A billion-dollar headline. A more specific conversation.

What this establishes—and what it leaves open

This is an announced commitment, not a report of $1 billion already spent. The source does not establish a price for a particular dataset, an allocation for an individual seller, or that any submitted asset will be accepted.

The company has a public Company Data Partnerships route. Current eligibility, permitted uses, onboarding, and commercial terms need to be discussed through that route.

Our reading: context may be the product

The practical question for a seller is whether its records preserve the context needed to understand a workflow. A final answer without the preceding state, permitted actions, or outcome can be much harder to evaluate as a learning asset.

This is our interpretation, not a statement of micro1’s unpublished procurement criteria. A supplier should ask which elements are actually needed before doing expensive preparation.

Five questions to take into a conversation

1. Which workflows, industries, and task types are currently relevant?
2. Does the buyer need historical records, annotated tasks, or a functioning interactive environment?
3. Which rights are required for training, evaluation, derived assets, and onward licensing?
4. Who handles privacy and confidentiality review, and how are rejected samples treated?
5. What evaluation, commercial terms, and decision process apply to this specific product?

What to prepare first

Create a short dataset card with a workflow description, available outcomes, coverage, known gaps, and the rights you can document. Include a schema rather than unreviewed production records. If a real sample is appropriate, settle its permitted use before sharing it.

Our guide to selling data to AI companies explains the distinction between training material, evaluation sets, and interactive environments. Use the dataset card builder to prepare an offer around your own asset.