Best Places to Sell Data to AI Companies in 2026
The best places to sell data to AI companies in 2026: DataVendor, Mercor, Turing, and SimpleClosure compared on price, preparation, risk, and repeat sales.
By DataVendor team
If you own a production codebase or proprietary business data, AI labs may pay to train on it. Not every asset qualifies, and not every route pays the same. Where you sell decides who sets the price, who does the work of the sale, and whether you can sell the same asset more than once.
At DataVendor, we broker sales of code and data to AI labs. This guide compares DataVendor with Mercor Data, Turing's Project Lazarus, and SimpleClosure's AssetHub. DataVendor wrote this guide. The details about each company come from its own public pages and are current as of November 2026.
Know which lane fits your asset
There are two main ways to sell company data to AI labs, and the right one depends on the quality of what you have.
The first is a broker that sells your asset to labs on your behalf. DataVendor works this way. It helps you prepare the asset, lists it for the labs that buy on DataVendor, and can license it to several of them. Because more than one lab can buy the same asset, a high-quality asset can earn more this way than from a single sale. It also carries a risk: an asset that labs do not want will not sell.
The second is a managed vendor that sets the price for your data and handles the sale to labs itself. Mercor works this way. The vendor does more of the work, so your team does less. In exchange, the vendor sets the price, and for a strong asset it is often lower than the asset could earn through a broker. Payment terms vary by vendor. Mercor pays once, when it delivers the dataset to a buyer, so you do not share in later sales.
| Broker (DataVendor) | Managed vendor (such as Mercor) | |
|---|---|---|
| Who sets the price | You, with DataVendor's help | The vendor |
| Who finds the buyer | DataVendor, across many labs | The vendor |
| How you are paid | Each time a lab licenses the asset | Set by the vendor; Mercor pays once, on delivery to a buyer |
| Earnings from a high-quality asset | Can add up across several licenses | One price, often lower |
| Your work | DataVendor's team helps you assess, prepare, and package the asset, including PII removal and secrets cleanup | The vendor extracts and packages the data |
| Best for | High-quality assets that labs want | Assets of uncertain value, or when you want the vendor to set the price and do the work |
If you have a high-quality asset, sell it through DataVendor. If you are not sure, get an estimate first. If the asset is unlikely to sell on its own merits, a managed vendor that sets the price and handles the sale may be the better choice.
Compare the four options
The table shows how the four options differ, and the sections below explain each one.
| Platform | How it works | Best for | Main trade-off |
|---|---|---|---|
| DataVendor | A broker that helps prepare your asset and sells it to several labs | High-quality codebases and proprietary data that labs want | Weaker assets may not sell |
| Mercor Data | A managed vendor that packages your data and handles the sale to labs | Companies that want Mercor to set the price and do the packaging | One payment, often lower for a strong asset; exclusivity and post-deal rights are not published |
| Turing Project Lazarus | Turing prepares your records and licenses them for a revenue share | Companies, open or closed, that want Turing to run the sale | Turing is the only seller for 90 days, and the split is not published |
| SimpleClosure AssetHub | A broker attached to a shutdown service | Founders recovering value while they close | The success fee is not published |
1. DataVendor
DataVendor has relationships with all of the leading AI labs and brokers sales of code, data, and training tasks to them. It suits a founder with one production codebase, a consulting company with several repositories, or an operator with proprietary records from a specialized business. You do not need to be shutting down, and you do not need to sell your company. Sellers on DataVendor include several publicly traded companies.
Get help preparing the asset
DataVendor's platform fee is a share of your earnings, charged when you withdraw them, so it has every reason to help you turn your data into an asset that labs will pay for. Its team helps you assess buyer fit, prepare the asset, and list it. Legacy enterprise codebases, including COBOL, Java, and mainframe systems, are welcome.
You can list private code repositories with their history and raw data delivered as ZIP files. Tasks imported through HUD sell through private listings submitted to buyer requests. DataVendor imports repositories from GitHub, GitLab, and Bitbucket, with their pull requests and reviews. It also imports project data from Jira and Confluence, though those imports cannot be listed for sale yet. Slack, email, and other exports upload as raw data, and the team can help package records that need special handling. If a codebase is not ready to upload, you can register it in Inventory for an initial assessment, then attach the code before you list it.
DataVendor's team removes personal information and committed secrets from codebases and workplace exports, such as email, Slack, Google Drive, and document exports, with the scope agreed for each asset. You do not need to finish that cleanup yourself before you ask for an assessment.
Before a purchase, buyers see metadata, redacted file trees, and quality evidence, but never the raw files. That lets you show what the asset contains without handing it over. To see current demand, open the Wanted tab after you accept the vendor agreements. It lists open requests from buyers, and you can submit a listing that meets one.
Know what high-quality data earns
Each lab decides what to pay based on its own training priorities. An asset that does not interest one lab may fill an important gap for another. Because DataVendor has relationships with all of the leading labs, a listing puts your asset in front of the buyers most likely to pay for it. High-quality codebases from large engineering teams sell to the most lab buyers.
DataVendor's strict confidentiality policies prohibit us from naming buyers or sellers, or from sharing the figures from specific deals, so the numbers here are directional, current as of November 2026, and not guarantees. About $30,000 per license is the floor for a good codebase, and some deals for a single codebase have exceeded $10 million. A codebase built by several dozen engineers, with thousands of pull requests and over a million lines of mostly human-written code, recently licensed in the low six figures.
Engineering quality matters as much as size. A codebase with similar history but weaker engineering can sell for about a third as much as one built by top engineers. Strong assets also move quickly. For a complex codebase with a rich engineering record, procurement can take as little as two to four weeks.
Labs pay the most for code written mostly by people and for complex products, rather than simple apps that only store and edit records. They value five or more years of git history, issue tickets, and documentation, with more than five employees on the team in each of those years. Systems languages such as Rust, Go, C, and C++ sell for more, and COBOL and Swift are in high demand. Our guide to selling source code covers the full criteria.
Know the terms
You set the price of each listing. DataVendor charges a 20% platform fee on earnings when you withdraw them, so you keep 80% of each sale before any payout fees. Earnings are available as soon as the buyer pays, and withdrawals are usually reviewed within three business days.
A sale on DataVendor licenses your asset for a commercial use. You keep ownership by default and grant the rights that the use requires. The exact rights depend on the agreement behind the listing. If a buyer asks for exclusivity, compare the price with the other labs you would have to turn away.
Know when DataVendor is not the right fit
Poor-quality data usually does not sell on DataVendor, and lower-quality codebases stay on the marketplace longer. If your asset is small, thin on history, or unlikely to interest labs, DataVendor is probably not the best place for it. A managed vendor that sets the price and handles the sale may serve you better. An estimate from DataVendor can tell you which case you are in.
2. Mercor Data
Mercor Data licenses workplace data from operating companies. Mercor connects to more than 100 tools, including Slack, Gmail, GitHub, and Salesforce. It extracts and de-identifies the data, packages it, and sells it to labs itself. Mercor says no lab or third-party vendor sees the raw data.
Mercor bases compensation on the volume of data, the tools you connect, and the depth of the records. It states that there are no upfront costs, setup fees, or deductions. Payment arrives by wire when the anonymized dataset reaches the buyer, typically within two to four weeks.
Mercor takes on much more of the packaging than a broker does, and it handles the sale to labs, so your team does little work. The trade-off is the price itself. Mercor sets one price for the dataset and keeps the margin from selling it on, so a high-quality asset that several labs would want often earns less this way. The public page also does not say whether the license is exclusive, who holds which rights after the deal, or which labs buy the data. Ask for those terms in writing before you connect any systems.
Mercor suits a company with workplace data of uncertain value, or one that wants a set price and the packaging done for it more than the highest possible return.
3. Turing Project Lazarus
Project Lazarus licenses company operating records, including chat, email, tickets, wikis, CRM records, and databases. Source code is optional. Turing says communication, tickets, and documents often matter more than code.
The process starts with a mutual NDA, followed by a volume estimate, samples, and a quality check. Turing prices from a real sample rather than from a form. It accepts companies that are operating, winding down, or already shut down, and it covers de-identification and preparation, recovering those costs from sales.
Turing's model sits between the two lanes. It runs the sale for you, but you are paid a share of each sale rather than a set price. Licenses are non-exclusive, ownership does not change hands, and you owe nothing if nothing sells. Two terms are worth noting. Turing is the only one selling your record for the first 90 days, after which you can renew or opt out. The page also does not publish the revenue split.
Lazarus suits a company that wants Turing to prepare and sell a broad operating record, including a company that has already shut down.
4. SimpleClosure AssetHub
SimpleClosure AssetHub sells source code and workspace data from tools such as Slack, Google Drive, email, Notion, and Figma, along with IP, domains, and other startup assets. SimpleClosure assesses each asset, processes datasets for personal information, packages them, and matches them with buyers.
A deal can license the data while you keep the underlying assets, or sell the assets outright. SimpleClosure takes a percentage only when an asset sells or is licensed, but it does not publish the rate, the exclusivity terms, or the payment timing. Operating companies can use AssetHub, but its main draw is its connection to SimpleClosure's shutdown service.
AssetHub suits founders who want asset recovery handled as part of closing their company.
See how repeat licensing adds up
Repeat licensing is how a broker can earn more for a strong asset. Consider a hypothetical codebase. A managed vendor such as Mercor offers $100,000 for the dataset and handles the sale itself. You are paid once, and later sales earn you nothing.
Now suppose four labs each buy a $100,000 non-exclusive license through DataVendor instead. After the 20% platform fee, you keep $80,000 from each sale, or $320,000 in total.
| Scenario | Buyer payments | Seller keeps |
|---|---|---|
| One set-price sale to a managed vendor | $100,000 | $100,000 |
| Four non-exclusive licenses on DataVendor | $400,000 | $320,000 after the 20% fee |
The fee matters in this comparison. At 20%, a single DataVendor license must sell for $125,000 to match a $100,000 set price with no deductions. Two licenses of $62,500 each also match it, and every license after that adds to your total.
The second outcome needs four interested labs, so treat it as an illustration rather than a forecast. It only applies to an asset that labs want. For a weaker asset, a firm $100,000 offer may be worth more than a listing that never sells.
FAQ
What is the best place to sell data to AI companies?
It depends on the quality of the asset. For a high-quality codebase or proprietary dataset that labs want, DataVendor can earn you more, because it brokers the sale and can license the asset to several labs. For data of uncertain value, a managed vendor such as Mercor sets the price and handles the sale, usually for a lower return. If you are closing your company, SimpleClosure's AssetHub ties asset sales to the shutdown process.
How much does DataVendor charge sellers?
DataVendor charges a 20% platform fee on earnings when you withdraw them. Converting earnings to credits is free. Stripe payout fees may apply in some countries.
Do I keep ownership of my data when I sell it?
On DataVendor, a sale licenses your asset, and you keep ownership by default. Turing's Project Lazarus also says ownership does not change hands. SimpleClosure offers both licenses and outright sales, and Mercor does not publish its ownership terms.
Can I sell the same codebase to more than one AI lab?
Yes, if your agreements allow it. Through a broker such as DataVendor, a non-exclusive license leaves you free to license the same asset to other labs. When a managed vendor such as Mercor sets one price for your data and sells it on, you are paid once.
Should a startup that is shutting down try to sell its data?
It is worth trying. A startup that closes without checking whether its code and records have buyers may miss a chance to recover some value. Not every startup qualifies, and data that is not high quality will not sell. Get an estimate before accounts close, and preserve the exports and admin access you would need. Turing's Project Lazarus also accepts companies that have already shut down, and SimpleClosure's AssetHub is built around shutdowns.
Find out which lane fits your data
Start by finding out whether your asset is one that labs will pay for. Get an estimate for your codebase, or book a call for a quote on other data. If your data is high quality and you want to earn from more than one sale, DataVendor is the right place to sell it.