The National Library of Norway pays NOK 45,000,000 for 2026 under an agreement signed on 2025-12-19.
This page covers public bodies on the acquiring side and how a data holder reaches them, as of November 17, 2026. Public buyers are a small corner of AI training data licensing, and the record is thin: one priced license, two platforms and one fee schedule.
Three Roles a Public Body Can Play in Acquiring Training Data
A public body acquires training data in one of three roles on record: as a licensee that pays for data, as a platform operator that holds datasets for others to use, or as a member that pays a distributor for corpora.
- Licensee: the National Library of Norway, which pays a collective for a training license.
- Platform operator: Korea's AI Hub and India's AIKosh, which hold datasets for others to use.
- Paying member: US government members of the Linguistic Data Consortium, which pay a membership fee.
A fourth role, funder of a program that buys from private holders, has no priced record in the sources used here: no other priced purchase by a public body is recorded here as of November 17, 2026.
The National Library is a public institution, so its payment is not a payment by a commercial AI developer.
Public bodies far more often release their own data than buy any. The charging and exclusivity rules for that release belong to the page on governments as data sources, so how public bodies release their own data is covered under governments and open data as AI training data sources.
Public Institutions and National Programs on Record
The table lists five public institutions and programs on record in alphabetical order, each with its role and the money terms its own source publishes.
Five public institutions and programs on record, in alphabetical order, as of 2026-11-17 (dates differ in kind: check dates, signature and announcement dates):
| Institution or program | Country | Role | What it acquires or holds | Money terms published (stated by) | Date | Source |
|---|---|---|---|---|---|---|
| AI Hub | South Korea | government AI data platform (Ministry of Science and ICT, National Information Society Agency) | data in 14 fields (NIA) | not recorded | checked 2026-11-09 | aihub.or.kr |
| AIKosh | India | national datasets platform (IndiaAI) | 16,231+ datasets from 689 contributing organizations (IndiaAI) | not recorded | as of 2026-11-05 | aikosh.indiaai.gov.in |
| Creative Content Exchange | United Kingdom | government pilot marketplace for licensing, 12 cultural organizations | digitized cultural and creative assets | not published | announced 2026-01-26; status after summer 2026 not recorded | gov.uk Report on Copyright and AI |
| Linguistic Data Consortium membership | United States | distributor; US government bodies pay as members | speech, text and lexical corpora | USD 2,400 Standard, USD 3,850 Subscription for not-for-profit and US government members (LDC) | checked 2026-11-09 | ldc.upenn.edu |
| National Library of Norway (Nasjonalbiblioteket) | Norway | licensee | Norwegian press publications licensed from Kopinor for training Norwegian and Sami language models | NOK 45,000,000 for 2026, CPI-indexed from 2027 (agreement s.17) | signed 2025-12-19 | kopinor.no agreement PDF |
Programs often named in this context, such as the EU AI Factories, the UK AI Research Resource, the Netherlands' GPT-NL and India's Bhashini, are not listed because their data terms are not recorded in the sources used here. The same holds for national corpus programs in Japan and Korea.
The National Library of Norway: A State Library on the Paying Side
The National Library of Norway, a state library, pays Kopinor NOK 45,000,000 for 2026, index-adjusted from the 2027 remuneration, under an agreement signed on December 19, 2025 to build training corpora and train Norwegian and Sami language models. The license is non-exclusive, and the resulting models may be shared publicly under the license terms.
Kopinor, the Norwegian reproduction rights organization, licenses on behalf of its member organizations. Every clause is set out in the record of Kopinor and the National Library of Norway.
National Data Platforms: Korea's AI Hub and India's AIKosh
Korea's AI Hub and India's AIKosh are government platforms that hold training data for others to use, and neither publishes terms on which it buys data from private holders in the sources used here.
AI Hub is operated by the Ministry of Science and ICT and the National Information Society Agency, which states data in 14 fields, checked November 9, 2026. How that data is built, by whom and for what payment is not recorded in the sources used here.
IndiaAI states that AIKosh holds 16,231+ datasets from 689 contributing organizations, as of November 5, 2026. Payment terms for contributors are not recorded in the sources used here.
Government Research Users as Corpus Licensees
Public research bodies also acquire training corpora by paying a distributor, and the Linguistic Data Consortium publishes a fee band for not-for-profit and US government members. It states USD 2,400 for Standard and USD 3,850 for Subscription membership, checked November 9, 2026, without labeling the year.
Members hold perpetual rights to data licensed during their membership years, and providers reach the distributor through a submissions portal. Membership terms are in the Linguistic Data Consortium profile.
How Public Money Reaches Data Holders
In the one priced public license on record, money reaches data holders once a year through a collective, at a sum fixed in the agreement and indexed to consumer prices. Three terms of that agreement fix the flow:
- Remuneration falls due no earlier than June 30 each year.
- The total is NOK 45,000,000 for 2026, indexed by consumer prices from the 2027 remuneration.
- The agreement rolls yearly, with the first termination taking effect on January 1, 2028.
Public money depends on public budgets. On October 26, 2026, Kopinor stated that the 2027 state budget proposal holds no funding for its separate books agreement with the National Library. That is Kopinor's claim, and the terms of the books agreement are not recorded in the sources used here.
How a Data Holder Can Supply a Public Program
A data holder can reach a public program through four routes on record: a collective that licenses on its members' behalf, a national platform that takes contributions, a distributor whose members include public bodies, and a government pilot marketplace. The routes run from the priced record to the unpriced, as of November 17, 2026:
- Collective: Kopinor licenses for member organizations, among them Mediebedriftenes Landsforening. The agreement can extend to rights holders Kopinor does not represent if the Norwegian Industrial Property Office approves Kopinor under Copyright Act s.63 second paragraph, and rights holders may opt out with effect for future training.
- National platform: AIKosh lists 689 contributing organizations (IndiaAI, as of 2026-11-05), and terms for contributors are not recorded.
- Distributor: the Linguistic Data Consortium takes provider submissions through a portal, and payment to providers is not recorded.
- Pilot marketplace: the UK government announced the Creative Content Exchange pilot on January 26, 2026 with 12 cultural organizations and planned an operational pilot by summer 2026. Its status after summer 2026 is not recorded. It is a marketplace for licensing, not a purchaser.
None of these routes is ranked or recommended, and each figure is the party's own statement with its check date. Rights holders weighing the first route can read how collective licensing for AI training works for members.
What Changes When the Licensee Is a Public Body
A public licensee changes three things a holder would otherwise negotiate: what happens to the trained model, how long a withdrawal takes to bite, and whether the licensed material can be forced open later. In the Norwegian agreement of 2025-12-19, four terms show this:
- Models may be shared publicly under the license terms.
- A rights holder's opt-out works only for future training.
- Termination does not affect models already trained and shared.
- The license is non-exclusive and bars the training of specialised models that deliver media and news services.
In the EU, Directive (EU) 2019/1024 Recital 55 keeps documents in which third parties hold intellectual property rights outside the re-use rules.
For libraries, museums and archives, a document first owned by a third party with unexpired protection counts as third-party property. These terms come from one agreement, so they describe that record, not every public license.
Licensing Data to a Public Institution: Benefits, Risks, Downsides and Alternatives
For a data holder, a public licensee offers a fixed, published sum and a public purpose, and costs control over the model and dependence on a public budget.
Benefits
- A fixed annual sum indexed to consumer prices, as in the 2025 agreement.
- A published agreement with a dated payment rule.
- A collective can license for many members at once.
Risks
- Public funding can lapse with a budget, as Kopinor states for the books agreement.
- Models trained on the data may be shared publicly.
- Opting out does not remove data from models already trained.
Downsides
- Few priced public buyers are on record, and no count is available.
- Contribution terms for the national platforms are not published.
- An extended collective license can bind non-members only after approval under s.63.
Alternatives, including not licensing at all
- License to a commercial buyer through the routes on the parent page.
- Join a collective.
- Contribute openly to a national platform.
- Not license at all.
Not licensing is a valid outcome. The full answer to whether a holder should sell or license its data to AI companies sets out the same trade-offs for every buyer type.
Where Public Buyers Sit Among Other Buyer Types and Public Data Pages
Public institutions are one buyer category in the directory, and the pages below cover the other categories, public bodies as data holders and the national rules around them. The parent page lists who buys AI training data across every category.
- Libraries, archives and museums licensing collections (libraries as holders)
- Norway's rules for licensing data for AI training
- South Korea's rules for licensing data for AI training
- India's rules for licensing data for AI training
- Multilingual and low-resource language data
- Enterprises buying data to train their own models
- Dataset vendors and data labs
Institutions and Programs Named on This Page
The program figures above come from each institution's own pages and agreements, checked November 9, 2026, unless another date is given.
- AI Hub (Ministry of Science and ICT and National Information Society Agency)
- AIKosh (IndiaAI)
- Creative Content Exchange (UK government pilot)
- Kopinor
- Linguistic Data Consortium
- National Library of Norway (Nasjonalbiblioteket)
Frequently Asked Questions
Do governments buy AI training data?
Rarely on the public record: the clearest priced example is the National Library of Norway paying Kopinor NOK 45,000,000 for 2026 to train Norwegian and Sami language models. No other priced purchase by a public body is recorded here as of November 17, 2026.
How does Korea's AI Hub get its training data?
AI Hub, run by Korea's Ministry of Science and ICT and the National Information Society Agency, states that it holds data in 14 fields, but how that data is built and paid for is not recorded in the sources used here. The figure was checked on November 9, 2026.
Do national libraries pay for AI training data?
One does on record: Norway's state library pays a collective for press publications to train its own language models, which places a library on the paying side of a training license. The sum is NOK 45,000,000 for 2026.
Can a company supply data to a national AI program?
A company can supply through a collective, a national platform that takes contributions, or a distributor with public members, though none of the platforms publishes payment terms for contributors in the sources used here. The routes are listed above.