“AI companies are purchasing digital archives from shuttered startups to source training data, with founders like Keelan Miskell earning around $25,000 for code bases and additional sums for other assets. This emerging market turns failed ventures into valuable data suppliers for large language model development. As high-quality training data becomes increasingly scarce, defunct startups represent an untapped and largely unregulated resource.”
Key Takeaways
- Keelan Miskell netted roughly $25,000 selling his shuttered crypto startup's code base to an AI lab.
- AI labs are actively sourcing training data from defunct startups as publicly available data grows scarce.
- Startup archives, even from anonymous user bases, hold value for training code and language models.
Dead startups are becoming unlikely goldmines as AI labs hunt for training data.
trending_upWhy It Matters
As the low-hanging fruit of internet data is exhausted, AI labs are turning to unconventional sources, raising serious questions about user consent and data privacy — even when users were anonymous. Founders winding down companies may not realise they are sitting on sellable assets, creating an informal and largely unregulated secondary market. Regulators and privacy advocates should watch whether data sold this way complies with original terms of service or applicable laws like GDPR. This trend could also incentivise founders to collect and retain more user data than necessary, anticipating a future sale.
FAQ
Is it legal to sell user data from a shuttered startup to an AI company?
It depends on the startup's original privacy policy and applicable regulations like GDPR or CCPA. If users consented to broad data-sharing terms, a sale may be permissible, but legal grey areas remain, particularly around repurposing data for AI training.
What kinds of startup data are AI labs most interested in buying?
Code bases are among the most sought-after assets, as seen with Miskell's $25,000 deal. Labs are also interested in user interaction logs, proprietary text datasets, and other digital archives that can improve model performance.
Why are AI companies turning to defunct startups for training data?
Publicly available internet data is increasingly exhausted or restricted, pushing labs to find alternative sources. Failed startups offer unique, unlicensed datasets that haven't already been scraped by competitors, making them attractive despite their origins.
