The Graveyard Has an Asset Manager Now
When Keelan Miskell shut down his crypto startup this year, he assumed the company’s digital records were worthless. He was wrong – and $40,000 richer for finding out.
A Market Built on Operational Wreckage
Miskell netted roughly $25,000 for his company’s code base, with another amount – likely around $15,000 – still pending for workspace data pulled from 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 said. His sale was brokered through SimpleClosure, a startup-shutdown services firm that added data sales to its platform in April 2026.
SimpleClosure is one node in a growing ecosystem built entirely around corporate death. Shutdown rates have climbed sharply: estimates from Carta and AngelList put annual closures rising between 26% and 56% in recent years. The resulting market – which includes coaches, legal advisors, tax consultants, and what some in the industry call “death doulas” – now has a data-sales arm grafted onto it. Dying startups have something AI labs want, and a new class of intermediaries has positioned itself to make the match.
SimpleClosure’s Asset Hub platform connects founders with three categories of buyers: foundation AI labs, vendors that build simulated work environments for AI training, and mature tech companies developing internal models. The platform has facilitated over 150 deals and expects to return more than $10 million to founders by the end of 2026. CEO Dori Yona has watched transactions that were once rare and “somewhat taboo” become standard operating procedure – 47% of SimpleClosure’s deals now include a code or data sale component.
Industry rival Sunset runs a parallel operation. Founder and CEO Brendan Mahony said his firm assesses a dying client’s data value based on head count, years in operation, data volume, and how actively the team used its tools. His analogy: “This operational exhaust data is like the wood shavings that everyone thought had no value, but it actually has a ton.”
Why AI Labs Are Shopping at the Morgue
Demand for defunct companies’ data surfaced sharply in 2025, when AI training labs began running up against the limits of public datasets and synthetic data generation. The problem with synthetic data is that it introduces errors that real-world archives don’t – references to nonexistent dates, for example, or conversational patterns that don’t reflect how people actually work through difficult problems. Real workplace communications offer something synthetic environments can’t manufacture: the texture of genuine disagreement, judgment under pressure, and the kind of context that only emerges over years of operational history.
Surge AI is one buyer actively working in this space. The company purchases data archives to build what it calls “reinforcement-learning gyms” – training environments where AI agents practice complex tasks like debugging intricate IT architectures or drafting appellate briefs. Nick Heiner, who leads Surge AI’s environment-building work, looks specifically for data reflecting a wide range of situations. He says one company’s Slack archive – if it captures varied, substantive conversations – can be worth more than another company’s entire suite of Slack, email, and calendar data combined. Depth matters more than breadth.
Pricing in this market remains volatile because there’s no established method for valuing the data. Heiner says buyers quote wildly different numbers because valuations rest on “research theses” – essentially educated guesses about how many useful training tasks can be extracted from, say, 100,000 lines of code. Whether real-world operational data actually produces better model performance than alternatives remains an open question. No company has published comparative test results.
Despite the uncertainty, large transactions are moving forward. In August 2026, as part of Spirit Airlines’ bankruptcy proceedings, Google outbid AI data company Mercor to acquire the carrier’s data holdings: 100 million emails, 500 million Microsoft Teams messages, and 176,000 employee records. The purchase price was $10 million. That same month, Bloomberg reported that SpaceX’s AI unit was weighing acquisitions of data from struggling startups, signaling that the practice is attracting interest well beyond the startup-services niche.
The scale of Google’s Spirit Airlines purchase illustrates why defunct companies with long operational histories are particularly attractive. A decades-old airline generates a volume and variety of internal communication that a two-year-old crypto startup simply cannot match – which is why Miskell’s $40,000 haul, while real money for a founder winding down, is a fraction of what richer archives command. As AI agents take on increasingly complex tasks, the training data behind them – where it comes from, and what it actually contains – carries consequences that extend well beyond model performance metrics.
Legally Radioactive, Practically Lucrative
The legal exposure surrounding this market is substantial. Spirit Airlines’ flight attendants’ union formally objected to the Google data sale on privacy grounds. Greg Ewing, an attorney who advises AI startups, has described this category of data as “legally radioactive” – and for specific reasons. Sellers employ data scientists to scrub personally identifiable information and exclude HR records before any transfer. Buyers say they run their own scrubbing processes. But Ewing points to what survives even careful cleaning: references to doctor appointments, school pickup logistics, and personal scheduling details embedded in emails and calendar entries.
SimpleClosure and Sunset both emphasize compliance as a selling point, aware that the legal framework governing this kind of transaction is still forming. There is no settled standard for what constitutes adequate anonymization of workplace communications, no regulatory body that has specifically addressed the sale of bankrupt companies’ internal data to AI labs, and no litigation that has yet tested where liability lands when something scrubbed turns out not to be scrubbed well enough. The question isn’t whether a lawsuit eventually targets one of these transactions – it’s which one breaks first and how much it rewrites the economics for everyone else.
