AI labs are running out of fresh public data, and synthetic alternatives have distinct limits. To keep training frontier models, developers are turning to the digital graveyards of failed tech companies. With startup shutdowns rising between 26% and 56% annually in recent years, according to data from Carta and AngelList, a secondary market has emerged for the digital exhaust left behind: proprietary codebases, internal messaging logs, and project workflows.
Intermediaries are turning liquidation archives into training commodities. SimpleClosure launched its Asset Hub platform to help shuttered companies sell off software code and workspace data, handling over 150 transactions and projecting more than $10 million in returns to founders. According to SimpleClosure CEO Dori Yona, 47% of their wind-down deals now feature code or data sales. Rival liquidation service Sunset also runs a data resale business, pricing assets based on historical headcount, operational lifespan, data volume, and internal tool usage patterns.
For failed founders, this market provides liquidity from assets that used to be worthless. Keelan Miskell, who shut down a crypto venture, secured roughly $25,000 for his company's codebase, with an additional payout pending for workspace archives across Google, Notion, Figma, and Slack. The buyer pool spans major AI labs, mature enterprises training internal models, and specialized vendors building simulated workspaces.
Surge AI acquires workplace archives to construct reinforcement-learning environments where AI agents train on tasks like complex IT debugging and appellate brief drafting. Real workplace history eliminates synthetic data flaws, such as fictional calendar dates. As Surge AI's Nick Heiner points out, a single company's Slack history can be worth more than another's combined Slack, email, and calendar records if the dialogue offers sufficient depth and variety.
Valuations, however, remain opaque. Buyers offer wildly diverging bids based on internal research hypotheses about how many productive tasks can be extracted from a specific dataset. Formal benchmarks measuring real-world workplace data against alternatives have not been published, yet deals continue to close.
These liquidation sales carry legal and privacy complications. Despite intermediaries employing data teams to strip personally identifiable information and exclude HR records, internal archives routinely contain private employee details, ranging from medical appointments to family schedules.
The economics also raise questions for the workforce. Proceeds from data sales bypass former employees and flow directly to senior creditors and investors. Emory University law professor Ifeoma Ajunwa points out that while workplace monitoring is legally defended during active operations to assist productivity, that justification collapses once a business dissolves. While Ajunwa argues that explicit consent should be required before selling employee records, no federal law currently restricts liquidating workplace communications.