Who buys company data for AI training?
Last updated
Short answer: AI labs mostly buy company data through data vendors such as micro1, Scale AI, Handshake AI and Turing, through data marketplaces, and through wind-down platforms for closed startups. Each sets its own eligibility and payment terms. micro1, for example, publishes $100K to $2M+ per approved data package, subject to review.
Several kinds of buyer pay for company data, and they work very differently. Some license private operational records to train AI agents; others resell datasets to many customers; others help closed companies realise their remaining assets. This guide compares them so you can see which route fits your company.
Cadabra is an independent referral partner of micro1 and may earn a referral fee. Terms for the other companies named here are set by them; check their published information before applying.
Buyer types compared
| Type | Examples | Typical eligibility | How they pay | Best for |
|---|---|---|---|---|
| AI data vendors | micro1, Scale AI, Handshake AI, Turing | Set by each vendor. micro1 publishes 30+ full-time employees, mature operations, substantial history and mostly English records. | A licence fee for an approved, de-identified data package, on the vendor's own terms. micro1 publishes $100K to $2M+ per approved package. | Operating companies with years of connected internal records |
| Data marketplaces | Datarade, AWS Data Exchange, Snowflake Marketplace | Usually open listing; the buyer decides whether the data fits. | Per-dataset sales or subscriptions agreed with each buyer. | Companies with structured, repeatable datasets they can keep updating |
| Wind-down platforms | Startup wind-down and closure service providers | Companies that have closed or are winding down, with someone authorised to act for the company. | Proceeds from selling or licensing assets, including data, as part of the wind-down. | Closed startups that want one process covering assets and data |
How to choose
If your company is operating and has years of history across chat, docs, tickets and code, a data vendor is usually the most direct route, because AI labs want connected records of real work. If you produce a structured dataset you can refresh, a marketplace may suit you better. If the company has closed, involve whoever controls its assets first, then compare a wind-down provider with a vendor's closed-company track.
Whichever route you take, agree the scope in writing, de-identify personal data and check contracts first. See privacy and consent when licensing company data and published AI training data prices. Figures above are published ranges, not promised payouts.