AI labs short of fresh training material are buying the digital remains of failed startups. MIT Technology Review reports that wind-down firms have added data sales to their shutdown services: SimpleClosure, through its Asset Hub platform, has handled over 150 deals and expects to return over $10 million to startups by the end of 2026, with 47 percent of its deals now including code or data sales. Rival Sunset runs a similar resale business, valuing archives by headcount, years in operation and data volume.
The assets are code bases and workspace records: email, chat, documents and calendars. Buyers include foundation model labs, vendors building simulated workspaces where AI agents practice tasks, and larger technology companies training their own models. One founder cited in the report netted roughly $25,000 for a code base, with a further amount, likely around $15,000, pending for workspace data.
The largest disclosed example came in August, when Google outbid the AI data company Mercor in Spirit Airlines’ bankruptcy to buy the carrier’s data archive, including 100 million emails and 500 million Teams messages, for $10 million, as reported by Reuters and cited in the report. Bloomberg reported in September that SpaceX’s AI unit is weighing similar purchases from struggling startups.
The legal position is unsettled. Suppliers and buyers say they scrub personal information, but the report notes objections from Spirit’s flight attendants’ union on privacy grounds, and legal scholars argue employees never consented to their workplace data being sold after a shutdown. Proceeds typically flow to senior creditors and investors. No published test yet shows how much real archives improve model performance over synthetic alternatives, so pricing remains thesis-driven.