The Afterlife of Failed Tech Startups: How AI Labs Are Buying Dead Company Data for Model Training
As venture capital tightens and startup failures rise, a fascinating and controversial new market has emerged in the tech industry: the monetization of dead company data for artificial intelligence training.
With major foundation labs hitting the limits of public and synthetic datasets, they are increasingly turning to the digital archives of defunct startups. This "operational exhaust"—including code bases, workspace chats, emails, and internal documents—is finding a lucrative second life as training fuel for the next generation of AI models.
The Rise of Tech's "Undertakers"
Recent data from Carta and AngelList highlights an annual increase in startup shutdowns ranging between 26% and 56%. To help founders navigate the legal, operational, and financial headaches of going under, a specialized end-of-life ecosystem has sprouted, featuring coaches, "death doulas," and specialized wind-down firms.
Companies like SimpleClosure and Sunset have pioneered data resale as part of their wind-down services:
- SimpleClosure: Launched its Asset Hub platform to help founders sell code and workspace data (Slack, Notion, Figma, Google archives). By late 2026, Asset Hub expects to return over $10 million to startups through more than 150 deals. Roughly 47% of SimpleClosure deals now incorporate code or data sales.
- Sunset: Valuing assets based on headcount, operating history, data volume, and internal tool usage, Sunset treats this operational exhaust much like wood shavings in a particle board factory—previously thought to be waste, but carrying immense hidden value.
Why AI Labs Want Defunct Startup Data
Foundation AI labs, mature tech companies, and specialized vendors are hungry for authentic human workflows. Companies like Surge AI purchase these archives to construct "reinforcement-learning gyms" where AI agents practice complex tasks, such as debugging IT architecture or drafting legal briefs.
Real-world archives offer distinct advantages over synthetic data:
| Data Source | Advantages | Potential Drawbacks | | :--- | :--- | :--- | | Synthetic Data | Clean, highly scalable, customizable | Lacks nuanced human friction, prone to logical/date errors | | Defunct Startup Archives | Reflects real disagreements, deep context, and complex judgment calls | Privacy concerns, highly variable quality and valuation |
According to industry experts, a single, highly nuanced Slack history capturing deep team debates can sometimes be worth more than another company's combined Slack, email, and calendar archives.
High-Stakes Deals and the "Wild West" of Privacy
The market is no longer limited to small startups. Major transactions are commanding staggering sums:
- Spirit Airlines Bankruptcy: Google outbid AI data company Mercor to acquire decades of corporate data—including 100 million emails, 500 million Teams messages, and 176,000 employee records—for $10 million.
- Corporate Exploration: Reports indicate that units like SpaceX's AI division are also weighing data acquisitions from struggling ventures.
However, this rapidly expanding market operates in a legal gray area. Privacy advocates and legal experts have raised alarms. Employee records often contain sensitive personal information, ranging from doctor appointments to family schedules. While suppliers and buyers claim to scrub personally identifiable information (PII) before training, critics argue that the practice remains legally radioactive.
Furthermore, questions persist regarding profit distribution. Proceeds from these sales typically flow to senior creditors and investors rather than the employees who actually generated the data. Without federal laws explicitly barring or regulating such sales without employee consent, tech's new undertakers will likely continue appraising and monetizing what remains of failed ventures.
在本站开通 Claude / GPT / Gemini / Grok 套餐
如果你正在寻找稳定、合规、可立即使用的 Claude、GPT、Gemini 或 Grok 接入方案，可以直接在本站完成开通：
新用户开通后即可获得可用额度，按量计费、随用随充。
