AI Labs Turn to Defunct Startups for Training Data as Open-Web Sources Dwindle
A New Data Market Emerges
As the race to train large AI models intensifies, companies are increasingly looking beyond publicly scraped web data to secure training material. A notable trend has emerged: purchasing digital assets from defunct startups, including codebases, user interaction logs, and internal datasets.
Startup Archives as Commodities
Individual founders winding down their companies are finding unexpected value in data they assumed had little worth. One founder, Keelan Miskell, netted approximately $25,000 from selling his crypto startup's code base and digital archives to an AI company, with additional payments pending. The buyers included established AI labs seeking diverse data to improve model training.
Why This Market Is Growing
Several factors are driving this trend. Legal challenges around web scraping have made traditional data collection riskier. Additionally, many major platforms have tightened their data-sharing policies, limiting what AI companies can legally obtain. Startup archives represent a relatively untapped source of structured, domain-specific data that may not be available elsewhere.
Implications for the AI Industry
This shift highlights the increasing value placed on proprietary data as a competitive advantage in AI development. For startups, even those that failed commercially, their operational data may hold significant value in the AI training market. The development also raises questions about data provenance and how AI companies source their training material as traditional channels become more restricted.