Wikimedia, LLC, a subsidiary of the Wikimedia Foundation, named Amazon, Meta, Microsoft, Mistral AI and Perplexity as new Enterprise partners on January 15, 2026, and the Associated Press licensed part of its text archive to OpenAI on July 13, 2023, each without a disclosed value.
Mission, donor terms, beneficiary privacy and tax status decide what a nonprofit may license at all.
Nonprofit data is one part of AI training data licensing, and deciding not to license, or to release data openly, is a valid outcome.
DataPacts Editorial Team checked the figures here against their primary sources on November 9, 2026, and re-read the Mode page on November 16, 2026; the page on how we source, date and check every figure sets out the rules.
What Nonprofits Hold That AI Developers Ask For
Nonprofits hold seven kinds of data AI developers ask for: reference content, news archives, web archives, research corpora, speech datasets, members' works and operational records about programs, beneficiaries and donors. The table compares each kind with an example and what decides licensing (as of 2026-11-16).
| Holding | Example on record | What decides licensing | Source |
|---|---|---|---|
| Reference content | Wikipedia: "more than 65 million articles in over 300 languages" (Wikimedia Enterprise, 2026-01-15) | contributors' license terms; subsidiary contracts | enterprise.wikimedia.com |
| News archives | AP text archive, part licensed to OpenAI (2023-07-13) | staff and contributor rights | ap.org |
| Web archives | Common Crawl CC-MAIN-2026-43: 2.16 billion pages (2026-10-21) | its terms of use (updated 2024-03-07); site owners' opt-outs | commoncrawl.org |
| Research corpora | Ai2 Dolma 3 pool: over 9 trillion tokens (Ai2 claim); LAION-5B: 5.85 billion image-text links | public license chosen | huggingface.co; laion.ai |
| Speech datasets | Common Voice 27.0: 42,593 hours, 295 locales (2026-09-11) | contributor consent; platform terms | github.com/common-voice |
| Members' works | STIM members' music under its opt-in AI license (from September 2025) | members' explicit consent | stim.se |
| Operational records | program, beneficiary and donor records; Mode names nonprofits among eligible sellers (read 2026-11-16) | mission, gift and grant terms; privacy law | data.mode.inc |
Operational records are the kind most bound by the people they describe. Nonprofits are one of the types among who sells data to AI companies.
How Nonprofits License Data: Routes on Record
Five routes are on public record for nonprofits: a commercial subsidiary, a direct archive license, open release under a public license, a company data program and a platform or collective; none has a published price. The table compares the routes by example and published terms (as of 2026-11-16).
| Route | Example | Terms as published | Money terms published | Source |
|---|---|---|---|---|
| Archive license by a cooperative | The Associated Press and OpenAI (2023-07-13); Google (2025-01-15, real-time feed for Gemini) | part of AP's text archive; AP uses OpenAI technology | not disclosed | ap.org; blog.google |
| Collective license | STIM (opt-in AI license); RSL Collective (paid and attribution licenses) | STIM: only works whose creators consented; RSL: declared in robots.txt | STIM: fixed fee plus a share of AI service income; RSL: rates not published | stim.se; rslcollective.org, checked 2026-11-09 |
| Commercial subsidiary | Wikimedia, LLC (single-member LLC of the Wikimedia Foundation) | Enterprise APIs: on-demand, snapshot, realtime | not disclosed; no pricing | enterprise.wikimedia.com, checked 2026-11-09 |
| Company data program | Mode (buyer) | buys a filtered, de-identified copy; home page: 30+ employees "for all other businesses or nonprofits" | Mode states "$100K–$5M", indicative only | data.mode.inc, checked 2026-11-09, eligibility re-read 2026-11-16 |
| Open release | Common Crawl (free archive); Ai2 Dolma (ODC-BY); LAION (CC BY 4.0 metadata; Re-LAION-5B Apache-2.0) | public license; no fee | none (no fee) | commoncrawl.org; huggingface.co; laion.ai |
| Platform listing | Mozilla Data Collective (platform, not a buyer) | provider sets price and license (Compensated Datasets, 2026-07-29) | "keep 100% of revenue" (its claim); no platform fee stated | mozilladatacollective.com, checked 2026-11-09 |
A Commercial Subsidiary Selling API Access: Wikimedia, LLC
The Wikimedia Foundation, Inc. does not sell Enterprise access itself: the Enterprise APIs are offered by Wikimedia, LLC, a single-member limited liability company of the Foundation, as the Enterprise terms state (checked November 9, 2026).
On January 15, 2026, Wikimedia Enterprise listed Amazon, Meta, Microsoft, Mistral AI and Perplexity as new partners, with access on demand, by snapshot and in realtime. No value or pricing is published, and the partners are listed under Wikimedia Enterprise AI deals and their terms.
A Not-for-Profit Cooperative Licensing Its Archive: The Associated Press
The Associated Press, a US not-for-profit news cooperative, licensed part of its text archive to OpenAI on July 13, 2023, with no value disclosed, and AP states that it uses OpenAI technology and product expertise.
On January 15, 2025, Google announced that AP delivers a real-time feed for the Gemini app, again without a value.
Open Release Under a Public License: Common Crawl, Ai2 and LAION
Several nonprofits release training data openly instead of licensing it for a fee, and the license they choose sets what AI developers may do with it.
The archive itself is described under Common Crawl's contents, license and legal status. The Common Crawl Foundation states on its FAQ that it is a 501(c)(3) nonprofit offering free archives collected since 2008.
The Allen Institute for Artificial Intelligence, "a Seattle based non-profit AI research institute" in its terms publishes Dolma under ODC-BY since April 15, 2024. LAION e.V. published LAION-5B, 5.85 billion links and captions without the images, with metadata under CC BY 4.0, on March 31, 2022.
It took the dataset down on December 19, 2023, and released Re-LAION-5B with 5,526,641,167 pairs under Apache-2.0 on August 30, 2024. The images keep their own copyright.
Operational Records Through Company Data Programs That Accept Nonprofits
At least one company data program names nonprofits among eligible sellers: Mode's home page asks for 30 or more employees "for all other businesses or nonprofits", read November 16, 2026. Mode states that it buys a filtered, de-identified copy, so Mode is a buyer and the nonprofit is the seller.
Mode's pages give three versions of eligibility: the home page asks for 30 or more employees, the referral page for 50 or more, and the agreement for 50 office-based employees.
Mode states that a company's data "could be worth $100K–$5M" and labels its calculator "Indicative only" (checked November 9, 2026). The full Mode program terms and eligibility set out each version with its date.
Listing on a Mission-Locked Platform or Joining a Collective
A nonprofit that wants to set its own price and terms can list a dataset on a platform, and one whose members create works can license through a collective. Mozilla Data Collective Limited (Companies House 17054959) describes itself as a "mission-locked British company, incubated by Mozilla Foundation".
Its Compensated Datasets launched on July 29, 2026, with providers setting price and license, and it states "Own your data and keep 100% of revenue" with no platform fee stated (checked November 9, 2026). It is a platform and not a buyer, and the page lists Mozilla Data Collective listing terms.
STIM describes itself as a nonprofit membership organization with an opt-in AI license since September 2025, and the RSL Collective describes itself as a "nonprofit rights organization" that does not publish its rates.
Mission, Donor and Funder Terms That Decide What a Nonprofit May License
Before any privacy rule applies, a nonprofit checks whether its mission, its gift and grant terms and its agreements allow a license, because data gathered for a charitable purpose may be bound by that purpose. This page states no rule on donor-restricted gifts, grant data clauses or charity regulators. Three checks apply:
- Mission and governing documents: the purposes the organization states.
- Donor and grant conditions: each gift agreement and grant.
- Data held for partners or sponsors: one buyer listing excludes "Data owned by a sponsor or partner".
Under 17 U.S.C. 201(b), the employer owns a work made for hire unless a signed written instrument provides otherwise, and who owns what volunteers create is not stated here.
Funding also shapes the decision: Common Crawl says that donations from AI companies are "a small fraction of our overall operating needs" (November 4, 2025). Gift agreements and grants are read in a rights audit before licensing.
Beneficiary, Member and Donor Privacy Before a Nonprofit Licenses Records
Records about the people a nonprofit serves often reveal health, religion, political opinion or similar traits, and in the EU such data needs an Article 9(2) exception on top of a legal basis before any reuse for AI training.
Regulation (EU) 2016/679, applicable since May 25, 2018, prohibits in Article 9(1) processing data that reveals, among others, ethnic origin, political opinions, religious beliefs or health unless an exception applies, as explained under special category data in AI training.
The Dutch Autoriteit Persoonsgegevens states in its guidance of July 13, 2026 that the model builder, as acquirer, must check whether the personal data was lawfully obtained and may be used for the new purposes. It calls informed and specific consent for training "vrijwel onmogelijk" (practically impossible). Three situations need checks:
- Special category data in the EU: records that reveal protected traits.
- Patient data at nonprofit health providers: a covered entity's sale of protected health information needs authorization under 45 CFR 164.502(a)(5)(ii), while de-identified data falls outside the Privacy Rule under 164.502(d).
- State privacy laws: whether a given state law covers nonprofits is not stated on this page, so the statute itself is read.
Whether reuse fits the purpose of a service record is a question of purpose compatibility when reusing data for AI training.
Tax Status Questions a Nonprofit Takes to the Primary Rules
Licensing income raises tax questions for a tax-exempt organization that this page does not answer, because the rules on unrelated business income and on royalties are not part of the record used here. Three questions go to the primary rules and a tax adviser:
- Does licensing income count as unrelated business income in the nonprofit's country?
- Does a commercial subsidiary, as in the Wikimedia route, change the analysis?
- Does the payer withhold tax or charge VAT on the license fee?
The general rules are set out under tax on data licensing income, VAT and withholding.
Research Status for Nonprofit Associations That Build Datasets
A nonprofit association that mines works for scientific research can rely on the EU research exception, and German law withdraws that status from organizations controlled by a private company with privileged access to results.
Directive (EU) 2019/790 Article 3 lets research organizations mine works for scientific research without an opt-out, and Article 7(1) protects it from contract override.
Section 60d of the German Copyright Act (UrhG) excludes organizations controlled by a private company with privileged access to results. LAION e.V. is the defendant in Kneschke v. LAION: the Higher Regional Court of Hamburg (5 U 104/24) dismissed the appeal on December 10, 2025, relying on section 44b and, alternatively, section 60d.
The Federal Court of Justice (I ZR 281/25) heard the case on September 3, 2026 and has scheduled its pronouncement for December 17, 2026, as of November 9, 2026. See the Kneschke v. LAION rulings and status. The university side sits under the related seller types below.
Licensing or Not: Benefits, Risks, Downsides and Alternatives for Nonprofits
Licensing can fund a nonprofit's work from data it already holds, but it can conflict with its mission, its donors' terms and the privacy of the people it serves.
Benefits
- major AI developers take access from a nonprofit's subsidiary, as with Wikimedia, LLC
- a company program names nonprofits among eligible sellers
- a platform listing lets the provider set its own price and license
Risks
- beneficiary data falls under special category and purpose rules
- donor and grant terms may bar a license
- commercial ties can cost an association its research status under section 60d
Downsides
- no public price benchmark exists for nonprofit AI licenses
- an open dataset cannot be recalled from copies already made: LAION-5B was taken down on 2023-12-19 and re-released as Re-LAION-5B on 2024-08-30
- the tax questions stay unresolved on this page
Alternatives
- open release under a public license
- a collective license for members' works
- listing with its own terms
- not licensing at all
The general decision is set out under should you sell or license your data.
How Nonprofits Differ From Operating Companies and Other Seller Types
A nonprofit licenses like an operating company where it holds operational records, but its mission, donors and the people it serves add limits a trading business does not have. Process, record families and core-business risks are covered under operating companies licensing their data.
Nonprofits also acquire data: TTAM Research Institute, a California nonprofit public benefit corporation, bought substantially all 23andMe assets for USD 305.0 million in cash, closed on July 14, 2025, in a bankruptcy asset sale and not an AI training license. Related seller types are:
- universities and research groups licensing data
- libraries, archives and museums licensing collections
- governments and open data
- collective licensing for AI training
- platforms licensing user content
- news publishers licensing content to AI
Related Deal, Dataset, Law and Platform Pages
The routes and rules on this page come from the organizations' own pages, filings, statutes, a regulator's guidance and court records, each dated, and this page is general information and not legal advice. Further pages:
- LAION-5B and Re-LAION-5B contents and license
- HIPAA de-identification and sale of PHI
- PII removal and de-identification before licensing
- alternatives to selling data, including not selling
- Associated Press AI licensing deals
- reputational risk of licensing data
- privacy risk to people in licensed data
- openly licensed material in AI training
- Dolma dataset
Frequently Asked Questions
Can a nonprofit license its data to AI companies?
Yes, where it holds the rights and its mission, gift terms and privacy law allow it; the Wikimedia Foundation does so through a subsidiary and AP licensed part of its archive. Neither announcement discloses a value.
Do AI company data programs accept nonprofits?
Some say so: Mode's home page names nonprofits with 30 or more employees, read November 16, 2026, while its referral page and agreement give 50. Each program sets its own terms.
Does licensing data affect a nonprofit's tax-exempt status?
It can raise unrelated business income and withholding questions, and the answer depends on the tax rules of the nonprofit's country, which this page does not state. The primary rules and a tax adviser settle it.
How much do AI companies pay nonprofits for data?
No nonprofit AI license value is on public record: Wikimedia's and AP's announcements disclose none. Mode's range is a program claim and not a payment. How buyers value data is explained under what company data is worth to AI companies.