AI Labs Are Buying Up Dead Startups' Data Archives. Defunct Companies' Records Are Worth Millions

A dead startup, live money

When Kilan Miskey shut down his crypto startup earlier this year, he didn't think its archives were worth anything. The users were anonymous, and the company had barely lasted a few years. But the archives turned into real money: roughly $25,000 for the source code and another roughly $15,000 (the final figure isn't settled yet) for the data from its workplace tools — emails, chats, and documents in Google, Notion, Figma, and Slack.

"At least everything we did for two and a half years is worth something to someone," Miskey says.

The deal was arranged by SimpleClosure, part of a new sector in tech: businesses that specialize in dead startups. Shutdowns are growing 26–56% annually, and a whole ecosystem has sprung up around them — coaches, "death doulas," firms that help with the legal, tax, and operational headaches of winding down. Now these tech undertakers are selling data to AI labs.

The commerce of dead data

SimpleClosure launched Asset Hub in April. The platform helps founders sell source code and workspace data — emails, chats, documents. Buyers include foundation AI labs, vendors building workplace simulations to train AI, and large tech companies developing their own models. Asset Hub has processed more than 150 deals and expects to return over $10 million to startups by the end of 2026. CEO Dori Yona says these transactions, once rare and "somewhat taboo," have become routine: 47% of SimpleClosure's deals now include the sale of code or data.

Competitor Sunset is also rapidly growing its data-resale business. Founder and CEO Brendan Mahoney values a dying company's assets based on headcount, years in operation, data volume, and how the team used its tools. His analogy: particleboard. "This operational exhaust data is like wood chips that everyone thought were worthless, but it's actually worth a huge amount."

Demand for dead companies' data surged suddenly last year, when AI labs hit the limits of open and synthetic datasets. Surge AI, for instance, buys data to build "reinforcement-learning gyms" — environments where AI agents practice tasks like debugging complex IT architecture or drafting appeals. Nick Heiner, who leads this work at Surge AI, looks for archives that reflect a range of situations — from disagreements to complex decisions requiring judgment. A single company's Slack history, if it has enough diversity and depth, can be worth more than another company's entire email, calendar, and document archive combined.

Real archives avoid common mistakes found in synthetic worlds — like references to dates that don't exist. But how much of a performance boost this actually delivers remains unclear. No company has published benchmarks comparing results from training on real data versus alternatives. Prices reflect that uncertainty: buyers quote wildly different numbers, because valuations are based on "research theses" estimating how many useful tasks, say, 100,000 lines of code might yield.

Big deals and legal clouds

But major deals are already happening. In August, as part of Spirit Airlines' bankruptcy, Google outbid AI company Mercor to buy the airline's multi-year archives — 100 million emails, 500 million Teams messages, 176,000 employee records. Price: $10 million. In September, it emerged that SpaceX's AI division is weighing whether to buy data from struggling startups.

Legally, all of this can be murky. Spirit Airlines' flight attendants' union objected to the deal on privacy grounds. Attorney Greg Ewing, who advises AI startups, calls this kind of data "legally radioactive." Sellers hire data scientists to strip out personal information and exclude HR records, and buyers say they scrub it too. But Ewing points to private details buried in emails and calendars: doctor's appointments, school pickups, dinner plans. He describes a "Wild West of AI training" that's only beginning to evolve toward proper licensing agreements and vetted use.

A second question: who actually profits? Proceeds typically go to senior creditors and investors, not employees. For Ifeoma Ajunwa, a law professor at Emory University and author of books on workplace ethics, this is a form of "trapped capital." Precedent allows monitoring that's meant to help workers. But once a company is dead, that justification evaporates. "Employers should get explicit consent before selling workers' data," Ajunwa says. But there's no federal law that bans such sales.

Many see this market as a chance for bankrupt businesses to recover some value. But the deals have limits. Founders rarely see who's buying or why. And there's no guarantee they're getting a fair offer. Data prices will likely stabilize over the next year — unless the AI bubble bursts first. Either way, the new tech undertakers will be standing by to appraise what's left behind.