Withholding tax on cross-border data license payments

This page is general information, not legal advice. Confirm requirements with counsel for your jurisdiction and use case.

It is written for the finance or procurement lead who pays the licensor. Suppliers asking whether their own license income is taxable should read is licensing data to AI taxable instead.

Why characterization decides whether withholding applies

Withholding regimes key off the type of income, so the first question is what the buyer is paying for, not what the contract is called. Most domestic laws and the royalty articles of tax treaties treat payments "for the use of, or the right to use" copyright or confidential information (know-how) as royalties. Payments for services performed, or for buying a copy outright with no right to exploit it, are often treated differently and may fall outside withholding entirely.

An AI training data license tends to sit on the royalty side because the buyer obtains a right to reproduce records, train on them and keep derived model weights. That conclusion is not automatic. A one-time purchase of a static dataset with no reproduction or exploitation rights beyond internal use can be argued as a purchase, and many countries draw that line differently. US tax practice, for example, distinguishes a transfer of copyright rights from the sale of a copyrighted article in its software rules, and advisers sometimes reason by analogy for datasets.

The grant clause drives the analysis. A broad AI training rights grant that lets you reproduce, modify, train on and create derivatives is strong evidence of a royalty. A narrow fine-tuning-only license still conveys use rights, so do not assume narrowness removes royalty treatment.

Two directions of payment, two sets of rules

The withholding obligation sits with the payer under the payer's own country's law, so a non-US buyer paying a US licensor applies its home rules, not US rules. If your company is resident in, say, an EU member state, India or Korea, your local tax code decides whether outbound royalties carry withholding and at what domestic rate. The US licensor then claims treaty relief from your tax authority, not from the IRS.

The reverse direction is governed by US law. When a US payer pays a foreign licensor US-source royalties, the Internal Revenue Code generally requires the payer to withhold under sections 1441 or 1442 unless a treaty reduces the rate. US royalty withholding is computed on the gross payment, before any deduction for the licensor's costs. A US-based affiliate of your group paying a foreign supplier is therefore a withholding agent with its own obligations.

Many AI teams operate both ways at once: a non-US parent licenses from US suppliers while a US subsidiary licenses from European or Asian suppliers. Map each payment flow (payer entity, payee entity, residence of each, source of the income) before the first invoice arrives.

Treaty relief and the residency paperwork behind it

Tax treaties frequently cap or eliminate source-country withholding on royalties, but the payer can usually apply the reduced rate only if it holds proof of the licensor's treaty residence before payment. The exact proof is set by the payer's country. Some accept a self-certification form, others require a residency certificate from the licensor's own tax authority, and some require pre-approval from the local tax office.

For a US licensor, the standard proof is IRS Form 6166, a letter certifying US residency for income tax purposes. Many treaty partners ask for this certificate before granting treaty benefits. The licensor requests it from the IRS on Form 8802, so build that lead time into your payment calendar. Because the certificate covers a stated tax year, plan to request a fresh one for multi-year or recurring purchases.

For a foreign licensor paid by a US entity, the document is Form W-8BEN-E (W-8BEN for individuals). A US withholding agent relies on a properly completed form to treat the payment as going to a foreign beneficial owner and, where a treaty applies, to use a reduced rate or exemption. Collect it before the payment is made or credited; check the current form and instructions on irs.gov.

Treaty claims also depend on beneficial ownership and limitation-on-benefits tests. A licensor that is a conduit or a holding company passing fees to an affiliate in a non-treaty country can lose treaty access, so ask who actually owns the licensed rights.

Tax document request checklist for a data license payment

A short checklist collected at contracting, before the first invoice, prevents most over-withholding and refund claims. Use it as a starting point and let your adviser add country-specific items.

Illustrative example: invented to show structure; it does not describe an available dataset.

Gross-up, tax cooperation and invoice structure in the license

Who bears withholding is a commercial term, so the license should say it expressly instead of leaving the default, which usually means the licensor receives the net amount. Licensors in the US often ask for a gross-up clause that obliges the buyer to increase the payment so the licensor nets the invoiced figure. Buyers typically resist a full gross-up because it shifts the cost of the licensor's paperwork failures onto them.

Common buyer fallback positions include:

• Gross-up capped at the treaty rate, so the buyer does not bear extra withholding caused by a late or missing residency certificate.
• No gross-up where the licensor failed to supply forms, and a duty to cooperate on refund claims.
• A covenant that the buyer will provide official withholding receipts promptly so the licensor can claim a foreign tax credit.
• A narrow audit right tied to royalty calculation only; licensors use audit clauses to check compliance and that the correct royalties were paid [1], and usage-based fees make that right more likely.
• Separate line items for license fees and for genuine services such as annotation, de-identification or delivery engineering, priced at arm's length.

Splitting invoices is legitimate only when the components are real and separately priced. Relabeling a license fee as "data services" to avoid withholding invites reclassification, interest and penalties. Fold these terms into your data license negotiation checklist and the tax section of your data license term sheet.

How pricing structure changes the tax picture

Payment structure affects when withholding arises and how hard it is to compute, even though it rarely changes the characterization. A flat upfront fee creates one withholding event that is easy to document. Per-record, per-token or revenue-share pricing creates recurring payments, and each one needs a valid residency certificate on file for the period paid.

Revenue-share or usage-based fees are the strongest royalty signal because payment varies with exploitation of the licensed rights. If you are comparing models, read data license pricing structures compared alongside your adviser's view on recurring compliance cost. Ongoing purchases under a master agreement also need a calendar for certificate renewals.

Indirect taxes are a separate question

Withholding on income is not the only cross-border tax on a data license, and indirect taxes run on different rules. Many countries apply VAT or GST to imported digital services or intangibles, often through a reverse charge that the buyer self-assesses. A payment can be exempt from withholding and still create a reverse-charge VAT entry, or the reverse.

Some jurisdictions also levy digital services taxes or equalization levies that may reach certain cross-border digital payments. Ask your adviser to cover income withholding, indirect tax and any digital levies together, because the same contract language (what is supplied, where it is used) drives all three.

Where SourceX fits for buyers outside the US

SourceX sources operational datasets from US companies on request and manages the commercial process, including licensing agreements and ongoing purchases, for AI teams wherever they are based. Every dataset is rights-reviewed for ownership and consents and delivered under a license that defines the records, uses, term and delivery; share that grant with your tax adviser. Diligence materials covering source, rights, preparation and allowed use are prepared per dataset. Non-US buyers can read licensing data to a non-US lab for the supplier-side view, then describe the data you need on the buyers page.

Get licensed operational data with a clear rights grant

SourceX sources operational datasets from US companies on request and manages licensing and ongoing purchases; nothing is contracted until a supplier agrees, and SourceX does not publish prices, so terms are agreed per deal. Start in the AI data licensing guide or the AI data hub, then tell us what data your team needs.