Guide · 4 min read

Exclusive vs non-exclusive licensing for AI training data

Most training data is licensed non-exclusively: the same footage can be licensed to several buyers. Exclusivity costs more and is usually only worth it for custom collections.

Non-exclusive

You receive the right to train on the data; the vendor may license the same data to others. This is the default for DataClue's in-stock datasets and is reflected in the published per-hour prices.

  • Lowest cost per hour
  • Fast delivery from stock
  • Best for pretraining at scale

Exclusive

The vendor agrees not to license the data to anyone else, either permanently or for a fixed term, sometimes limited to a field of use (for example, household robots). Exclusivity on in-stock data is quoted case by case and depends on whether the data has already been licensed.

  • Competitive moat on a specific domain
  • Usually paired with custom collection
  • Priced per project

Questions to settle in any licence

  • Can you sell or deploy models trained on the data commercially?
  • Can contractors (labelling vendors, cloud providers) access it?
  • What happens to the data and models when the licence term ends?
  • Who is the end buyer, and can the licence be assigned in an acquisition?

Find matching data in two minutes

Describe the volume and scenario you need — see what's in stock and request a quote.

Search datasets

Related datasets