AI Labs Buy Data From Dead Startups to Train Models

Overview

A growing niche in the technology sector is emerging around the liquidation of failed startups, with a new twist: their digital data is being sold to artificial intelligence developers. Keelan Miskell, founder of a crypto startup that wound down this year, discovered that his company’s abandoned digital archives held significant value. What he initially believed was negligible data has since netted him roughly $25,000, with an additional $15,000 pending for workspace files stored in tools like Slack, Notion, and Figma.

This trend reflects a broader shift where tech "undertakers" are monetizing the operational exhaust of dying companies. As shutdowns rise, an end-of-life ecosystem has formed to help founders navigate the legal and tax headaches of going out of business. Now, firms within this sector are brokering deals that sell codebases and communication histories to AI labs hungry for high-quality training data.

The Rise of the Data Brokers

Companies like SimpleClosure have entered the market specifically to handle the final stages of a startup’s life cycle. Originally focused on closure logistics, they added data sales to their services in April. Their Asset Hub platform allows founders to list code and workspace data for sale to a variety of buyers, including foundation AI labs, vendors that build simulated workspaces, and mature tech companies developing proprietary models.

"At least all that stuff we were doing for two and a half years is worth something to someone somewhere." — Keelan Miskell, former crypto startup founder

The market for these assets is expanding rapidly. SimpleClosure reports handling over 150 deals and expects to return more than $10 million to startups by the end of 2026. CEO Dori Yona notes that transactions once considered rare or taboo have become routine. Currently, 47% of SimpleClosure deals include the sale of code or data, signaling a major shift in how founders view their final assets.

Industry rival Sunset is also capitalizing on this demand. Founder Brendan Mahony explains that their valuation process considers head count, years of operation, data volume, and tool usage patterns. He compares the operational data to wood shavings from particle board production: once considered worthless waste, it is now recognized as having substantial value.

Why AI Labs Are Buying

The surge in demand for defunct companies’ data is a direct response to the limitations facing current AI training methodologies. Major labs have hit the ceiling of available public datasets and synthetic data, which often lack the nuance and complexity required for advanced reasoning. Synthetic data, for instance, frequently contains errors such as references to dates that do not exist, creating a distorted view of reality for AI agents.

Real archives offer a solution by providing authentic human interactions. Nick Heiner, who leads environment creation at Surge AI, seeks data archives that reflect a wide range of situations, including disagreements and tricky ethical judgment calls. He argues that a single company’s Slack history, if it captures diverse and deep conversations, can be more valuable than another company’s combined emails, calendars, and chats.

Prices for these assets vary wildly because valuations are currently based on "research theses" rather than standardized metrics. Buyers estimate how many useful training tasks can be derived from a specific volume of code or messages. For example, a dataset containing 100,000 lines of code might be priced based on estimates of its utility for debugging complex IT architectures. No company has yet released public tests comparing performance gains from real-world data versus synthetic alternatives, leaving the market in a state of pricing uncertainty.

High-Profile Sales and Legal Gray Areas

Despite the lack of standardized valuation, major deals are already taking place. In August, Google outbid AI data firm Mercor to purchase Spirit Airlines’ data as part of the carrier’s bankruptcy proceedings. The package included 100 million emails, 500 million Teams messages, and 176,000 employee records, with a reported price tag of $10 million. Additionally, Bloomberg reported in September that SpaceX’s AI unit is actively considering purchasing data from struggling startups.

"This operational exhaust data is like the wood shavings that everyone thought had no value, but it actually has a ton." — Brendan Mahony, CEO of Sunset

The legal implications of these transactions are significant and potentially contentious. Greg Ewing, an attorney advising AI startups, has described such data as "legally radioactive." While suppliers employ data scientists to scrub personally identifiable information and exclude sensitive HR records, private details often remain hidden in plain sight. Emails and calendars may contain doctor appointment times, school pickup schedules, and personal plans that were never intended for public or commercial consumption.

Spirit Airlines’ flight attendants’ union formally objected to the sale on privacy grounds, highlighting the tension between corporate asset liquidation and worker privacy. Ewing describes the current landscape as a "Wild West of AI training," suggesting that proper licensing agreements and vetted use policies are only beginning to emerge.

Equity and Future Outlook

Another point of contention is who benefits from these sales. Proceeds typically flow to senior creditors and investors, leaving regular employees without compensation for the value generated by their work. Ifeoma Ajunwa, a law professor at Emory University and author of The Quantified Worker, describes this dynamic as "captured capital." She argues that while employers may have justified monitoring rights while a company is active, that justification vanishes once the firm ceases to operate.

Ajunwa advocates for express consent from workers before their data is sold, noting that no federal law currently bans such sales. This legal gap leaves many workers vulnerable as their digital footprints are commodified without their knowledge or input.

Founders also face challenges regarding transparency. Many have little visibility into who purchases their data or how it will be used. Furthermore, there is limited assurance that they are receiving fair market value for their assets. While data prices are expected to stabilize in the coming year, the market remains volatile. If an AI bubble were to pop, the demand for these archival assets could contract sharply. Regardless of market fluctuations, the ecosystem of tech support companies will likely remain to appraise whatever value is left in failing businesses.