Data monetization for AI means earning money from company data by letting AI developers train on it, through one of four models: a license, a sale, a revenue share or ongoing data-as-a-service access. Each model decides who keeps the data, when money arrives and who controls later use. micro1's Data License Master Terms (updated 2026-11-08) take a perpetual, non-exclusive grant and leave the data with the company, a bankruptcy court records Google LLC's successful bid of USD 10,000,000 for Spirit Airlines' data as an outright sale (2026-08-14, not approved as of November 9, 2026), Turing states a revenue share with "Nothing upfront" (checked 2026-11-09), and Reddit's S-1 of 2024-02-22 shows USD 203.0 million in aggregate for API-based licenses with unnamed counterparties.
Choosing a model is one decision within AI training data licensing: the license is the most common published form, a sale ends the right to license again, a revenue share depends on resale the company does not control and data-as-a-service pays over a term. Not monetizing remains a valid result.
DataPacts Editorial Team checked every figure here against its primary source on November 9 and 10, 2026; the page on how we source, date and check every figure sets out the rules.
What Data Monetization Means When the Data Trains AI Models
Data monetization is turning data a company holds into revenue or savings, and with AI buyers it means external monetization: another party pays for the right to train models on the data. Internal monetization, such as better operations or data-enhanced products, is not developed here. The external models differ by what transfers to the buyer, not by the type of data, and advertising, consumer profiles and people data fall outside this page.
The four models used on this page (license, sale, revenue share and data-as-a-service) are this reference's grouping, not an industry standard, and each one is evidenced below by a published agreement, filing, court record or program page. No public figure exists for how many AI data deals follow each model.
Four Ways to Earn From Data With AI Buyers, Side by Side
The four models differ in what the buyer receives, whether the company keeps the data and when money arrives. The table compares them as of 2026-11-09, each with a public example and its source.
| Model | What the buyer receives | Company keeps the data | When money arrives | Public example | Source |
|---|---|---|---|---|---|
| License | right to train, scope set by the grant | yes | at signing, delivery or acceptance; or per year | micro1 master terms, non-exclusive and perpetual; Kopinor NOK 45,000,000 a year for 2026; Tempus USD 100 per delivered record | micro1.ai (updated 2026-11-08); kopinor.no agreement s.17; Tempus 10-Q |
| Sale | the data itself, free and clear | only copies for wind-down or legal compliance | on delivery, after court approval in bankruptcy | Spirit Airlines Bill of Sale; successful bid USD 10,000,000, not approved as of 2026-11-09 | Bankr. S.D.N.Y. 25-11897, Dkt. 1463, 1779 |
| Revenue share | data placed with labs or resold | yes (license to the intermediary) | each use or resale; percentage mostly unpublished | Turing Project Lazarus, Protege, SuperAnnotate, Gradient Health | their own pages, checked 2026-11-09 |
| Data-as-a-service | continuing access to updating data via API, feed or subscription | yes | over the term | Reddit continuous API access, USD 203.0 million aggregate; Tempus and BMS USD 137.5 million to 2031 | Reddit S-1; Tempus 10-Q |
The models combine in practice: Tempus licenses records to Pathos AI and gives Bristol Myers Squibb platform access in the same filings, so one company can run a license and a data-as-a-service deal at once.
License: The Company Keeps the Data and Grants Training Rights
A license gives an AI buyer the right to use data for training while the company keeps ownership and its own copy, and most published company data programs use this model. Labelbox states on its Prism FAQ that the seller keeps full ownership and continued use of the data (checked 2026-11-09).
Grant Scope: Who Else May Train on the Same Data
A license grant ranges from non-exclusive and perpetual to exclusive by default, and the scope decides whether the company can license the data again. micro1's master terms grant a worldwide, royalty-free, fully paid-up, non-exclusive, irrevocable, perpetual and sublicensable right (s.1.2, updated 2026-11-08), and the licensor also warrants that it has not granted a third party rights to the same data on the same or substantially similar terms (s.5.1), which narrows what a non-exclusive grant leaves free. The licensee owns derivatives, models, weights, outputs and synthetic data (s.2.2), and deletes the data at the earlier of 120 days after delivery or termination with no duty to delete or retrain models (s.4.4).
At the other end, Labelbox states that Prism licenses are exclusive by default and that "Models already trained on the data cannot be unlearned", Turing states a 90-day window in which it is the only seller of a record, and the Tempus Neuro order form of 2026-10-22 carries an 18-month exclusivity side letter. Kopinor's agreement with the National Library of Norway is non-exclusive. The price and lock-in of exclusive and non-exclusive data licenses are compared on their own page.
License Fees: One Sum, Per Record, Per Work or Per Year
A license is paid as one fixed sum, a fee per delivered unit or a recurring annual fee, and each basis appears in a filed or published agreement. Four bases are documented:
- Fixed sum: Tempus AI reports in its 10-K for fiscal 2025 (filed 2026-02-24) a data license with Pathos AI of USD 200 million over three years, including USD 50 million upfront, for foundation model training only; micro1's grant is "royalty-free, fully paid-up", so any sum is agreed outside the master terms.
- Per record: the Tempus Neuro order form of 2026-10-22 sets up to USD 55.0 million, made up of USD 100 per delivered record (USD 50.0 million for 500,000 records over five years) and USD 5.0 million for record updates.
- Per work: the Authors Guild states that a HarperCollins AI licensing offer of 2024-11-19 pays USD 5,000 per title, split 50-50 with the opted-in author; the buyer is not named in that statement.
- Per year: the Kopinor agreement of 2025-12-19 sets NOK 45,000,000 for 2026, index-adjusted from 2027 (s.17).
How much a license fee should be belongs to valuation and is not estimated here. License income can also fall: Shutterstock's Data, Distribution, and Services revenue was USD 15.8 million in the third quarter of 2026 against USD 65.7 million a year earlier, down 76 % (10-Q for the period ended 2026-09-30). Each license appears in the AI data licensing deals tracker with its source.
Sale: Ownership of the Data Passes to the Buyer
In a sale the company transfers the data itself, gives up the right to license it again and usually keeps copies only for narrow purposes, and the clearest public example is a court-supervised bankruptcy sale. The Form of Bill of Sale in Spirit Airlines' sale motion (Bankr. S.D.N.Y. 25-11897, Dkt. 1463, 2026-08-14) sets these terms:
- The seller "irrevocably sells" the assets free and clear under 11 U.S.C. 363.
- Personal data is excluded, and a Deidentification Agent de-identifies the data to the CCPA standard (45 CFR 164.514 for health data) at the buyer's cost.
- The seller keeps copies only for wind-down or legal compliance and may not sell the assets to others.
- Payment is contemporaneous with delivery.
- Representations do not survive closing, and the assets are sold as is.
On 2026-08-14 the Notice of Auction Results records Google LLC as successful bidder at USD 10,000,000 and Mercor.io Corporation as alternate bidder at USD 7,500,000 (Dkt. 1463); the sale is not approved as of November 9, 2026, and the hearing is set for November 17, 2026 (Dkt. 1779). A sale of consumers' personally identifiable information in bankruptcy against a restrictive privacy policy needs a consumer privacy ombudsman (11 U.S.C. 332) and court findings under 11 U.S.C. 363(b)(1), as set out in the rules for US bankruptcy data sales. The 23andMe sale is a contrast: substantially all assets, including the genetic database, passed to TTAM Research Institute for USD 305.0 million in cash plus assumed liabilities (sale order 2025-06-27, closed 2025-07-14, 8-K), and it was an asset sale, not an AI-training license.
Revenue Share: Income Each Time the Data Is Used or Resold
Under a revenue share the company is paid a percentage when the buyer or intermediary uses or resells the data, so income depends on sales the company does not control. Four programs state the model, listed alphabetically, each as its own statement checked 2026-11-09:
- Gradient Health states a revenue share each time data is used and gives a worked example of its own: a USD 3 million deal, a 40 % contribution and a "typical" 20 % share give USD 240,000.
- Protege states revenue share payouts "with each use" and does not publish the percentage; its provider terms (s.1.1) take a worldwide, royalty-free, non-exclusive license, and existing licensees keep access 12 months after termination (s.7.2).
- SuperAnnotate states that payment comes "on delivery, and on each relicense", with an indicative valuation of USD 100K to 1M+ (checked 2026-11-09).
- Turing Project Lazarus states "Nothing upfront", a non-exclusive license in which "the same record can go to more than one lab", and a first 90 days in which Turing is the only seller.
Apart from Gradient's example, no program publishes its split, and the share continues only while the intermediary keeps selling. The full Turing Project Lazarus revenue share terms are on its profile.
Upfront Fee Plus a Share of Later Use
A hybrid pays a fee at signing or delivery and a share of later use, which moves some income forward and keeps a stake in resale. Labelbox states for Prism an upfront licensing fee at signing plus a revenue share on downstream use, exclusive by default, and Replay states upfront cash plus a perpetual revenue share (both checked 2026-11-09).
- Labelbox Prism: upfront fee, then a share of downstream use.
- Replay: upfront cash, then a perpetual share.
Neither publishes the percentage.
Marketplace Listings Where the Platform Keeps a Commission
On a data marketplace the company lists a dataset, sets its own terms and the platform keeps a commission on each sale. Datarade states three provider plans: commission-only at USD 0 a year plus 30 %, Bronze at USD 6,000 a year plus 20 % and Silver at USD 12,000 a year plus 15 % (checked 2026-11-09), and Opendatabay's fees page (2026-10-06) lists 30 % on sales of USD 1 to 100 and 5 % on sales of USD 250,001 and above.
A platform lists data and earns a commission, so it is not the buyer. Fees and eligibility are compared among AI training data marketplaces with their fees.
Data-as-a-Service: Ongoing Access Through an API or Feed
Data-as-a-service gives the buyer continuing access to data that keeps updating, through an API, a feed or a subscription, and is paid for over the term rather than per copy.
Reddit's S-1 of 2024-02-22 discloses data licenses signed in January 2024 with aggregate contract value of USD 203.0 million over two to three years, continuous API access and unnamed counterparties, and a minimum USD 66.4 million of revenue in 2024. Google announced on 2024-02-22 Data API access to "display, train on, and otherwise use" Reddit content, and OpenAI announced on 2024-05-16 Data API access for ChatGPT and new products; neither announcement states a value.
Reddit's 10-K for fiscal 2025 states that licensing revenue is generally recognized ratably over the license period, as a right to access, and its 10-Q for the third quarter of 2026 states that "substantially all" licensing contract value comes from two licensees "approaching the end of their respective current terms". The deal terms are set out in Reddit's AI data licensing deals with Google and OpenAI.
Tempus shows the model with company data. Its Onco statement of work with Bristol Myers Squibb (2026-10-14) provides de-identified oncology records plus Lens platform access and services for USD 137.5 million in aggregate fixed payments through 2031-12-31, and BMS may terminate on 12 months' notice. The record-update fee in the Pathos order form is a feed element inside a license.
Subscription also exists on the buyer side: the Linguistic Data Consortium charges for-profit members USD 40,000 for a Subscription membership (checked 2026-11-09), which shows the model for corpora, not for a company's own archive.
Per-Request Access for Website Content
Website owners can charge AI crawlers per request, a metered form of data-as-a-service for public web content rather than company archives. Cloudflare's pay per crawl is in closed beta (documentation of 2026-07-28): the site owner sets a price per zone, Cloudflare is Merchant of Record, the response code is HTTP 402 and payouts are monthly (payouts page of 2026-04-23). The model is covered under pay-per-crawl and crawler licensing.
Choosing a Monetization Model: What Each One Decides
The choice of model decides five things: whether the company keeps the data, when cash arrives, who controls resale, how long income lasts and how much work stays with the company. The table sets the four models against those decisions as of 2026-11-09.
| Decision | License | Sale | Revenue share | Data-as-a-service | Source |
|---|---|---|---|---|---|
| Ownership after the deal | company | buyer | company | company | micro1 s.1.2; Spirit Dkt. 1463 |
| Cash timing | fixed or per unit | once on delivery | uncertain, per use | recurring over term | Tempus 10-K, 10-Q; Spirit Dkt. 1463; Turing page |
| Who controls resale | company if non-exclusive; limited by the no-prior-grant warranty | buyer | intermediary | company | micro1 s.5.1; Spirit Dkt. 1463; Protege terms |
| Income duration | term of the grant; can fall | none after the sale | while the intermediary sells | while access continues; concentration risk | Reddit 10-Q Q3 2026; Shutterstock 10-Q |
| Work left with the company | delivery and acceptance | delivery and de-identification support | ongoing reporting checks | continuous delivery and updates | Tempus 10-Q; Spirit Dkt. 1463 |
| Revenue timing (ASC 606, IFRS 15) | point in time (right to use) or over time (right to access) | not a license | royalty exception for usage-based fees | over time (right to access) | ASU 2016-10; IFRS 15 B56, B63 |
A sale gives up the most control and a non-exclusive license the least. Because the model shapes what company data is worth to AI companies, what company data is worth to AI companies belongs after the choice, not before it.
Benefits, Risks, Downsides and Alternatives of Each Model, Including Not Monetizing
Every model trades control for income, and none removes the option of keeping the data unmonetized.
Benefits
- A non-exclusive license keeps ownership and leaves room to license the data again (micro1 s.1.2; Kopinor).
- A sale pays once, on delivery (Spirit Dkt. 1463).
- A revenue share needs no upfront price (Turing: "Nothing upfront").
- Data-as-a-service recurs over the term (Reddit; Tempus and BMS).
Risks
- Trained models cannot be unlearned (Labelbox, checked 2026-11-09).
- The licensee owns models and weights and has no duty to retrain (micro1 s.2.2, s.4.4).
- Income can fall: Shutterstock's Data, Distribution, and Services revenue is down 76 % in the third quarter of 2026.
- Revenue-share percentages are mostly unpublished (Protege, Turing).
- Reddit reports licensing value concentrated in two licensees near the end of their terms.
Downsides
- Exclusivity limits other licenses (Labelbox default; the Tempus Neuro side letter).
- A sale ends the right to license the data again.
- The no-prior-grant warranty limits what a non-exclusive grant leaves free (micro1 s.5.1).
- Ongoing delivery and updates stay with the company (Tempus and Pathos; Tempus and BMS).
Alternatives, including not monetizing at all
- Use the data internally to improve operations or products.
- Offer a de-identified subset instead of the full archive.
- Grant a non-exclusive license instead of selling.
- Choose a hybrid instead of a pure revenue share.
- Keep the data and monetize nothing.
Each path is set out among the alternatives to selling data to AI labs.
Where the Monetization Model Sits in the Process to Sell or License Company Data
The monetization model is chosen early, after the inventory and rights audit and before a buyer is picked, because it decides which buyers fit. The full sequence to sell or license company data to AI companies places it between the rights audit and the choice of buyer.
- Data inventory
- Rights audit
- Choose the monetization model (this page)
- Choosing a buyer for your data
- Selling direct or through a broker
- Evaluating an offer for your data
- AI data licensing agreements
- When and how AI data buyers pay
Accounting and Tax Treatment by Model
The model changes when license income is recognized and how US tax rules classify the payment. Under ASC 606-10-55-58A a right to access is recognized over time and under 55-58B a right to use at a point in time (per ASU 2016-10), while IFRS 15 B56 and the B63 royalty exception apply to the same distinction. For US tax, 26 CFR 1.861-18(f)(2) treats a transfer of a copyrighted article as a sale if the benefits and burdens of ownership pass and otherwise as a lease, and 1.861-19 treats cloud transactions as services (T.D. 10022, tax years beginning on or after 2025-01-14). Self-created works are not capital assets under 26 U.S.C. 1221(a)(3). See accounting for data licensing revenue for recognition.
Tax on data licensing income covers VAT and withholding. This is information, not tax or accounting advice.
Monetization Models for Healthcare Data
Healthcare data is monetized under the same four models, with de-identification and a revenue share common in published programs. Gradient Health states a revenue share, and the Spirit Bill of Sale refers to 45 CFR 164.514 for health data; how healthcare providers licensing data to AI choose a model is covered on the healthcare page.
Programs and Deals Cited on This Page
The program terms above are each program's own statement, checked November 9, 2026; deal figures come from filings, agreements and court records. The programs and deals cited are listed alphabetically.
- Cloudflare pay per crawl
- Datarade
- Gradient Health
- Kopinor agreement
- Labelbox Prism
- Linguistic Data Consortium
- micro1: micro1 data buyer profile
- Opendatabay
- Protege
- Replay
- Spirit Airlines data sale
- Tempus deals
The overview of company data partnership programs compares program terms, AI data partnership payouts and program terms compared lists them with dates, and who buys AI training data covers every buyer.
Frequently Asked Questions
What is data monetization?
Data monetization is turning data into revenue or savings, internally through better operations or externally by licensing, selling or providing access to it. With AI buyers the external route applies: a license, a sale, a revenue share or ongoing access, each with different rights for the company.
How do companies monetize data for AI training?
Companies license data to AI developers, sell it outright, take a revenue share through an intermediary or provide ongoing access through an API or feed. The license is the model in micro1's master terms, the sale is the Spirit Airlines Bill of Sale, the share is Turing's and the API is Reddit's.
What are examples of AI data monetization?
Public examples include Reddit's API licenses, Tempus AI's per-record license to Pathos AI and Kopinor's annual fee from the National Library of Norway. Reddit's S-1 shows USD 203.0 million in aggregate (2024-02-22), Tempus sets USD 100 per delivered record (2026-10-22) and Kopinor sets NOK 45,000,000 for 2026.
Is a data monetization platform a buyer?
No; a marketplace, broker or aggregator places or lists data and earns a commission or fee, while the AI developer that receives the license is the buyer. Datarade and Opendatabay publish commissions, and Endless states "No commission on your side" and names micro1, Mercor and Handshake as the buyers it works with (checked 2026-11-09). Intermediaries are compared among AI data licensing brokers and sell-side advisers.