DataVendor by HUD: What It Buys, Fees and Payouts

DataVendor is a real marketplace, run by HUD, where companies and teams license private data to AI companies: codebases, raw company records, task sets and RL environments, on listings they price themselves. HUD, which builds software for AI training environments, raised a $16 million Series A led by Standard Capital in June 2026 and says vendors on it are already doing millions a month (its own claim, with no total). DataVendor keeps 20% of your earnings, charged only when you withdraw them; in its own example, a $2,000 withdrawal releases $1,600. It says about $30,000 per licence is the floor for a good codebase, that some single-codebase deals have exceeded $10 million (no deal named), and that requests for tasks are priced at around $500 to $1,500 a task. Anyone can sign up to sell, but buyers join a waitlist and are invited at a managed pace, so DataVendor decides who sees your listing.

This guide is for the founder, CTO or counsel of a company that found DataVendor, was told about it, or searched for it to check whether it is genuine, what sells there and what it costs. It reads DataVendor's public documentation as it stood on 9 November 2026: what it buys, how a listing is priced and checked, what DataVendor keeps, how and when you are paid, what you sign and what a buyer sees. It then sets DataVendor beside the other ways to license code and records, and walks a software company through it.

AIData.Broker introduces holders to DataVendor through a referral link held by AITraining.jobs, a company under the same ownership as ours. DataVendor pays that referral fee, 5% of what DataVendor receives on the vendor's deals within 180 days, out of its own share, so a vendor's price and DataVendor's fee stay the same. That is how we work with every partner: we introduce a holder to every buyer whose published rules take its data, free for the holder, because the buyer pays us its published referral payout out of its own money; where a sale run by us earns more, and only for holders based in the United States, the United Kingdom and Canada, we act as the holder's agent under a mandate and the holder pays our fee only if the sale closes. Either way the holder itself licenses its data to the buyer, and we do not buy it, take ownership of it, or license or sell it in our own name. Everything below comes from DataVendor's public pages, cited to its record; nothing seen inside a DataVendor workspace is used here.

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

Contents

1. What DataVendor is
2. What sells on DataVendor
3. What it pays: prices on DataVendor
4. What DataVendor keeps, and when you are paid
5. Signing up and what you accept
6. Listing, checks and the estimator
7. Buyers, offers and requests
8. Rights, cleaning and what a buyer sees
9. DataVendor beside other ways to sell code and records
10. Worked example: a software company with a private codebase
11. Questions people ask
12. What this means for you

What DataVendor is

DataVendor is a marketplace with two sides. On one side are vendors, the companies and teams that hold private data, code or tasks; on the other are buyers, AI companies that need material to train and test their models. Vendors package what they hold into listings and set a price; buyers either buy a listing at that price or make an offer, and they also post requests for supply they cannot find. Everything belongs to an organisation rather than a person, and only an organisation's admins can move money.

It belongs to HUD, a company that builds a platform and an open-source kit for building environments and tasks, evaluating AI agents against them and training models on the results. That is why DataVendor's own kinds of supply include task sets and RL environments built on HUD's platform, beside codebases and raw records: HUD's customers build agent training material, and DataVendor is where that material and the private data behind it can be sold. HUD's Series A of $16 million, led by Standard Capital, was announced on its own blog in June 2026, and the same post says it opened DataVendor so independent builders could supply that data.

Is DataVendor legitimate? It is a named company's product, with public documentation, published fees and agreements, and an investor of record. What it does not publish is a list of its buyers or a count of completed sales, and DataVendor itself estimates that vendors typically complete at least one deal within two to eight weeks, adding that a listing does not guarantee a sale. Its documentation lives at docs.datavendor.ai and its sign-up at datavendor.ai; nothing legitimate about it asks for your code by email (see Rights, cleaning and what a buyer sees).

What sells on DataVendor

DataVendor names four kinds of supply. Codebases are private repositories with their history. Raw data is the underlying material, delivered as a ZIP: documents, logs, examples, enterprise records such as emails or org charts. Task sets are structured sets of tasks a person or a model can attempt, built on HUD's platform. RL environments, interactive setups where an AI agent acts and is scored, are never sold alone: they ship with the task sets that run on them.

It also says what buyers want now: strong demand for enterprise software codebases and for cybersecurity, legal, and finance data. The common thread is that the supply is private. DataVendor is closed to public repositories, since a buyer can already clone them, and it judges value by whether your rights are clear, how strong demand is, how well documented the asset is, how hard it would be to recreate and how directly it fits a buyer's need.

For a company's records, DataVendor is one of several buyers whose own pages take them; our guide to selling company data to AI labs compares them, and what labs buy shows what each asks for. For code, our guide to selling a codebase explains why buyers count the tasks a repository yields rather than its lines, and what raises or kills its value.

What it pays: prices on DataVendor

DataVendor does not buy anything itself. Vendors set the price of each listing, and buyers either pay it or offer their own. What DataVendor publishes are guides to what supply fetches: about $30,000 per licence is the floor for a good codebase, some deals for a single codebase have exceeded $10 million, and poor-quality codebases usually do not sell. Buyers' requests for tasks are priced around $500 to $1,500 a task, with hundreds to thousands of tasks per request.

Two things about those figures matter. The codebase floor is per licence, and DataVendor's licences are for a use rather than a transfer, so the same codebase can in principle be licensed more than once, each licence priced on its own. And the floor is DataVendor's statement, not a record of sales: no completed DataVendor sale is public. It is higher than the rest of the market's published offers for a repository, whose median is $2,121 per repository, and it sits at the top of the middle half of public code prices, $1,478 to $18K per repository.

Because a vendor names its own price, the number on a listing is a decision, not a quote. DataVendor's estimator returns an advisory value band, dated when it was priced, for each private repository you connect, and buyers who disagree with a price make an offer. Our valuation gives a second, independent range from the market's published prices, and the prices guide sets out what each kind of data fetches.

What DataVendor keeps, and when you are paid

DataVendor's fee is simple and comes at one moment: it keeps 20% of vendor earnings, assessed only when earnings are withdrawn. Its own worked example is that a $2,000 withdrawal releases $1,600 to Stripe, the payment company it pays out through. Buyers currently pay no transaction fee. In some countries DataVendor also passes Stripe's payout fee on to you; elsewhere it pays that fee itself, and the withdrawal quote shows which applies.

You can avoid the fee in one way: earnings can be converted into HUD credits one for one, with no fee. But credits only pay for things on HUD's own platform, such as optional environment checks, and the conversion cannot be undone, so credits are not money. A vendor that wants cash pays the fee.

The last row is the one to understand before you sell. All sales are final, with no refunds on either side, but if a buyer's bank reverses a card payment and the dispute is lost, the sale amount is taken back from your wallet with a $15 dispute fee. Until a payment can no longer be reversed, a sale is not quite settled money, so keep a record of what you delivered and when. Our guide to when AI data buyers pay compares DataVendor's timing with other buyers', and comparing offers net of the platform's cut explains why a listed price and a direct offer are not the same number: DataVendor takes its share at withdrawal, while some buyers quote what the holder receives with nothing deducted.

Signing up and what you accept

Selling is self-serve. You sign up at datavendor.ai with a work email, which creates an organisation, choose Sell data, and DataVendor lets a vendor organisation list as soon as its setup finishes, with no application review. To open the workspace you accept four agreements for the whole organisation, once: a mutual NDA, the Vendor Platform Access Agreement, a Data Processing Addendum and the Vendor Data Sharing Agreement. Each acceptance box starts ticked.

Read the four before you accept them, and let whoever can sign for the company accept them, because they bind the organisation rather than the person clicking. The NDA protects what both sides share, including your code and buyers' requirements; the platform agreement sets the rules for using DataVendor; the data processing addendum governs any personal data that passes through it; and the data sharing agreement governs what you license. A listing does not transfer your rights: a sale licenses your assets for a commercial use, you usually keep ownership, and the exact rights depend on the agreement behind the listing. By submitting a listing you also confirm that you own the supply and may license it.

What DataVendor does not publish is the licence a buyer receives for each listing in a form you can read before you list, whether a listing can be sold exclusively to one buyer, or whether a buyer may pass the data on. Ask before your first sale. Our guide to what a data licensing agreement holds lists the terms worth checking, and the licence terms page sets DataVendor's published clauses beside other buyers'.

Listing, checks and the estimator

A listing starts in your Inventory, where you import what you could sell. Codebases come in through a GitHub app limited to the repositories you choose, or through GitLab and Bitbucket access, or as a ZIP; raw records come as a ZIP; Jira data as its CSV export. You then package assets into a listing, write its title and description, set the price and submit it.

Automated checks run after you submit. The mandatory ones, which decide whether the listing can go live, cost nothing; for a codebase they include credentials and personal data in the listing text, and a readable copy of the repository with its Git history and captured pull requests. Optional checks on environments are paid in HUD credits, and a listing that passes every check earns a verified badge that buyers can see. Editing the listing removes the badge until the checks pass again.

Before listing a codebase you can run DataVendor's estimator. It archives each private repository's source and history and returns an advisory value band, dated when it was priced; a public repository cannot be estimated, and nothing is shown to buyers. The estimate binds nobody, so use it as one view beside the market's published prices. If preparing a listing would take real work, DataVendor asks first for a short inventory, ideally with a small representative sample, and tells you not to send data or code by chat or email.

Buyers, offers and requests

DataVendor's buyers are AI companies it has let in. Buyer organisations join a waitlist, and DataVendor invites them at a managed pace, so a listing reaches only the buyers DataVendor has chosen. It does not publish who they are.

Buyers reach your supply in two ways. On a listing, a buyer either buys it at your price or makes an offer; an offer names the buyer organisation to you, so you choose whom to accept, while a buyer who buys outright stays anonymous. On a request, which DataVendor calls an opportunity, a buyer posts its requirements and a deadline, and DataVendor decides which vendors see it; you answer by submitting a listing that meets it. Requests often come in batches around model training runs, so they open and close quickly, and you see only those DataVendor matches you to.

Once a buyer has paid, its access to what it bought stays available even after you edit, hide or archive the listing, and the pages give no end date for that access. That is normal for a licence of training data, since a model already trained on your data cannot unlearn it, but it is a reason to list only what you are content to license for good. Our guide to exclusive and non-exclusive licences explains how the length and scope of a licence move its price.

Rights, cleaning and what a buyer sees

DataVendor leaves the cleaning to you. It asks vendors to remove secrets, credentials, customer information and anything you do not have the right to share before connecting or uploading anything. Private code needs this more than most people expect: in GitGuardian's 2026 study, 32.2% of internal repositories held at least one hardcoded secret, against 5.6% of public ones. Scan the whole history, not only the latest version, and revoke every key you find.

The rights question is who wrote the code and what it contains. Code your employees wrote is the company's as a work made for hire in the United States; code an agency or contractor wrote is yours only if a written assignment moved it, since a transfer of copyright is not valid unless it is in writing and signed. Code you built for a client under contract is usually the client's, and open-source parts keep their own licences. Raw company records raise the further question of the people in them, which our guide to whether it is legal to sell customer data covers.

What protects you during a sale is that buyers never see the raw files first: before buying, they see metadata, redacted file trees and quality evidence, never your raw files. That is the pattern most careful buyers follow, pricing from a profile or a metadata scan before any file moves. Our readiness check walks through the rights and cleaning questions for your own data.

DataVendor beside other ways to sell code and records

DataVendor is a listing site: you set the price and pay a share when you take the money out. Other routes work differently, and the difference shows in who sets the price, what is kept and how fast you are paid. Platforms that only carry the payment keep little; a payment rail keeps 2 to 5 percent, a listing site 15 to 30. DataVendor's 20% of earnings sits in the second group.

Each route has its own guide: Turing's Project Lazarus, Google's Content Offer Pilot, and the data platforms compared. Nothing in DataVendor's published terms makes a listing exclusive, so a holder can usually list on DataVendor and talk to other buyers too; read each other buyer's clause on earlier licences before you sign, since some ask you to promise you have not licensed the same data elsewhere.

Worked example: a software company with a private codebase

An example, not a real company. A 30-person company that sells software to insurers has a private monorepo going back to 2018, with Jira tickets linked to its pull requests and a test suite. Its staff wrote all of it. The founder hears about DataVendor and wants to know whether the code is worth listing.

First, fit. Enterprise software code is what DataVendor says buyers want now, and its history with tickets and tests is what DataVendor asks vendors to send. The repository is private, so it can be listed and estimated.

Second, cleaning. A scan of the full history finds old cloud keys in early commits; the company revokes them and rewrites the history before importing. A folder of customer test fixtures holds real policyholder data and is taken out, along with anything written for a client under contract.

Third, a price. The founder runs DataVendor's estimator, which returns a dated advisory band, and a valuation for a second view. DataVendor's own floor for a good codebase is about $30,000 per licence, above the market's median offer of $2,121 per repository a repository, so the founder prices the listing from both and leaves room for offers.

Fourth, the money. If a buyer pays, the earnings are available at once; withdrawing them costs 20%, as in DataVendor's own example where a $2,000 withdrawal releases $1,600, and the review typically takes three business days before Stripe pays. The founder compares that net figure with what other code buyers would pay directly, keeps the record of what was delivered in case a charge is ever disputed, and stays free to license the same code elsewhere.

Questions people ask

Is DataVendor legit?

Yes. It is HUD's marketplace, with public documentation and published fees, and HUD raised a $16 million Series A led by Standard Capital in June 2026. It publishes no list of buyers or completed sales, and it says vendors typically complete a first deal within two to eight weeks, with no guarantee of a sale.

What does DataVendor charge?

20% of your earnings, only when you withdraw them; in its own example a $2,000 withdrawal releases $1,600. In some countries it also passes on Stripe's payout fee, and a lost card dispute costs the sale plus a $15 fee. Buyers currently pay no fee.

How much can I make selling a codebase on DataVendor?

DataVendor says about $30,000 per licence is the floor for a good codebase, that some single-codebase deals exceeded $10 million, and that poor-quality codebases usually do not sell. You set the price yourself; its estimator gives an advisory band first.

How do I get paid by DataVendor?

Through Stripe, to a bank account in the country you choose when you set up payouts, a choice that is permanent for that account. Withdrawals start at $1,000 before fees, and DataVendor reviews each one, typically within three business days.

Who buys on DataVendor?

AI companies that DataVendor invites from a waitlist; it admits buyers at a managed pace and does not publish their names. A buyer who makes an offer is named to you; one who buys outright stays anonymous.

Can I sell company records on DataVendor, not just code?

Yes. Its raw data listings take documents, logs, examples and enterprise records such as emails or org charts, delivered as one ZIP, and it names cybersecurity, legal and finance data among what buyers want now. You must remove customer information and anything you have no right to share first.

Do I keep ownership of what I sell?

Usually. A sale licenses your assets for a commercial use; you keep ownership and grant the rights that use needs, and the exact rights depend on the agreement behind the listing. A buyer's access stays available after you edit or archive the listing.

What this means for you

DataVendor suits a holder that has private code, records or tasks, is willing to prepare and price its own listing, and accepts a 20% fee on what it takes out. These steps make the most of it.

1. Check the fit: private, documented supply in the areas buyers want now. The readiness check asks the rights questions.
2. Clean before you connect: secrets out of the full history, customer data and client-owned code out.
3. Read the four agreements before whoever can sign for the company accepts them.
4. Price from two views: DataVendor's estimator and a valuation against the market's published prices.
5. Compare net figures: a DataVendor price less its fee against what other buyers would pay you directly. Our offer check reads an offer against other buyers' clauses.
6. Keep your options open: DataVendor's terms make no listing exclusive, so other buyers remain open to you.

If you hold code or records and want them in front of every buyer whose rules take them, AIData.Broker introduces you free, paid by the buyers' own referral programmes, DataVendor's among them. For the wider market, see the buyers page, what labs buy and the guide to how AI companies get their training data.