According to MIT Technology Review, a new niche of tech businesses is helping failing startups sell their digital remains to AI labs. SimpleClosure, one such broker, has handled over 150 deals and expects to return more than $10 million to startups by the end of 2026. Its CEO says 47% of deals now include code or data sales, and rival Sunset is expanding a similar resale business. Founders who once thought their archives were worthless are earning real money: one crypto startup founder netted roughly $25,000 for code and an expected $15,000 for workspace data from Google, Notion, Figma, and Slack.
AI training labs began seeking this data last year after hitting limits with public and synthetic datasets. Surge AI, for example, buys archives to build "reinforcement-learning gyms" where AI agents practice tasks like debugging or drafting legal briefs. Real-world archives avoid errors common in synthetic worlds, though no company has released tests comparing performance gains. Pricing remains uncertain, with buyers quoting widely different numbers based on research theses. Larger deals are happening too: in August, Google outbid AI data company Mercor to buy Spirit Airlines' decades of data—including 100 million emails and 500 million Teams messages—for $10 million during bankruptcy.
The practice raises legal and ethical concerns. Spirit's flight attendants' union objected on privacy grounds, and attorney Greg Ewing calls such data "legally radioactive" because emails and calendars can contain doctor appointments and personal plans. Suppliers say they scrub personally identifiable information, but Emory law professor Ifeoma Ajunwa notes there is no federal law banning the sale of workers' data and argues employers should get express consent. Proceeds typically go to senior creditors and investors, not employees, which Ajunwa describes as a form of "captured capital." Founders also have little visibility into who buys the data or whether they received a fair offer, though prices may stabilize in the coming year unless the AI bubble pops first. The source is a single MIT Technology Review report; no differing sources were available for comparison.{