AI labs are buying data from dead startups to train their models
Startup wind-down services have begun selling the digital remains of dead companies — code repositories, Slack and Teams histories, email archives, Notion and Figma workspaces — to AI labs that need training data. MIT Technology Review reports that SimpleClosure, a firm that helps founders shut down their businesses, added data sales to its services in April through a platform called Asset Hub. Its CEO, Dori Yona, says 47% of SimpleClosure deals now include the sale of code or data, and Asset Hub has handled over 150 deals. The company expects to return more than $10 million to startups by the end of 2026. Rival Sunset also runs a growing data resale business; its founder and CEO, Brendan Mahony, values a dying client's assets by head count, years in operation, data volume and how the team used its tools.
One seller, Keelan Miskell, wound down an anonymous-user crypto startup this year and netted roughly $25,000 for the code base plus a still-pending amount likely around $15,000 for workspace data held in Google, Notion, Figma and Slack. "At least all that stuff we were doing for two and a half years is worth something to someone somewhere," he told MIT Technology Review. The shift is part of a wider end-of-life ecosystem that has grown around failing startups — coaches, so-called death doulas, and firms handling the legal, tax and operational work of going under. Shutdowns have risen between 26% and 56% annually in recent years, according to estimates from Carta and AngelList.
Demand appeared suddenly last year as AI training labs hit the limits of public and synthetic datasets. Buyers include foundation model labs, vendors that construct simulated workspaces for AI training, and large tech companies building their own models. Surge AI buys archives to build reinforcement-learning gyms, environments where clients' AI agents practice tasks such as debugging a complex IT architecture or drafting appellate briefs. Nick Heiner, who leads that work, says he looks for archives reflecting 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 enough variety and depth. Real archives avoid errors common in synthetic worlds, such as references to dates that do not exist, though the size of the performance gain is not established: no company has released tests comparing training on real-world data with the alternatives. Pricing reflects that uncertainty, Heiner says, with buyers quoting widely different numbers because valuations rest on research theses about how many useful tasks can be extracted from, say, 100,000 lines of code.
Large deals are nonetheless happening. In August, as part of Spirit Airlines' bankruptcy, Google outbid the AI data company Mercor to buy the carrier's decades of data — including 100 million emails, 500 million Teams messages and 176,000 employee records — for $10 million. Bloomberg reported in September that SpaceX's AI unit is weighing buying data from struggling startups.
The legal picture is unsettled. Spirit's flight attendants' union objected to that sale on privacy grounds, and Greg Ewing, an attorney who advises AI startups, has called such data "legally radioactive." Suppliers employ data scientists to scrub personally identifiable information and exclude HR records, and buyers say they scrub as well. Ewing notes that emails and calendars can contain doctor appointments, school pickup arrangements and date-night plans. He characterizes the current period as a Wild West of AI training that is only starting to give way to licensing agreements and vetted use.
Who profits is a separate question. Proceeds typically flow to senior creditors and investors rather than employees. Ifeoma Ajunwa, a law professor at Emory University and author of The Quantified Worker, calls this a form of captured capital: legal precedent permits monitoring so workers can do their jobs, but that justification disappears once a firm is out of business. She argues employers should obtain express consent before selling workers' data, while noting no federal law bans such sales.
The market has practical limits too. Founders have little visibility into who buys their data or what buyers plan to do with it, and little assurance that offers are fair. Prices will likely stabilize in the coming year, though that assumes no AI bubble bursts first.
For developers and product teams, the mechanics matter more than the novelty. The archives on sale are the everyday artifacts of how a company worked: commit histories, code review comments, incident threads, design files and support tickets. Vendors pitch them as grounding for agents that must navigate messy, ambiguous situations rather than clean textbook tasks. This suggests a durable market for structured operational exhaust only where it is rich enough to teach judgment — a quiet Slack channel is worth little — which is why buyers weigh variety and depth over sheer volume.
A likely trade-off is that thorough scrubbing erodes exactly the context that makes archives valuable, since the ambiguities buyers prize often sit near personal details. Expect pressure toward narrower licensing, retention limits and provenance tracking rather than outright bans, because there is no federal prohibition to lean on. Teams that keep clean records and clear consent policies will find their archives easier to sell, and easier to price, than those that do not.
Why it matters: Founders shutting down now have a new line item on the balance sheet: their code and workplace data. Developers and product teams should assume that chat, email and repository history can become a saleable asset in a wind-down, which makes retention policies, data minimization and employee consent decisions that carry financial consequences. Inference: as more labs compete for scarce real-world archives, expect pricing benchmarks and standard licensing terms to appear within the next year.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
When Keelan Miskell began winding down his crypto startup this year, he didn’t think its data was worth much: the users were anonymous, the company just a few years old. But he’s since turned the firm’s digital archives into real money. He’s netted roughly $25,000 for his code base and a still-pending amount (likely around…