Mercor, data brokers circle dying startups for AI training data
AI labs' hunger for real company data has created a new kind of distressed-asset buyer: data brokers who swoop in when startups shut down or get acquired, offering to buy or license the internal datasets that used to die with the company.
Data firms that supply AI labs are driving a growing market for the internal datasets of startups that are shutting down or being acquired, according to The Information. Mercor and other companies that gather training data for OpenAI, Anthropic and their rivals are now actively courting founders on their way out the door, the report says — a pitch that used to be rare and is now routine. The report opens with the case of Warmly, an AI agent startup that agreed to be acquired by HubSpot in late June: eight days after the deal, CEO Maximus Greenwald found an unusual email in his inbox, the kind of unsolicited offer for internal data that brokers now send as a matter of course.
The trend is a direct consequence of what frontier labs actually need next. Public web text and synthetic data can't teach a model how a company really operates; labs increasingly want Slack threads, support tickets, sales records and workflow traces to train agents that work inside enterprises. The Information's reporter made exactly that point three days earlier, noting that OpenAI and Anthropic's demand has turned startups' Slack threads into prized assets. Mercor has built its $10 billion business on that same insight — TechCrunch reported last fall that the company pays experts up to $200 an hour to produce and license proprietary data, paying out more than $1.5 million a day to its contractor network.
It's a strange afterlife for startup data. A company's internal corpus was never collected with model training in mind — it contains customer conversations, employee chatter and unreleased product details — and when a startup dies, someone else now decides what happens to it. The timing is also awkward for Mercor specifically: the company spent spring and summer rebuilding trust after confirming a major security breach in April that prompted Meta to pause work with it. Aggregating the most sensitive data of defunct companies is a growth strategy, but also an invitation for the next regulatory fight over who really owns what a startup's users and employees said.
What to watch: whether data-buying deals start appearing in shutdown announcements and acquisition filings — and whether customers of dying startups get any say.
If your startup ever got cold-emailed about its data, we'd love to hear how the pitch went. Tell us in the comments.
Sources: The Information · Techmeme · TechCrunch
