AI companies have run out of public data. What happens now?
For years, AI companies trained their models on the public internet. That well has now run dry — the public web has been read, and the models need more. The next training data lives inside real companies: their documents, workflows, and project histories.
That data cannot be scraped. It has to be licensed, company by company, with the owner's consent. And the buyers are paying for it — AI lab micro1 committed $1 billion over 12 months to license enterprise operational data.
What buyers are actually looking for
- Standard operating procedures and internal documents
- Workflows, policies, and training materials
- Project and job histories, including CAD files for design and engineering firms
- CRM data, quality-assurance processes, and decision records
- Years of history across multiple systems — depth matters as much as size
What never qualifies
Personal information, customer details, and patient records are never part of a licensing deal. The data is operational only — how the business runs, not who its customers are.
Who qualifies
As a rule of thumb, buyers look for established businesses with 25 or more employees, at least three years of records, three or more software systems in use, and records kept mainly in English. Smaller firms in the right industries — law firms at 10 or more, medical practices and accounting firms at 15 or more — can qualify through other buyers.
The five questions every buyer asks
Before any offer, a buyer will want to know:
- What kind of data is it?
- How many records, and how many years back?
- Do you own it?
- Does it contain any personal information?
- Has it been licensed before?
If you can answer those five, the check goes quickly — and you keep ownership of your data throughout. The business pays nothing, and the buyer pays only if a licensing deal closes.
Curious whether your business qualifies? The free check takes about ten minutes, and there is no obligation.
Free Data Check →