor and Turing state they cover them; listing fees of 1.5 to 3 % on AWS Data Exchange | mercor.com; lazarus.turing.com; docs.aws.amazon.com |
| Payment trigger | signing, delivery, acceptance, invoicing, approval, each sale | at signing (Labelbox); on delivery, typically 2 to 4 weeks after signing (Mercor); acceptance vs invoicing on one micro1 page | prism.labelbox.com; mercor.com; micro1.ai |
| Review window | length, what silence means, what survives | 30 business days then deemed acceptance, 60-day latent defects, clawback survives (micro1 terms) vs page "30 days"; Patronus up to 30 days | micro1.ai master terms; data.patronus.ai |
| Grant | exclusivity, term, sublicensing, ownership of derivatives | nonexclusive, irrevocable, perpetual, sublicensable (micro1) vs exclusive by default (Labelbox) vs 90-day sole selling window (Turing) | micro1.ai; prism.labelbox.com; lazarus.turing.com |
| Deletion and models | deletion deadline; models trained before deletion | at most 120 days after delivery completes, no duty to retrain (micro1) vs kill clause 30 days, "cannot be unlearned" (Labelbox) | micro1.ai; prism.labelbox.com/faq |
| Open conditions | due diligence, eligibility, excluded data | Handshake AI: due diligence and acceptance of its terms; micro1 exclusions "among other categories" | joinhandshake.com; micro1.ai |
Headline Price vs What the Company Keeps
The headline price is not what the company keeps when extraction, de-identification or an intermediary's fee comes out of it. Mercor states that it covers extraction, anonymization and transfer with no setup fees or deductions, Turing states that it covers de-identification and packaging, and Polyshares states "no fee and no commission" (all checked 2026-11-09).
Regulr states that the buyer pays Regulr when a license closes, and AWS Data Exchange, a marketplace, charges listing fees of 3 % on public offers and 3 %, 2 % or 1.5 % on private offers by contract value (effective 2024-01-05). All programs are listed under AI data partnership payouts and program terms compared.
What Triggers the Payment
The payment trigger decides when money arrives and what the buyer can still refuse, and buyers publish six different triggers, each checked 2026-11-09.
- At signing: Labelbox states an upfront licensing fee at signing.
- On delivery: Mercor states a wire on delivery of the anonymized dataset, typically 2 to 4 weeks after signing.
- On acceptance: Patronus AI states payment "Direct, on acceptance".
- After invoicing: micro1 states payment within 30 days of invoicing.
- On approval plus Net terms: Polyshares states Net 30 to 60 after approval.
- On each resale: Turing states "Nothing upfront", with payment every time the record sells.
The micro1 data-partnerships page states payment both "in full within 30 days of acceptance" and "within 30 days of invoicing", so it does not say which event starts the clock.
Review Windows Before the Offer Becomes Final
Most offers become final only after the buyer reviews the delivered data, so the review length and the effect of silence matter as much as the price. micro1's Data License Master Terms give 30 business days after delivery to reject, after which the data is deemed accepted, and 60 days after acceptance for latent-defect claims (s.1.3). Its data-partnerships page says "30 days to review and approve", and the two are not reconciled. The s.4.3 clawback, a refund if the Data Manifest is materially inaccurate, survives acceptance.
Under s.1.2 the license attaches only after acceptance and payment (re-read 2026-11-10), and Patronus AI states up to 30 days of review. The clause itself is explained under acceptance, rejection and deemed acceptance of delivered data.
Grant Terms That Change What the Price Buys
The same price buys very different rights, because buyers' published grants range from a perpetual, sublicensable license to an exclusive license with a kill clause. micro1 states that the licensee owns derivatives, models, weights and outputs (s.2.2).
- Perpetual and sublicensable: micro1 states a nonexclusive, irrevocable, perpetual license (s.1.2).
- Exclusive by default with a kill clause: Labelbox states deletion within 30 days under the clause, triggered for example by material breach, and that "Models already trained on the data cannot be unlearned".
- Time-limited sole selling window: Turing states that it is sole seller for 90 days after signing, then renew or opt out.
- Defined term: Endless states a non-exclusive license with deletion at the end of the agreed term, and Polyshares states "Defined material for a defined term" on its home page.
Each grant clause of an AI data licensing agreement is explained on the contract page.
Conditions Still Open When the Offer Arrives
An offer that still depends on due diligence, eligibility or data exclusions is a range, not a price. Handshake AI states that payouts depend on the outcome of its due diligence and acceptance of its licensing terms, which it does not publish (re-read 2026-11-10). micro1 lists excluded categories such as PHI and biometric identifiers "among other categories", so the list is not exhaustive (s.5.7, re-read 2026-11-10).
Questions to Put to a Buyer Before Answering an Offer
Eight questions turn an offer into terms that can be compared, and each answer belongs in writing.
- Which unit and quantity is the price based on?
- Who pays for extraction and de-identification?
- What triggers payment, and on which date?
- How long is the review, and what does silence mean?
- What survives acceptance, such as a clawback?
- What are the scope, term and exclusivity of the grant?
- What happens to the data and to trained models when the grant ends?
- Which conditions are left open?
The answers are the material for negotiating an AI data licensing deal.
Accept, Counter or Decline: Benefits, Risks, Downsides and Alternatives
Accepting, countering and declining are all valid answers to an offer, and declining is the right one when the terms cost the company more than the price pays.
Benefits of accepting
- A written price replaces an estimate.
- Some programs carry extraction and de-identification.
- A term-limited grant ends on a known date.
Risks
- A perpetual, irrevocable grant cannot be recalled (micro1 s.1.2).
- Models already trained "cannot be unlearned" (Labelbox).
- The no-prior-grant warranty (micro1 s.5.1) limits later sales of the same data.
Downsides
- Money may wait for deemed acceptance, invoicing or approval.
- Exclusivity by default or a sole selling window blocks parallel use while it runs.
- Licensing income can concentrate: Reddit's 10-Q for the quarter ended 2026-09-30 states that "substantially all" contract value comes from two partners approaching the end of their terms.
Alternatives
- Counter on terms.
- Ask other buyers for written offers.
- License a narrower set of the data.
- Decline and keep the data, covered under alternatives to selling data to AI labs.
When an Offer Calls for Caution
An unsolicited offer that names a large sum before anyone has seen the data is a reason to check who is offering before replying. SimpleClosure's CEO, whose company is a competing intermediary, reported on 2026-10-01 that SHFA.ai ads promised founders USD 1 million, 3 million and 5 million.
The checks on whether a data licensing email is legitimate are on the scams page.
Where Offer Evaluation Sits in the Process to Sell or License Company Data
Offers are evaluated after the company has a value range and before it signs, so the steps on either side decide how much an offer can be trusted. The full sequence to sell or license company data to AI companies runs as follows, with this page after valuation.
- Should you sell or license your data
- Company data valuation
- Choosing a buyer for your data
- Running several AI data buyers in parallel
- Due diligence in an AI data deal
- Term sheets, order forms and LOIs
- Delivering licensed data to an AI buyer
Offers in a Court-Supervised or Defunct-Company Sale
When a company is closed or insolvent, offers usually arrive as bids, and a court or administrator decides which one is accepted. In Spirit Airlines' chapter 11 (Bankr. S.D.N.Y. 25-11897 (SHL), Dkt. 1463, auction 2026-08-14), the successful bid for de-identified data was Google LLC's USD 10,000,000 and the alternate bid was Mercor.io Corporation's USD 7,500,000; the sale was not approved as of 2026-11-09, with a hearing set for 2026-11-17 (Dkt. 1779). Selling data from a defunct company follows the same pattern. See data assets in insolvency for how administrators sell them.
Programs and Profiles Cited on This Page
The offer terms above come from these programs' own pages and filings, listed alphabetically.
- Endless (aggregator): Endless profile
- Handshake AI: Handshake AI data buyer profile
- Labelbox Prism: Labelbox Prism profile
- Mercor: Mercor data buyer profile
- micro1: micro1 data buyer profile
- Mode
- Patronus AI: Patronus AI profile
- Polyshares: Polyshares profile
- Protege
- Regulr (broker)
- Tempus and Pathos
- Turing Project Lazarus: Turing Project Lazarus profile
The overview of company data partnership programs compares them, and who buys AI training data covers every buyer.
Frequently Asked Questions
How do AI data buyers calculate offers?
Buyers start from company size, years of history, systems and country, then set a firm price from a sample, and none publishes the formula. Handshake AI's estimator takes employees, tools, years and country (checked 2026-11-09). Advertised and confirmed figures are compared under reported payouts to data sellers.
Is a buyer's estimate an offer?
No; buyers label their calculators indicative or illustrative, and only a written offer tied to an order form proposes a price. Handshake AI calls its result an indicative estimate and Mode marks its calculator "Indicative only" (checked 2026-11-09). Every published calculator is run under buyer estimators side by side.
When does an offer for company data become binding?
Under published master terms, the price that binds is the one in the signed order form, and the license itself may attach only after acceptance and payment. micro1 states both (checked 2026-11-09, condition re-read 2026-11-10). This page states no general rule on when an emailed offer or letter of intent becomes binding; the document class is covered under term sheets, order forms and LOIs.