Dead startups' data finds buyers in AI labs

When Keelan Miskell began winding down his crypto startup this year, he assumed the accumulated data wasn't worth much — anonymous users, a company just two and a half years old. In practice, he has already collected roughly $25 thousand for the code, and another $15 thousand is on the table for the work history stored in Google, Notion, Figma and Slack. "At least all the things we did for two and a half years are worth something to someone somewhere," he says. The deal was brokered by SimpleClosure, part of a whole ecosystem that has sprung up around dying startups.

An ecosystem of closures

According to estimates from Carta and AngelList, startup shutdowns have jumped between 26% and 56% year over year in recent years, and a whole "end-of-life" industry has been born alongside them — coaches, "death doulas" and firms that handle the legal, tax and operational mess of dissolution. SimpleClosure added data sales to its services in April, and its Asset Hub platform lets founders sell code and workspace data — emails, chats, documents. The buyers: foundation-model labs, vendors building simulated environments for training, and mature companies developing their own models. More than 150 deals have been processed so far, with a projection of over $10 million returned to startups by the end of 2026. Chief executive Dori Yona says such deals, once rare and "a bit taboo," have become routine; 47% of SimpleClosure's transactions now include code or data sales.

How to value operational exhaust

There is competition. Industry rival Sunset runs a fast-growing data-sales business. Chief executive Brendan Mahony explains that valuation hinges on headcount, years active, data volume and how the team used its tools. His analogy: "particle board." "This operational exhaust data is like wood chips everyone thought had no value, but actually they have a ton." Demand for dead-company archives exploded last year when training labs hit a ceiling on public and synthetic data. Surge AI, for instance, buys data to build "reinforcement-learning gyms" where AI agents practice tasks — say, debugging a complex IT architecture or drafting legal appeals. Nick Heiner, who leads that work, looks for archives that reflect a range of states, disputes, complex judgment calls. A single Slack history with depth and variety, he says, can be worth more than another company's Slack, email and calendar combined.

The synthetic problem and the unsettled price

Real archives avoid common synthetic-data pitfalls, such as references to dates that don't exist, but the magnitude of any performance gain remains unclear. No company has yet published benchmarks comparing training on real data versus the alternatives. Pricing reflects the uncertainty: buyers offer wildly different numbers, Heiner says, because valuations rest on "research theses" that try to guess how many useful tasks might emerge from, say, 100 thousand lines of code.

The big deals and the legal risk

That hasn't stopped large transactions. In August, as part of the Spirit Airlines bankruptcy, Google outbid AI data company Mercor to buy decades of data — 100 million emails, 500 million Teams messages, 176 thousand employee records — for $10 million. In September Bloomberg reported that SpaceX's AI unit is considering buying data from distressed startups. Legally, the business can get messy. The Spirit flight attendants' union opposed the sale on privacy grounds, and attorney Greg Ewing, who advises AI startups, calls such data "legally radioactive." Vendors employ data scientists to scrub personally identifiable information and exclude HR records, and buyers say they clean the data too. But Ewing points to the private details that hide in emails and calendars.