Which AI Data Buyers Take Holders Outside the US?

Most AI data buyers do not shut out a company based outside the United States, but few say they want one, and the buyers that publish a price pay a company abroad much less. Of the 43 buyers and labs' programmes whose own pages we read on 18 October 2026, 13 take holders from anywhere or already work with holders abroad, 2 take holders only from a listed set of countries, 2 take holders abroad but rank the US first, 1 takes US holders only, and 25 say nothing about where a holder can be based. Among the 24 that buy a company's own records (support tickets, email, chat, documents, code), the split is 5 open, 1 on a closed list, 2 US first, 1 US only and 15 silent. Money already crosses the border: of 78 disclosed licensing deals whose licensor we could place, 32 were with licensors outside the US, in 13 countries.

The finding that matters most is price. The buyers' own estimators value the same company at 0.71 times its US figure when it is based in Europe, the UK or Canada, and at 0.27 times anywhere else, before anyone has looked at a single record.

This guide is for the founder, CFO or counsel of a company outside the US that holds data an AI buyer might want. The live version of every rule below is in the country register.

This guide is research from public sources, not legal, tax or investment advice.

Contents

1. The short answer, counted
2. Buyers of company records, one by one
3. What the estimators do with your headquarters
4. English, or any language
5. Marketplaces and the labs' own programmes
6. Buyers and brokers based outside the US
7. What "says nothing" means in practice
8. Worked example: a Canadian and an Indian company
9. Questions people ask
10. What this means for you

The short answer, counted

Ask a buyer "do you take companies outside the US?" and there are three honest answers: yes in writing, no in writing, and nothing in writing. Most give the third. So we read what each buyer publishes for holders (its programme page, FAQ, forms, calculator, payout help and terms) and recorded every sentence on where a holder may be based, language, payment, tax forms and sanctions, word for word. The verdict for each buyer is our reading of all its rules together.

A buyer's page is not its contract. Almost none publishes the agreement a holder signs, so a silent page may hide a rule, and "anywhere" may be narrowed by the contract. Our buyer board keeps the dated reading.

The "open" row mixes buyers that say "anywhere" with buyers that show holders abroad by what they do: Ooak runs a Paris entity that keeps EU partners' data in the EU, and Troveo's content side names licensors in Vietnam, the Netherlands, India and South Africa. Only one buyer of company records ties its rule to the US alone, and through payroll, not country: Nyne's programme wants at least 20 full-time U.S. W-2 employees, and it does not say whether a company abroad with that US payroll would qualify.

The deals already done with licensors abroad

The deal ledger shows money crossing the border often. The Financial Times licensed its journalism to OpenAI for training, GEDI licensed Italian-language content from La Repubblica and La Stampa to OpenAI, and Grupo Folha and Grupo UOL in Brazil signed a deal with OpenAI whose values are withheld. Meta signed Le Figaro and Prisa; Microsoft signed Nine Entertainment in Australia; Protege, a US broker, placed GrabMaps' map data from Singapore.

Two things matter for a company holder. Almost every licensor here is a publisher, broadcaster or music company, so the deals prove that US buyers pay licensors abroad, not that they pay a company abroad for its tickets and email at the US rate. And terms are rarely stated: every deal named above is recorded as not disclosed.

Buyers of company records, one by one

The buyers that take a company's own records are the ones a holder meets first, because they run public programmes with forms, estimators and referral rewards. They also publish the strictest rules about where and in what language. The first table covers buyers whose pages say something about where a holder can be based; the second, those that say nothing.

Every cell links to the buyer's own words. "Not published" means the pages say nothing on the point, which is different from "no".

The buyers that say something about where

Three deserve a closer look. Turing is the only records buyer naming the legal reason a country can be a problem: "Some jurisdictions carry data-export restrictions we'll flag early." Its English rule covers code ( pull requests included), not documents or chat. Mode's closed list sits on its referral pages, yet its calculator offers USA, Canada, UK, Australia, EU and Other and its referral reward has a separate rate outside the US, so its company-data page states a preference rather than a ban. micro1's track for companies that have shut down lists ten countries, India and Singapore among them, in its own posting.

The buyers that say nothing about where

The silent buyers include some of the largest. That is not a refusal: the answer, with the currency, tax form and governing law, arrives in the buyer's own agreement after you have scoped your data with it.

What the estimators do with your headquarters

This is the most important finding, so plainly: four buyers publish an estimator that asks where your company is based, and each cuts the estimate for a company outside the US, whatever its data. Two companies with the same head count, years of records and tools get different numbers because one answered "Europe" or "Other".

Handshake shows it most clearly, because its calculator's settings are readable in the page's own script file. Its four country answers carry editor's notes that read "Prices at 1.00", "0.75", "0.75" and "0.30" for USA, Canada, Europe and Other. A second note says the calculator shows a floor of $100,000 in place of a range, outside the US only, when the lower estimate falls below it. Visitors see only the result: for 20 people and three years, $100,000 to $296,551 in the US and "Up to $100,000" for the same company elsewhere.

The table sets the four estimators side by side, at the inputs where we hold a reading in each region.

The last row is our valuation's reading: every pair of one estimator's outputs that differ only in location, averaged within each estimator, then across them. micro1 and Handshake agree because they run the same formula. Nyne cuts deeper, while its programme text asks for US payroll staff. Mode puts Australia in its middle group; micro1 and Handshake can only call it "Other".

What the discount is, and what it is not

The discount is a location factor: one multiplier on the whole estimate. It is not a judgment of your records, which the estimator has not seen, and it is not a price, since every estimator calls its output indicative and every buyer prices only after reviewing a sample. What it sets is the anchor, the first number a holder hears, and first numbers are sticky. A Canadian company opening with Handshake's calculator starts at three quarters of its US twin; an Indian or Brazilian company starts below a third. Our negotiation page treats this as a lever, the headquarters discount.

No estimator says why it applies the factor, but the buyers' own words point to three reasons. Demand: micro1's is "strongest for companies based in the United States, United Kingdom, and Canada". Language: micro1 asks for English records and contacts. Law and export risk: Turing flags "data-export restrictions", and a buyer that must work through your privacy law and a cross-border transfer prices that work in; the GDPR guide and the privacy-law-by-country guide set it out by country.

Listed speech prices show the same pattern: price follows where the speakers are, not their language. One vendor lists conversational English from the US at $90 an hour and from South Africa at $60. Not every vendor does: one lists customer-service calls in English and in Hindi at the same $24,500 for 1,000 hours. These are list prices with no sale behind them.

How a holder abroad answers the discount

Treat the factor as an assumption to test. Is it about headquarters, or where the records were made? A Canadian subsidiary of a US group may hold records that look exactly like a US company's. Is language a real cost? If your records are in English, show it in a datasheet. Is the transfer solved? A holder with its rights file settled and its de-identification done (see our de-identification standard) has removed most of the legal work a buyer might be pricing. Answer all three, put one datasheet before several buyers, and you are no longer negotiating against a multiplier on a form.

English, or any language

Language is the second filter, and the buyers split cleanly. The buyers of company records that set a language rule want English; those that want other languages mostly buy speech, text and video.

Two want English. micro1 is "primarily looking for companies that operate in English", with documentation, workflows and knowledge bases in English and representatives who can deal with it in English throughout. Turing wants "English-language comments & issues" in codebases and sets no rule for documents or chat. Every other buyer of company records publishes nothing on language, which in practice means a sample in another language is judged by reviewers who may not read it.

These want other languages:

Read two of these closely before spending time on them. Meta's one named partner is the Government of Nunavut, sharing Inuit-language data; it offers no payment and its interest form is closed. OpenAI works with partners "eager to represent data from their country or industry", naming Iceland's government, but the archived copy we could read says nothing about a fee.

Much non-English data is also produced for free: Google funds Project Vaani, open Indian-language speech, and Cohere's Aya datasets are openly licensed. That caps what commodity text or read speech can fetch, and leaves room for real conversations and linked records. No buyer we read publishes a price, or a multiplier, for company records in a language other than English.

Marketplaces and the labs' own programmes

Marketplaces and labs' programmes publish country rules more often, because a marketplace must pay holders in many places.

On a marketplace the holder usually sets the price, so its country rule decides only whether you may list and how you are paid. A lab's programme first decides whether it wants your data.

AWS Data Exchange publishes the fullest mechanics for a provider abroad. Its list of eligible places leaves out Canada, India, Brazil and Singapore; AWS's general page on who may sell adds Colombia, India and South Korea; an older FAQ names only the US and EU. A provider without a US account can use a virtual US account from Hyperwallet. Ask AWS which list applies before you prepare a listing.

Datarade and Opendatabay take registered businesses anywhere, and both are based outside the US. Datarade keeps 30% of a sale, less on a paid plan and forbids selling data to sanctioned countries. Opendatabay lets the holder keep 70% of a small sale, rising to 95% of a large one, and its data rights terms make the holder warrant a lawful basis that covers "transfer outside the country of collection": a holder in the EU or UK must settle its transfer question before it lists.

The labs' programmes are mostly for publishers and language data. Google's Content Offer Pilot reaches further: it takes large datasets in specialised segments such as code, agrees a fee with each partner and licenses non-exclusively. Microsoft's marketplace, by its own product page, "is for Grounding and Display use cases, not training". Protege names a Cambodian broadcaster and an EMEA broadcaster that earned over $1M in 6 months among its partners.

Buyers and brokers based outside the US

A holder abroad does not have to deal with a US buyer at all.

These buyers, brokers and public programmes based outside the US take data from holders; the list holds only those whose own pages we read.

The public programmes are slow and publish structures rather than amounts, and some run the other way: Korea's government AI data hub requires a separate agreement before a foreign company may use the data it funded. For company records, the marketplaces and Appen are the practical routes here; for language data in Europe, GPT-NL and the Language Data Space are worth watching.

What "says nothing" means in practice

15 of the 24 buyers of company records say nothing about where a holder can be based. Read that as "not decided in public", never as "yes".

The agreement that would answer is rarely published; micro1's data licence master terms, one of the few, make the holder warrant compliance with the laws on "cross-border data transfers" that apply to it. That is the usual shape: the page is silent on whether it takes you, and the contract puts the cross-border risk on you. Silence has four practical consequences.

1. You learn the answer late, after scoping, an NDA call and a sample, from a buyer that may then decline or apply a factor it never mentioned.
2. The currency and payment rail are open. Only AWS Data Exchange ties payment to a bank account that can accept US dollars. The rail matters: a US platform paying through Stripe Connect reaches accounts only in the US, UK, EEA, Canada or Switzerland.
3. Tax is withheld unless you document yourself. A US buyer paying a royalty to a company abroad must withhold 30 percent unless the holder claims a treaty rate on Form W-8BEN-E, delivered before the payment is made. Few buyers name the form; Troveo points international licensors to the W-8 forms. The withholding guide gives each country's treaty rate.
4. Sanctions are checked under US law, mostly through general terms: Mercor's users confirm they are not on a US sanctions list or in a sanctioned country, and Prime Intellect names China, Iran, North Korea, Syria, Cuba and Crimea. A holder under US sanctions cannot be paid, whatever a page says about "anywhere".

The questions to send a buyer, in writing

Send these before any sample leaves your systems, and keep the answers.

Worked example: a Canadian and an Indian company

Example. Two software services companies, each with 40 people and eight years of records in Slack, Jira, GitHub, Google Workspace and HubSpot. One is in Toronto and works in English. The other is in Bengaluru; its documents, tickets and code are in English, some of its chat in Hindi and Kannada. Each approaches micro1, Handshake and Mode.

Will they be taken? At micro1 the Canadian company is in the favoured group, since demand is "strongest for companies based in the United States, United Kingdom, and Canada"; the Indian company is in "other regions" that "may still be considered", and must show it operates in English, which its documents do and part of its chat does not. At Handshake neither is excluded: the form asks USA, Canada, Europe or Other and states no rule. At Mode the Canadian company is on the referral list of four countries and the Indian one is not, though Mode's calculator still offers it "Other".

What will the estimators show? Our regional readings are at a larger size, so take the ratios. Handshake prices the Canadian company at 0.75 of its US twin and the Indian one at 0.30, and at this size the Indian estimate may fall under Handshake's floor for estimates outside the US and show only "up to" a figure; micro1 applies the same weights. Mode's readings run from $290K to $570K in the US to $210K to $430K for Canada and $90K to $170K for "Other". Same kind of records; the Indian company starts from an anchor under half the Canadian one's, at every buyer.

What arrives with the money? If the deal is a royalty, the Canadian company can claim a 0 percent treaty rate on know-how or copyright; the Indian company's rate is 15 percent on know-how and copyrights. Both deliver a W-8BEN-E before the first payment. A buyer paying through Stripe Connect can reach the Canadian account and not the Indian one. None of the three publishes its currency. Each company's own privacy law also applies to personal data in its records; see the Canada and India pages.

What should each do? The Canadian company is close to a US company here: it should treat the factor as negotiable and put one datasheet before every fitting buyer. The Indian company should not let three estimators define its value. It should add buyers whose pages say "anywhere" ( SuperAnnotate, Turing) and marketplaces open to any registered business ( Datarade, Opendatabay); AWS Data Exchange's provider list leaves India out. Both should get a valuation that prices their records against sales and offers before they hear a buyer's number.

Questions people ask

Do US AI companies buy data from companies in other countries?

Yes. 32 of the 78 disclosed deals whose licensor we could place were with licensors outside the US, such as the Financial Times with OpenAI, almost all publishers and media companies. For company records, most buyers take holders abroad or say nothing on the point.

Does micro1 accept companies outside the US?

Its page says the programme "accepts referrals globally" but prioritises US companies, followed by other Western markets, and wants records and contacts in English. Its estimator prices a company outside the US, Canada and Europe well below the same company in the US.

Will I be paid less because my company is not in the US?

At the estimator stage, almost certainly: the estimators average 0.71 times the US figure for Europe, the UK and Canada and 0.27 times elsewhere. That is an anchor set before the buyer sees your records, not a price, and English records, a settled rights file and competing interest can move it.

Do AI data buyers accept data in languages other than English?

Some want it: OpenAI asks for data "across any language" and Protege covers 70+ languages in speech. But the buyers of company records that set a language rule, micro1 and Turing, want English, and no buyer publishes a price for non-English company records.

Can a company in India, Brazil or Singapore sell data to AI companies?

Most buyers' pages do not stop it, and some say "anywhere" ( SuperAnnotate) or show partners there ( Troveo, Protege). Mode's referral list and AWS Data Exchange's provider list leave them out, and the estimators price them lowest. Each country page lists the buyers that take a holder there.

Do I need a US bank account or a W-8BEN-E?

A company abroad paid a royalty gives a W-8BEN-E before payment to claim a treaty rate. A US bank account is needed only by some programmes: AWS Data Exchange asks for US bank information and offers a virtual US account.

What this means for you

A company outside the US can sell or license its data to AI buyers, and many will talk to it.

What changes is the starting price, the language a buyer will read, and the paperwork around payment, tax and transfer, almost none of it written down in advance.

1. See which buyers take a company where you are. Your country page ( United Kingdom, Germany, Canada, India and the rest under /countries) lists each buyer's answer, your withholding rate and your privacy work; every buyer's own rules sit on one table, and the buyer board has their other terms.
2. Run the match. The buyer match tool checks your size, region, language and status against each buyer's published rules.
3. Get a reading of value that is not a buyer's anchor. The valuation prices your records against sales and offers.
4. Settle your own law first, with the GDPR guide or the privacy-law-by-country guide, and the withholding guide for tax.
5. Send the questions above in writing, and compare the answers with the programmes compared before you sign.

For the whole picture, read the hub guide on selling data to US AI companies from outside the US. Where we stand: we run sales only for holders based in the United States, the United Kingdom and Canada today; our valuation works for a holder anywhere and binds no one.