Artificial intelligence companies are paying defunct startups millions of dollars for their digital archives—emails, code, chats, and documents—to train AI models, according to a report published October 13 in MIT Technology Review. The practice emerged suddenly last year as AI training labs exhausted public and synthetic datasets, transforming what was once considered worthless digital debris into a lucrative commodity. SimpleClosure, a firm that helps startups wind down operations, has handled more than 150 data deals through its Asset Hub platform and expects to return over $10 million to failing companies by the end of 2026.

The transactions span a wide range. One crypto startup founder, Keelan Miskell, collected roughly $25,000 for his code base and an anticipated $15,000 for workspace data stored in Google, Notion, Figma, and Slack after shutting down this year. The biggest deal came in August, when Google paid $10 million as part of Spirit Airlines' bankruptcy to acquire decades of the carrier's data, including 100 million emails, 500 million Teams messages, and 176,000 employee records. SimpleClosure CEO Dori Yona reports that 47% of the company's shutdown deals now include code or data sales, compared to earlier years when such transactions were rare and "somewhat taboo." Startup shutdowns have climbed between 26% and 56% annually in recent years, according to estimates from Carta and AngelList, fueling the growth of what the report calls an "end-of-life ecosystem" that includes coaches, "death doulas," and firms handling the legal and tax complexities of going under.

Buyers determine value based on head count, years in operation, data volume, and how teams used their tools, according to Brendan Mahony, CEO of SimpleClosure competitor Sunset. He compares the data to particle board: "This operational exhaust data is like the wood shavings that everyone thought had no value, but it actually has a ton." Nick Heiner, who leads work on AI training environments at Surge AI, a data buyer, says his company seeks archives that reflect a range of situations, from disagreements to tricky judgment calls, and that one company's Slack history can be worth more than another's Slack, email, and calendars combined if it captures conversations with variety and depth. The report notes that pricing varies wildly because valuations rest on "research theses" estimating how many useful tasks emerge from, say, 100,000 lines of code, and no company has yet released tests comparing performance gains from real-world data training with alternatives.

The legal landscape remains murky. Spirit Airlines' flight attendants' union objected to the data sale on privacy grounds, and attorney Greg Ewing, who advises AI startups, has labeled such data "legally radioactive." While data suppliers deploy data scientists to remove personally identifiable information and exclude HR records, and buyers say they scrub data too, Ewing points to private details hidden in emails and calendars: doctor appointments, school pickup arrangements, date-night plans. Ifeoma Ajunwa, a law professor at Emory University and author of The Quantified Worker, argues the practice represents "captured capital" because proceeds typically flow to senior creditors and investors rather than employees, and she believes employers should obtain express consent before selling workers' data, though no federal law currently bans such sales. The report describes a "Wild West of AI training" that's only beginning to shift toward proper licensing agreements and vetted use.

Data prices will likely stabilize in the coming year, the report concludes, unless an AI bubble bursts first. Founders currently have little to no visibility into who buys their data and what buyers plan to do with it, and there's minimal assurance they're receiving fair offers. Either way, tech's new undertakers will be there to value what remains. The emerging market for startup data highlights how quickly ethical norms can lag behind technological incentives, and whether transparency requirements will follow may determine if this becomes a legitimate secondary market or a cautionary tale about unchecked data commodification.