AI labs buy failed startups’ data for model training

Failed startups are finding a new source of value in their code bases and workplace archives as AI companies look beyond public and synthetic datasets. SimpleClosure, a company that helps startups shut down, added data sales to its services in April through Asset Hub, which brokers sales of code and workspace data such as emails, chats, and documents.

The market is growing alongside a broader startup wind-down industry. Estimates from Carta and AngelList show shutdowns rising between 26% and 56% annually in recent years. SimpleClosure says Asset Hub has handled over 150 deals, and 47% of its transactions now include code or data sales. Rival Sunset is also building a data resale business, valuing assets by head count, years in operation, data volume, and how teams used their tools.

Buyers include foundation AI labs, companies building simulated workspaces for AI training, and mature tech firms developing their own models. Demand intensified last year as labs sought real archives that capture workplace complexity, including disagreements and judgment calls. In one larger deal tied to Spirit Airlines’ bankruptcy, Google bought data including 100 million emails, 500 million Teams messages, and 176,000 employee records.

The trade remains legally and ethically unsettled. Privacy advocates and attorneys warn that workplace archives can contain sensitive personal details, even after efforts to scrub personally identifiable information. Proceeds typically go to senior creditors and investors rather than employees, while founders may have limited visibility into buyers or how the data will be used.