DataPacts

Selling Data to AI Companies From Outside the US

By DataPacts Editorial Team · Checked

A company outside the US can license data to AI buyers, but most published programs are run by US companies, so it meets four extra questions: whether the buyer takes non-US sellers, whether its home law lets the data leave, which law and currency the contract uses, and how US tax forms apply.

Programs state their own geography. micro1 states that it prioritizes US companies, followed by other Western markets, while 18 of the 23 Mercor listings that state a location say "Anywhere", both checked November 9, 2026. This page sets those statements beside home-country export rules, contract forum, payment and withholding, all part of AI training data licensing.

DataPacts Editorial Team: how we source, date and check every figure.

What Changes When the Seller Is Outside the US

Four layers change for a non-US seller, and the process steps (inventory, rights, de-identification, contract) stay the same. The table compares the four layers with one example each, as of 2026-11-09.

Layer Question for a non-US seller Example on this page Source
Buyer geography Does the program take companies from my country? micro1 prioritizes US companies micro1.ai, checked 2026-11-09
Home-country export rules May personal data leave my country for the US? GDPR Arts. 45, 46 and 49 Regulation (EU) 2016/679
Contract forum and currency Which law, which forum, which currency? micro1: California law, JAMS arbitration in California micro1 master terms, 2026-11-08
Payment and US tax forms What will the payer withhold and which form stops it? 30 % under 26 U.S.C. 1442(a) unless a treaty reduces it uscode.house.gov

Buyer Programs and Marketplaces: Where Each States It Takes Non-US Companies

Each program states its own geography, often in different words on different pages, so the wording and the check date matter more than any summary. The table compares seven programs and marketplaces, alphabetical, with the geography each states itself, as of 2026-11-09.

Program or marketplace Type Geography as stated Where stated Checked Source
AWS Data Exchange marketplace (seller stays licensor) provider guide: 13 entries (AU, BH, EU member states, HK SAR, IL, JP, NZ, NO, QA, CH, AE, UK, US); FAQ: "US or EU member state" provider guide; FAQ 2026-11-09 docs.aws.amazon.com; aws.amazon.com
Handshake AI company data program no country rule; estimator asks USA, Canada, Europe, Other data partnerships page 2026-11-09 joinhandshake.com
Mercor company data program (listings) 18 of 23 listings that state a location: "Anywhere"; accounting and tax: United States listings feed 2026-11-09 api.data.mercor.com
micro1 company data program US companies first, then other Western markets; data licensed to US and allied-country buyers data partnerships and payout pages 2026-11-09 micro1.ai
Mode company data program home: US strongest fit, others case by case; referral page: US, Canada, UK, Australia; agreement: US, EU, Canada, Australia three pages 2026-11-09 (home re-read 2026-11-10) data.mode.inc
Nyne company data program 20+ full-time U.S. W-2 employees; selector USA, Canada, Europe, Other program page 2026-11-09 nyne.ai
Snowflake Marketplace marketplace (seller stays licensor) paid listings for billing addresses in 20 listed countries; USD payouts need a US billing address provider docs 2026-11-09 docs.snowflake.com

Programs That Put US Companies First or Test for US Staff

micro1 states it prioritizes US companies, then other Western markets, and licenses data only to labs and enterprises in the US and allied countries, checked November 9, 2026. That second statement restricts who receives the data; the master-terms grant itself is nonexclusive. Nyne states 20+ full-time U.S. W-2 employees, yet offers a country selector with USA, Canada, Europe and Other.

Mode states its geography three ways. Its home page says the US is the strongest fit and others are case by case (re-read November 10, 2026), its referral page lists the US, Canada, the UK and Australia, and its referral agreement lists the US, the EU, Canada and Australia. Handshake AI's estimator asks for the country but states no country rule, checked November 9, 2026.

Listings and Marketplaces Open to Other Countries

Mercor's data listings carry a location field, and 18 of the 23 listings that state one say "Anywhere", checked November 9, 2026. The buyer match tool shows published acceptance criteria by company size, country, status and data type; this page covers only the geography wording.

Mercor's Law Firm and Imaging & Diagnostics Center listings say "United States preferred", and Accounting & Tax Firm says "United States". AWS Data Exchange publishes 13 jurisdiction entries in its provider guide but "US or EU member state" in its FAQ, and the two statements are not reconciled here. Snowflake lists 20 billing countries for paid listings. Both are marketplaces where the seller stays the licensor.

Export Rules for Personal Data in the Seller's Home Country

Records about people need a lawful route out of the seller's country before delivery to a US buyer, and de-identified data that is no longer personal data needs none. The summaries below are information, not legal advice.

EU and EEA Sellers: Chapter V, the DPF List and SCCs

An EU or EEA seller may deliver personal data to a US buyer under one of three GDPR routes: an adequacy decision, appropriate safeguards or a narrow derogation (Regulation (EU) 2016/679, Arts. 45(1), 46(1) and 49(1)). Legal basis and notices follow the GDPR rules for AI training data.

The EU-US Data Privacy Framework (Decision (EU) 2023/1795 of 2023-07-10) covers only US organizations on the DPF List. The General Court upheld it in case T-553/23 on 2025-09-03, and appeal C-703/25 P is pending. Standard contractual clauses (Decision (EU) 2021/914 of 2021-06-04) are modular, and the controller-to-controller module fits a data sale.

Swiss and UK Sellers

Swiss personal data may go to US organizations certified under the Swiss-US Data Privacy Framework without further safeguards, under the US entry of Annex 1 to the Data Protection Ordinance, in force since 2024-09-15. UK sellers fall under the UK GDPR as amended by the Data (Use and Access) Act 2025 (c. 18), which received Royal Assent on 2025-06-19.

Canadian Sellers: PIPEDA and Québec

Canada's PIPEDA business-transaction exception does not cover a deal whose primary purpose or result is the sale or lease of personal information (s.7.2(4)), so a data license generally needs consent or de-identified data. The wider rules for licensing data for AI training in Canada sit on their own page.

Bill C-36 had its first reading on 2026-06-15 and is at second reading as of 2026-11-09, so it is not law. Québec has required, since September 2023, a privacy impact assessment and a written agreement before personal information is communicated outside Québec.

Sellers in China, Taiwan and Other Jurisdictions With Transfer Controls

China's PIPL, in force since 2021-11-01, requires separate consent and a transfer mechanism before personal information leaves China, on top of notice and separate consent for the sale itself.

Taiwan's PDPA art. 21 lets the authority restrict cross-border transfers, Türkiye's KVKK (Law No. 6698) Art. 9 regulates transfers, and Saudi Arabia's PDPL restricts transfers abroad. For each country's rules, see AI training data law by jurisdiction.

When a Home Regulator Reviewed a Cross-Border Archive Deal

Italy's data protection authority, the Garante, formally warned GEDI on 2024-11-27 that sharing its news archive with OpenAI under their 2024-09-24 agreement could breach the GDPR. The warning (provvedimento n. 741) is an avvertimento under Art. 58(2)(a), and it cites possible breaches of Arts. 9, 10, 13 and 14 and of Chapter III.

The Garante found that the archives hold special-category and judicial data, and that the data protection impact assessment did not adequately analyze the legal basis. The findings show that a home authority can review a transfer to a US buyer on the seller's side of the deal.

Contract Terms a Non-US Seller Signs: US Law, US Forum, US Dollars

Published buyer terms are written under US law with a US forum and US dollar amounts, so a non-US seller litigates or arbitrates abroad unless it negotiates otherwise. The clauses are covered in AI data licensing agreements, and micro1's master terms, last updated 2026-11-08, choose California law with JAMS arbitration in California after a 30-day negotiation.

micro1 states its payout tiers in US dollars (USD 100K to 500K, 500K to 3M and 3M+, checked November 9, 2026), and the recipient restriction is also a contract term.

Getting Paid From the US: Bank Accounts, Tax Forms and US Withholding

Payment routes assume a US bank or a US-style account more often than a local one, and the paying side asks a non-US seller for a W-8 form before it pays. AWS Data Exchange asks non-US providers for a W-8, a VAT or GST registration number, and US bank information or a Hyperwallet virtual US account, while Snowflake pays USD only to a US billing address (both checked November 9, 2026). Mercor states it pays by wire after the anonymized dataset is delivered, typically 2 to 4 weeks after signing, checked November 9, 2026.

US-source royalties paid to a foreign corporation face 30 % withholding under 26 U.S.C. 1442(a) unless a treaty reduces it. IRS Table 1 lists 0 % for the United Kingdom, Germany, Japan and Canada under income codes 10 and 12, while Croatia and Hungary have no treaty in effect and stay at 30 %. Form W-8BEN-E stays valid from signing to the last day of the third succeeding calendar year, absent a change in circumstances. Rates and forms by country sit in withholding on data license payments and treaty rates by country.

Non-US Companies in Disclosed AI Data Deals

Publishers and rights organizations in Europe have licensed content to AI companies and AI developers, most of them in the US, since 2023, and in every case below except one the value is not disclosed. The table lists nine deals by date, with the value as stated by a party, as of 2026-11-09.

Date Seller (country) Buyer Value as stated Source
2023-12-13 Axel Springer SE (DE) OpenAI not disclosed by the parties axelspringer.com
2024-03-13 Groupe Le Monde (FR) OpenAI not disclosed; multi-year lemonde.fr
2024-03-13 Prisa Media (ES) OpenAI not disclosed elpais.com
2024-04-29 The Financial Times Limited (GB) OpenAI not disclosed aboutus.ft.com
2024-09-24 GEDI Gruppo Editoriale (IT) OpenAI not disclosed; Garante warning 2024-11-27 garanteprivacy.it
2025-01-16 Agence France-Presse (FR) Mistral AI (FR) not disclosed; multi-year mistral.ai
2025-02-14 Guardian Media Group (GB) OpenAI not disclosed theguardian.com
2025-12-19 Kopinor (NO) National Library of Norway (NO) NOK 45,000,000 for 2026, indexed from 2027 kopinor.no
2026-10-12 Reach plc (GB) Google LLC not disclosed; "we pay for enhanced content rights" reachplc.com

Every deal with its stating source sits in the AI data licensing deals tracker.

Licensing to a Buyer in Your Own Country or Region

A domestic or regional buyer removes the US layers but brings its own, as two European deals and two buyer-side withholding rules show. Mistral AI and AFP, both French, announced a multi-year agreement for AFP stories in Le Chat on 2025-01-16, with amount and duration not disclosed. Kopinor and the National Library of Norway agreed on 2025-12-19 to a non-exclusive license for Norwegian and Sami language models, with annual remuneration of NOK 45,000,000 for 2026, indexed from 2027 (s.17).

A Canadian payer withholds 25 % Part XIII tax on royalties to non-residents (Income Tax Act s.212(1)(d)), and a UK payer withholds at the basic rate, 20 %, on IP royalties paid abroad (ITA 2007 s.906). One US rule runs the other way: 28 CFR part 202, in force 2025-04-08, binds US persons, so a US buyer's onward licensing of data may carry contractual bars on transfers to countries of concern.

Benefits, Risks, Downsides and Alternatives for a Non-US Seller, Including Not Selling

Licensing from outside the US opens the largest group of published programs and adds legal and payment steps that a domestic sale does not have.

Benefits

  • Mercor's listings mostly say "Anywhere", so geography is often not a barrier there.
  • Tiers stated in USD are easy to compare across countries.
  • A published home-country route (adequacy, DPF certification, SCCs) exists for the largest regimes.

Risks

  • A home regulator can review a cross-border archive deal, as the Garante did for GEDI.
  • The EU-US Data Privacy Framework faces a pending appeal, C-703/25 P.
  • A program's eligibility can differ between its own pages, as Mode's three versions show.

Downsides

  • Contract forum and law are often abroad, as in micro1's California terms.
  • US withholding is 30 % unless a treaty reduces it and a valid W-8 form is on file.
  • Some payment routes need a US bank account or a US billing address.

Alternatives

  • A domestic or regional buyer, as in the Kopinor and National Library of Norway agreement.
  • A marketplace where the seller stays the licensor, such as AWS Data Exchange or Snowflake Marketplace.
  • De-identified data only, which needs no personal-data export route.
  • Not selling at all, which is a valid outcome.

For the general decision, see should you sell or license your data.

Where Selling From Outside the US Sits in the Seller Process

Seller location is checked at step 1 of the seller process, before inventory and rights work, because it decides which buyers and which export routes are open. The full sequence to sell or license company data to AI companies continues from there.

Whether a buyer could train on the seller's published material without a license depends on the seller's copyright law, which changes what a license is worth to the buyer. See copyright and AI training across jurisdictions. Japan's Copyright Act Art. 30-4 has a proviso on uses that unreasonably prejudice the owner, covered in Japan's rules for licensing data for AI training.

Closed or Insolvent Companies Outside the US

Outside the US, an administrator, liquidator or trustee is the seller of a closed company's data, under the insolvency law of that country. For the UK, see selling company data in UK administration and liquidation.

Buyer and Marketplace Pages Cited on This Page

The geography statements above come from these programs' and marketplaces' own pages, checked November 9, 2026.

Frequently Asked Questions

Can a company outside the US sell data to AI companies?

Yes; several programs and marketplaces state they take non-US companies, and others put US companies first. Mercor's listings mostly say "Anywhere", while micro1 states it prioritizes US companies, both checked November 9, 2026.

Which AI data programs accept companies from Europe?

Mode's referral agreement names the EU, Mercor's listings mostly say "Anywhere", and AWS Data Exchange lists EU member states, each checked November 9, 2026. Nyne's selector includes Europe, but its page also states a US W-2 staff requirement.

Does a non-US seller need a US bank account?

Not always; AWS Data Exchange accepts US bank details or a Hyperwallet virtual US account from non-US providers, and Snowflake pays USD only to a US billing address. Both were checked November 9, 2026.