Can you sell a dataset on Kaggle or Hugging Face?

Not as a sale on the platform. Both sites are built for sharing. Kaggle's dataset metadata, as its API documentation lists it, covers a title, description, licence, files, keywords and update frequency. None is a price or an access request.

Hugging Face is similar: its terms cover fees for its own paid services, not a way for users to charge each other. What it offers is gating, a switch that makes people ask before they can download.

So the hubs are shop windows, not tills: money changes hands under a separate contract. A public sample proves the data exists, and a gated repo can hand the full set to buyers you approve.

The table sets the three ways to publish side by side, from each platform's own pages.

Because the licences a public upload carries are built to last, a point most guides to selling datasets skip.

The Creative Commons FAQ says its licences are not revocable once granted unless someone breaches them. A licensor can stop distributing, but anyone who already has a copy may keep using and sharing it. Some licensors charge for initial access, the FAQ notes, so a price only buys the first copy.

Hugging Face adds a layer: its terms say a public repository grants each user a perpetual, irrevocable, worldwide, royalty-free, non-exclusive licence to use, reproduce and distribute the content through its services. You still own the data, but you can't sell exclusive rights to something everyone already holds a licence to.

As a delivery channel for a deal agreed elsewhere. Requesters share their username and email unless you switch that off, answer any custom fields you add, and wait for approval in manual mode.

1. 1

   Agree the licence first

   Price, permitted uses and term go in a written licence signed before any access. The gate enforces nothing about payment.

2. 2

   Turn on manual approval

   In the dataset settings, enable access requests and choose manual approval.

3. 3

   Ask for what you need at the gate

   Add fields such as company name and a checkbox accepting the licence. Hugging Face says you are responsible for personal data collected this way.

4. 4

   Grant the buyer's account

   Approve the request, or grant access through the settings page or the API.

5. 5

   Close access when the term ends

   Reject or reset the user. Files already downloaded are governed by your contract, not the gate.

Access is granted to individual users, not whole organisations, so each person on a buyer's team needs approval.

Enough rows to show schema and quality, and nothing you'd mind every user keeping forever. A 2026 Hugging Face forum thread on selling datasets describes the trust problem: buyers want proof of label accuracy and no duplicates, but a full audit exposes the data. Replies suggest a representative, seeded sample, not hand-picked best rows.

Before posting, scan it. A PII scanner searches text for personally identifiable information, such as email addresses, phone numbers and government ID numbers, so it can be removed or protected before the text is shared. We ran ours on four fictional support tickets.

It found four matches: an email, a phone number, an IP address and a card number that passes the Luhn check. It ignored a parcel tracking number. It did not flag the two names or the street address, as the tool's own page warns: they have no fixed format. Scan, redact, then read every row.

On a marketplace with billing, or in a direct deal. AWS Data Exchange takes a 3% listing fee on public data offers, and private offers pay 3% under $1 million of contract value, 2% up to $10 million and 1.5% above that, per AWS's fee page effective January 5, 2024. Sellers must be registered, qualified and based in an eligible jurisdiction.

Snowflake Marketplace allows paid listings for providers billed in one of 20 listed countries, paid out through Stripe Express. Our guide to where to sell data compares these with listing sites and brokers. For one-off data that AI teams want, such as a codebase or recordings, a direct deal under a written licence is the usual route.

Platform terms change; we read them on one day. Our Kaggle facts come from its official API documentation, not its web interface. Whether you may sell a given dataset at all depends on how it was collected, the privacy promises made to the people in it, and contracts we can't see.