Key takeaways
- Licensing gives a buyer a copy of your records and limited rights to use it while your company keeps ownership; selling transfers ownership of the records themselves.
- Deals pay as a one-time flat fee, an annual license, a revenue share or per task, and flat fees are the most common.
- An exclusive license pays more because no other buyer can have the data, but it rules out every other sale for the term.
- Deletion removes the buyer's copy, not what a model learned from it, so anonymizing records before transfer and leaving sensitive ones out are what protect your company.
01
What is data licensing?
The short answer
Data licensing is a contract that lets a buyer use a copy of your data, for agreed purposes and for an agreed time, in exchange for payment, while your company keeps ownership of the data. Companies that license internal records to AI labs and data vendors are paid $100K to $2M or more per dataset.
A license is permission, not ownership
A license is permission to use something that someone else still owns. The owner, called the licensor, agrees to let the other party, the licensee, use it within agreed limits, and usually charges for that permission. Cornell's Legal Information Institute describes it this way: intellectual property owners who license "maintain ownership over their IP while profiting from a licensee's use."
Software, music and brand names are licensed the same way. You don't own Microsoft Word when you pay for it; you hold a license to use it on Microsoft's terms.
How a license applies to data
Data licensing is a license whose subject is a company's records. The buyer receives a copy of the data and a written list of what it may do with that copy. The original stays with you, and so does the right to keep using it.
A data license relies on its written contract more than a software or music license does, because copyright law protects far less of a company's records than it protects a program or a song. In the US, the Supreme Court held in Feist v. Rural Telephone (1991) that "facts, whether alone or as part of a compilation, are not original and therefore may not be copyrighted." So the limits on what a buyer may do with company records come mostly from what the agreement says, which is why each of its terms deserves close reading.
Who licenses company data today, and who pays for it
The fastest-growing kind of data licensing is companies licensing their internal records to the businesses building AI. Those records include help-desk tickets (the logged requests customers send to a support team), team chat, email, CRM history (the customer relationship management system where sales staff log every call and deal) and shared documents. Two kinds of buyers pay for them.
- AI labs
- Companies that build and train AI models, such as OpenAI, Anthropic and Google DeepMind. Large publishers and platforms often license to them directly.
- Data vendors
- Companies that license records from businesses, anonymize them (replace names, contact details and other identifying information with stand-ins) and turn them into training material for AI labs. Most companies license to a data vendor rather than directly to a lab.
- Training data
- The examples an AI model studies to learn a task. Company records are valuable training data because they show every step of a real job, such as a support ticket from the customer's first message to the fix, along with how it turned out.
Why those records are worth so much is its own subject, covered in why AI companies want your data. This guide covers the deal itself.
Data license, data license agreement, business data licence
The same arrangement goes by several names. A data license agreement (or data licensing agreement) is the contract document. A business data licence is the British spelling, used for the same thing when one business licenses its data to another. All three names mean the buyer gets permission to use the data while ownership stays with the company that created it.
02
Data licensing vs. selling data: what changes hands
The short answer
When you license data, the buyer gets a copy and limited rights to use it, and you keep ownership. When you sell data, ownership passes to the buyer and your company keeps no rights to it. Companies that are still in business almost always license, and outright sales mostly happen when a company shuts down and its assets are sold.
Who holds what, before and after
A license and a sale start from the same place: your company holds the records and owns them. The difference is what each side holds afterwards, shown in the diagram below. After a license, you still own the records and the buyer holds a copy it may only use as the contract allows. After a sale, the buyer owns the records outright.
The differences, side by side
| Data licensing | Selling data | |
|---|---|---|
| Ownership | Stays with you | Passes to the buyer |
| What the buyer gets | A copy, plus the rights the contract lists | The records and every right in them |
| Buyer's use | Limited to the permitted use | Any lawful use |
| License it again? | Yes, unless the license is exclusive | No |
| When it ends | At the end of the term, with the copy deleted | Never |
| Typical seller | Operating companies | Closed or bankrupt companies |
When a sale happens instead of a license
Outright sales of company data usually happen when a company stops operating and its assets are sold to pay creditors. A closed company has no further use for its records, so it loses nothing by giving up ownership of them.
Spirit Airlines is the clearest recent case. After the airline shut down in 2026, Google won a bankruptcy auction for its internal data with a $10 million bid, Skift reported, ahead of a $7.5 million backup offer from Mercor, a data vendor. micro1, the data vendor we typically refer companies to, then filed a late $12.5 million bid, Fortune reported, and the bankruptcy court still had to approve a buyer. The archive includes finance and operations data, revenue-management data (how the airline priced its seats) and about 100 million emails, all to be scrubbed of personal information before the buyer uses it.
Why an operating company should license, not sell
A running business still needs its records: its staff look up old tickets, its accountants reconcile old invoices, and its lawyers rely on old email. A license lets the company keep all of that and still be paid. A non-exclusive license also leaves the company free to license the same records to another buyer later, so the same records can be paid for more than once.
03
How data licensing deals are paid: four payment models
The short answer
Data licensing deals pay in one of four ways: a one-time flat fee for a copy of existing records, an annual license fee for ongoing access, a share of what the buyer earns each time it licenses the data onward, or a fee per completed task. The one-time flat fee is the most common for small and mid-size companies, and pays $100K to $2M or more per dataset.
The four models
Each payment model pays for a different kind of access. A one-time flat fee buys a copy of the records your company already has. The other three keep paying as the buyer goes on using the data or as your team creates new records.
Most common
One-time flat fee
The buyer pays once for an anonymized copy of the records you already have, such as five years of support tickets and team chat. The industry calls this copy a snapshot or corpus.
This model is simple to agree and quick to pay, but it doesn't pay for records you create after the deal.
Recurring
Annual license
The buyer pays every year for continuing access, often including new records as your business creates them. Publishers and platforms licensing large content libraries often use this model.
Reddit's Google deal was reported at about $60 million a year.
Recurring
Revenue share
The buyer, usually a data vendor, pays you a share of what it earns each time it licenses your records onward to an AI lab.
The same records can earn several payments, but none is guaranteed, because you're paid only when the vendor makes a sale.
Recurring
Workflows, paid per task
Your employees do their real work with frontier AI tools, the most advanced AI assistants available, and the data vendor records, anonymizes and pays for each finished task as a worked example of the job.
micro1 says it pays around $900 per task. Nothing is recorded without your approval.
The Reddit figure comes from Bloomberg and Reuters reporting, summarized by Engadget; neither company disclosed the terms.
How the models compare over five years
The four models pay out on very different schedules over the life of a deal, as the chart below shows. A flat fee arrives in one payment at the start. An annual license pays a steady amount each year, a revenue share pays unevenly as the data is used, and per-task pay arrives in small amounts for as long as your team keeps working.
What a one-time flat fee pays
Companies that license their internal records are paid $100K to $2M or more per dataset, according to the published ranges of micro1 and of Polyshares, another data vendor, which says "typical deals run $100K to $2M and above." The largest company archives draw eight-figure bids, such as Google's winning $10 million bid and micro1's late $12.5 million bid for Spirit Airlines' archive. micro1 publishes a breakdown on its data partnerships page: an entry tier of $100K–$500K for enterprise data that meets its requirements, $500K–$2M for large-scale datasets or ongoing participation across several teams, and $2M+ for highly unique data.
A dataset's price within that range depends mainly on how well its records link together into whole jobs, how hard similar records are to find elsewhere, and how much cleanup they need. What your data is worth explains each price driver, and the data value estimate gives a range for your company.
How recurring models work in practice
micro1 runs two separate programs. Its corpus program pays a one-time fee for an anonymized snapshot of past records. Its workflows program pays per task, so a company earns recurring revenue for as long as its team keeps working with AI tools. micro1's page also says it licenses qualifying agent trajectories, step-by-step records of an employee and an AI tool completing a task together, "for $1–2M annually, with your approval."
Revenue share shows up mostly with vendors that resell a dataset to several labs, because each onward license earns the vendor a fee it can split with you. If a vendor offers one, ask how often it expects to license the records onward and how it will report each sale to you.
Non-exclusive licenses can also earn more than one payment without a revenue share. HUMXN, a data vendor that licenses companies' operational records, says "a non-exclusive licence can be sold to several buyers, and an ongoing feed of new records renews."
04
Exclusive vs. non-exclusive data licenses
The short answer
An exclusive data license means only one buyer may use your data, for the length of the deal, so you can't license it to anyone else. A non-exclusive data license lets you license the same data to other buyers. Exclusive licenses pay more because the buyer gets records that no competing AI lab can train on.
What exclusivity means
Exclusivity is a promise by the licensor, here your company, not to grant the same rights to anyone else during the deal. In a non-exclusive license you make no such promise, so the buyer is one of several possible licensees. The diagram below shows the same dataset under both kinds of license.
Why exclusive licenses pay more
An AI lab pays for data so that its models handle real tasks better than its competitors' models do. Records that every lab can buy give no lab that edge, so a lab pays more to be the only one training on them. An exclusive license also costs your company every other sale of that dataset during the term, so a fair exclusive price has to cover the fees those other buyers would have paid.
Exclusivity comes in degrees
Exclusivity is rarely all-or-nothing. The contract can limit it by time (exclusive for the first year, then non-exclusive), by field (exclusive for AI training, but free to license for analytics), or by buyer type (no other AI lab, but other buyers allowed). A short exclusive period, after which you may license the records to others, is a common middle ground.
Data vendors start from different defaults. SourceX licenses are exclusive for AI training, so a company can't license the same records to anyone else for that purpose. Polyshares licenses are exclusive by default unless both sides agree otherwise. HUMXN lets the company choose for each license and says "Exclusive licences pay more up front for buyers who want sole access."
Don't confuse exclusivity with buyer restrictions
Some contracts limit who may receive the data without making the license exclusive. micro1's page, for example, says data is "licensed exclusively to labs and enterprises based in the U.S. and allied countries." That clause restricts where the data can go; it isn't a promise that only one buyer gets it. Ask the buyer directly whether the license stops you from licensing the same records elsewhere.
How to decide
Exclusivity is worth agreeing to only if the extra money the buyer offers for it outweighs every sale it rules out. Before you say yes, compare at least one other offer, find out how long the exclusivity lasts, and check whether it covers your future records or only the snapshot. If you might sell the business, ask your lawyer whether an exclusive license would complicate that sale.
05
Permitted use: what the buyer may do with your data
The short answer
The permitted-use clause of a data license lists exactly what the buyer may do with your data, such as train AI models, test them, or use the data inside a product, and whether the buyer may pass the data on. Anything the clause doesn't permit, the buyer may not do, so narrower use should mean a lower price and broader use a higher one.
The common uses, defined
Buyers of company data ask for some combination of five rights. Each one is a separate permission, and each changes what the data is worth to the buyer.
- Training
- The buyer feeds the records to an AI model so the model learns patterns from them, such as how a support agent resolves a billing complaint. This is the main use for AI labs and data vendors.
- Evaluation
- The buyer uses the records as a test. The model attempts tasks drawn from them without having trained on them, and its answers are compared with what your staff actually did.
- Building environments
- A data vendor uses anonymized records as seed material for a practice environment: a realistic copy of a company's software, filled with records, where an AI agent, an AI system that carries out multi-step tasks in software on its own, rehearses tasks and is scored on each attempt.
- Product use
- The buyer shows or retrieves the data inside a product its customers use, such as a chatbot that quotes from it.
- Sublicensing and resale
- The buyer licenses the data, or material made from it, onward to someone else. A data vendor needs this right to supply AI labs.
Why sublicensing is the clause to read twice
When you license to a data vendor, the vendor is a middleman: it needs the right to license what it builds from your records onward to AI labs. That onward right is normal, but your company can set conditions on it. The main ones are whether your data must be anonymized first, which kinds of buyers may receive it, and whether they may pass it on again.
micro1's page says its agreement includes "a commitment not to resell or distribute your data until personal and identifying information has been anonymized." How that anonymization works is covered in how company data is anonymized.
Permitted use can also set timing and form
A permitted-use clause can also control when the buyer may use the data and in what form. News Corp's deal with OpenAI, reported by Press Gazette citing the Wall Street Journal, reportedly ensures News Corp's articles don't appear in ChatGPT immediately after they're published. For company records, a similar clause might forbid the buyer's AI from reproducing any of your documents word for word, or require that only anonymized versions ever leave the data vendor.
06
Term, retention and deletion: how a data license ends
The short answer
The term of a data license is how long the buyer may use the data, and published deals have run from two to six years. When the term ends, the contract should require the buyer to delete its copy on a set schedule. Deleting the dataset removes the buyer's files, but it can't remove what an AI model has already learned from them.
How long data licenses run
Published content deals have run from two to six years. AP's 2023 deal with OpenAI, which licensed part of AP's news archive back to 1985, ran two years, PBS NewsHour reported. Reddit's 2024 data licensing deals ran two to three years, according to its IPO filing. Shutterstock's 2023 license to OpenAI runs six years, TechCrunch reported.
A company licensing its records should settle two separate questions about the term. The first is which records the license covers: only the records that exist on the day it's signed, or new ones created during the deal as well. The second is how long the buyer may keep and use them.
Retention and deletion
Retention is how long the buyer keeps your raw records before deleting them. A data vendor usually needs the raw copy only long enough to anonymize it and build training material, so retention can be much shorter than the license term. micro1 says raw data "is deleted according to an agreed schedule," and that data is deleted at the end of an engagement under the agreed terms.
A deletion clause should settle three things in writing. It should give the date or event that triggers deletion, such as the end of the term. It should list what gets deleted, including raw records, anonymized copies and backups. It should also require the buyer to confirm in writing once deletion is done.
What deletion can't undo
A trained AI model doesn't store your records as files; it adjusts millions of internal settings while it studies them. Deleting the dataset afterwards doesn't reverse those adjustments. Researchers call the effort to remove what a model learned machine unlearning, and an introduction to the field notes that retraining a model from scratch without the removed data works but is often too costly to be practical.
Deletion can't reach inside a trained model, so the protections that matter most are the ones applied before the data leaves your company. The data vendor should anonymize the records before anyone trains on them, and records you never want a model to learn from, such as HR files, should be left out of the deal entirely. The diagram below shows the whole life of a license, using micro1's stated timings.
07
What a data license agreement contains, clause by clause
The short answer
A data license agreement answers five questions: what data is covered, who may use it and how, for how long, what is paid and when, and who carries the risk if something goes wrong. The agreement settles each question in a few clauses, which range from exclusivity and permitted use to warranties (promises about the data) and indemnities (promises to cover the other side's costs if someone sues).
The five questions every agreement answers
Data license agreements vary in length and wording, but their clauses group around the same five questions, shown in the diagram below. The clauses outlined in violet are the four terms that most change an offer's price and risk: exclusivity, permitted use, term and deletion.
Each clause in plain words
| Clause | What it says | What to check |
|---|---|---|
| Scope | Which systems, date ranges and record types the license covers. | That it lists only what you mean to license, and says whether future records are included. |
| Exclusions | What is left out, such as HR files, medical information or trade secrets. | That anything you can't or won't share is named. Excluding more may lower the price. |
| Anonymization | Who removes names and identifying details, by which method, and before what point. | That it happens before any third party sees the data, and that you can review samples. |
| Grant and exclusivity | The rights you give, and whether anyone else can get them. | Whether it's exclusive, for how long, and whether it covers future records. |
| Permitted use | What the buyer may do: train, evaluate, build environments, use in a product. | That each use is named, and that broader use comes with a higher fee. |
| Sublicensing | Whether the buyer may pass the data, or material made from it, to others. | Who may receive it, and that it must be anonymized first. |
| Term and retention | How long the license runs and how long raw records are kept. | A clear end date, and retention no longer than the work needs. |
| Deletion | What is deleted at the end, and when. | Raw copies, anonymized copies and backups, with written confirmation. |
| Fee and payment | The amount, the payment model and when payment is due. | Whether payment follows signing, transfer or acceptance, and how long each step can take. |
| Acceptance | How the buyer confirms the data is complete before paying. | A time limit on the review, and what happens if data is rejected. |
| Warranties | Promises you make about the data, mainly that you have the right to license it. | That each promise is one you can keep; limit them to what you know. |
| Indemnities | Who pays if a third party sues over the data. | That your exposure is limited and matches the fee. |
| Liability cap | The most either side can owe the other if something goes wrong. | That a cap exists, often set at the fee paid. |
| Confidentiality and security | How the buyer stores, protects and limits access to your data. | Encryption, separate storage and who on their side can see it. |
Warranties and indemnities, explained
A warranty is a promise in the contract that something is true, such as "the licensor has the right to license this data." If the promise turns out to be false, the buyer can claim against you. An indemnity is a promise to cover the other side's costs if a specific problem arises, such as a lawsuit from a customer whose information was in the records.
These two clauses carry most of the seller's risk, because company records often hold other people's information. A support inbox holds customers' names and addresses, a contracts folder holds suppliers' confidential prices, and some software providers' terms limit what may be done with data exported from their apps. Owning the records isn't the same as having the right to license everything in them.
Is it legal to sell company data? covers what a lawyer checks.
Payment timing: read the steps, not just the number
Payment usually follows acceptance, not signing. micro1's page says "we pay in full within 30 days of acceptance," while its FAQ describes a 30-day review followed by payment within 30 days of invoicing. Mercor's data page says payment by wire "typically occurs within 2-4 weeks of the signed agreement." Plan your cash flow around the slowest timeline the contract allows.
Before you sign
Have a lawyer read the agreement and talk to your CPA about tax. On a six- or seven-figure deal, the cost of that review is small, and a serious buyer will wait for it. Don't send samples before legal review, because a sample already discloses real company records.
08
Data licensing examples: real deals and what they show
The short answer
The best-known data licensing examples are publishers and platforms licensing content to AI labs: AP to OpenAI in 2023, Shutterstock to OpenAI in 2023, Reddit to Google in 2024 and News Corp to OpenAI in 2024. The newest examples are ordinary companies licensing internal records to data vendors for $100K to $2M or more per dataset.
Published content deals
Large content deals are the best public evidence of how data licensing works, because news reports and filings have revealed part of their terms. Each deal in the table below shows a different term in practice.
| Deal | Year | What was licensed | What it shows |
|---|---|---|---|
| AP → OpenAI | 2023 | Part of AP's news archive back to 1985 | A two-year term; fee undisclosed, and AP also got access to OpenAI's technology |
| Shutterstock → OpenAI | 2023 | Images, video, music and their descriptions | A six-year term for a large library |
| Reddit → AI companies, reportedly incl. Google | 2024 | Access to Reddit's posts through its Data API | Recurring fees: $203M of contracts over two to three years |
| News Corp → OpenAI | 2024 | Current and archived articles from the WSJ, The Times and others | Reportedly over $250M over five years, partly in credits |
| Spirit Airlines → Google | 2026 | Internal email, chat and business data of a bankrupt airline | A sale, not a license: Google's winning $10M bid at a bankruptcy auction, then micro1's late $12.5M bid, pending court approval |
Reddit's IPO filing disclosed data licensing arrangements "with an aggregate contract value of $203.0 million and terms ranging from two to three years," signed in January 2024. Shutterstock's chief executive said its AI licensing business brought in $104 million in 2023, according to an interview with Bloomberg, with buyers reported to include Meta, Amazon and Apple.
Company records: the newer, quieter deals
The fastest-growing category is operating companies licensing their internal records to data vendors. These deals are rarely announced, because both sides usually treat them as confidential. The companies in them are paid $100K to $2M or more per dataset.
Some companies are paid per task instead of per dataset. The owner of a 14-person HVAC company, a business that installs and repairs heating and air conditioning, is paid about $1,000 per task to grade how AI models handle his firm's routine work, such as invoices, credit applications and payroll, The Information reported. At his planned pace, that work comes to more than $150,000 a quarter.
micro1 announced in October 2026 that it would spend $1B on enterprise data acquisition over the next 12 months. We collect more dated deals, with sources, in data monetization examples.
09
Data licensing companies: who licenses company data
The short answer
Data licensing companies that buy company records for AI include data vendors such as micro1, Mercor, Handshake AI, Polyshares, HUMXN and SourceX, which license records from businesses and supply AI labs. Each publishes different terms, so comparing more than one offer is the simplest way to get a better deal.
The data licensing business model
The data licensing business model runs in a chain. Your company licenses records to a data vendor. The vendor anonymizes them, turns them into training material such as practice environments and worked examples, and licenses that material to AI labs. The vendor earns the difference between what labs pay it and what it pays you.
For your company, the result is new revenue from records it has already created, and the company keeps ownership of them. Referral fees, such as the one micro1 pays us when a company we introduce is accepted, are paid by the vendor, not by you.
What the main data vendors say they offer
The figures below are each company's own published statements, checked in November 2026. They're stated ranges, not guarantees.
| Company | Stated pay | Stated terms |
|---|---|---|
| micro1 | $100K–$2M+ | Licenses data; "companies retain ownership." Pays within 30 days of acceptance. |
| Mercor | Not published | "We license it, package it, and pay you for it." Paid 2–4 weeks after signing. |
| Handshake AI | $100K–$4M+ | You define the records included and exclude sensitive information. |
| Polyshares | $100K–$2M+ | Buys the anonymized operating record directly for a flat license fee. Licenses are exclusive by default. |
| HUMXN | $250K+ typical | Says deals "start around $250,000 and run into the millions for large archives." You choose exclusive or non-exclusive for each license. |
| SourceX | Up to $4M | A one-time fee per dataset, paid within 60 days of a valid invoice after closing. Licenses are exclusive for AI training. |
Each of these vendors takes applications directly. You can apply to micro1, contact Mercor's data team, create a Polyshares seller account, ask HUMXN for a valuation or start a SourceX application. Comparing two or three written offers is the simplest way to see which terms suit your company.
Data marketplaces, online stores such as AWS Data Exchange where companies sell datasets by subscription, and data brokers, firms that buy and resell information about people, are other routes, with different buyers and different work involved. Where to sell your data compares them, and does my company qualify? covers the size and data each vendor asks for.
10
How to prepare for a data licensing deal
The short answer
Before a data licensing offer arrives, decide which records you'll license and which you'll exclude, have a lawyer confirm you hold the rights, and settle your position on exclusivity, permitted use, term and deletion. Offers in this market can come with short deadlines, so preparation is what lets you answer quickly without agreeing to the wrong terms.
The order to do it in
List what you have Day 1
Write down the systems your records live in, such as help desk, chat, email and CRM, how many years each covers, and roughly how large each is. Vendors call this a data manifest; it describes the data without sharing any of it.
Decide what stays out Week 1
Write down which records will stay out of the deal, such as HR files, medical information, trade secrets and anything a client contract says you can't share.
Have a lawyer check your rights Week 1–2
A lawyer confirms what you're allowed to license, including customer details and data held under software providers' terms, before anything leaves the company.
Set your position on the four terms Before the offer
Decide your answer on exclusivity, permitted use, term and deletion, and what price would change it.
Compare offers, then sign Offer
Get more than one offer where you can, read the payment steps, and have your lawyer and CPA review the final agreement.
Mistakes that cost sellers
- Granting exclusivity without a premiumAn exclusive license should pay noticeably more than a non-exclusive one, because it rules out every other buyer for the term.
- Leaving permitted use open-endedA license for "any purpose" gives the buyer nearly everything a sale would. Name each permitted use, and charge more for broader ones.
- Accepting deletion "on request" with no dateAsk for a schedule, a list of what is deleted, and written confirmation.
- Promising more than you knowMake warranties only about facts you can check. An indemnity with no cap can cost your company more than the whole fee if a third party sues.
- Sending samples before legal reviewA sample is a disclosure. Once it's out, it can't be taken back.
Only the company can license its data, never an individual employee. If you're unsure where your company stands, start with does my company qualify? and an estimate of what your data is worth.
Questions
Frequently asked questions
What is data licensing?
Data licensing is a contract that lets a buyer use a copy of your data for agreed purposes and an agreed time, in exchange for payment, while you keep ownership. The agreement sets whether the license is exclusive, what the buyer may use the data for, how long the deal runs and when the buyer must delete its copy. Companies licensing internal records to AI labs and data vendors are paid $100K to $2M or more per dataset.
What is the difference between licensing and selling data?
Licensing data gives a buyer a copy and limited rights to use it, while your company keeps ownership and can usually license the same data again. Selling data transfers ownership, so the buyer can use the records in any lawful way and the seller keeps no rights. Operating companies almost always license their data; outright sales mostly happen when a company closes, as with Spirit Airlines' archive, which drew Google's winning $10 million bid at a 2026 bankruptcy auction.
How much does data licensing pay?
Companies that license internal records to AI labs and data vendors are paid $100K to $2M or more per dataset, and the largest company archives draw eight-figure bids, such as the $10M and $12.5M bids for Spirit Airlines' archive. micro1, one data vendor, breaks this into an entry tier of $100K–$500K, $500K–$2M for large-scale datasets, and $2M+ for highly unique data. Most small and mid-size companies receive a one-time flat fee; annual, revenue-share and per-task deals also exist.
What is the difference between an exclusive and a non-exclusive data license?
An exclusive data license gives one buyer the sole right to use your data for the length of the deal, so you can't license it to anyone else. A non-exclusive data license lets you license the same data to other buyers. Exclusive licenses pay more because the buyer gets data its competitors can't have, but the higher fee has to cover every sale you give up during the term.
What should a data license agreement include?
A data license agreement should state the scope of data covered and what is excluded, how the data is anonymized, whether the license is exclusive, the permitted uses and any right to sublicense, the term, retention and deletion schedule, the fee and when it's paid, and the warranties, indemnities, liability cap and confidentiality terms. A lawyer should review it before any data or samples are shared.
Does deleting licensed data remove it from an AI model?
No. Deleting a licensed dataset removes the buyer's copy of the records, but it can't remove what an AI model already learned from them during training. Removing that learning, called machine unlearning, usually means retraining the model, which is costly. That's why anonymization before transfer and clear exclusions for sensitive records matter more than the deletion clause.
What are data licensing companies?
Data licensing companies are businesses that license data from its owners and supply it to buyers. For company records used in AI, the main ones are data vendors such as micro1, Mercor, Handshake AI, Polyshares, HUMXN and SourceX, which anonymize business records and turn them into training material for AI labs. Each publishes different pay ranges and terms, so comparing offers is worthwhile.
What is a business data licence?
A business data licence is the British spelling of a business data license: a contract that lets one business use another business's data under set terms, without ownership changing hands. It covers the same terms as any data license agreement, including scope, exclusivity, permitted use, term, deletion and payment.
Bottom line
Know the terms before the offer arrives
Offers in this market can come with short deadlines. Knowing what to ask about exclusivity, use and deletion ahead of time means you can say yes quickly without saying yes to the wrong deal.
Not legal, tax or financial advice. Figures come from micro1's published pages and public reporting as of November 2026.