AI Labs Are Training on Data from Dead Startups

Code and workspace data from shuttered startups is being sold as training material for AI. As the competition over AI performance intensifies, the data of companies that have closed their doors has become a new commodity for sale. According to estimates from Carta and AngelList, the number of startup shutdowns has risen in recent years, from 26% to as much as 56% annually. Amid this trend, an ecosystem has emerged to help companies wind down—liquidation coaches and 'death dungeons' among them—and as part of that, deals are being struck to hand over defunct companies' digital assets to AI labs.

Broker in these deals is SimpleClosure, a firm specializing in wind-downs. In April of this year, the company added a data sales service and now operates as an Asset Hub platform, supporting the sale of code, emails, chats, documents, and more. Buyers range from frontier AI labs and builders of simulated workspaces to mature tech companies developing their own models.

The Asset Hub has already processed more than 150 transactions and expects to return over $10 million to startups by the end of 2026. Dori Yona, CEO of SimpleClosure, said 47% of all transactions include the sale of code or data.

Demand has come from AI training labs that hit the limits of public and synthetic data last year. Surge AI buys data to build 'reinforcement-learning gyms' where clients' AI agents can practice tasks like debugging complex IT architectures or drafting appellate briefs. Nick Heiner, who leads this work, said he looks for archives containing a wide range of situations, from disagreements to difficult judgments, noting that a single company's Slack history, if voluminous enough, can be worth more than another company's Slack, email, and calendar combined.

The scale and flow of these deals are being closely watched to see whether the sale of defunct startups' data is solidifying into a market of its own.