When Keelan Miskell began winding down his cryptocurrency startup, he assumed the data it had accumulated carried little commercial value. According to the account relayed by MIT Technology Review, that assumption proved wrong: the archives turned out to attract interest from an unexpected buyer category — AI labs constantly searching for new material to train their models.
The anecdote points to a broader pattern emerging among failing startups. When a tech company shuts down, its most visible assets — code, patents, brand — aren't always the ones that draw buyers. Datasets built up during ordinary operations, even modest ones or those from sectors unrelated to AI, can hold meaningful residual value for labs facing a shrinking pool of readily available training content.
This trend reflects a wider context in which large language models have already consumed much of the freely accessible text on the open web, pushing AI companies to look for alternative sources: specialized archives, transactional records, content generated by niche communities. The digital leftovers of defunct startups, often overlooked during liquidation, are becoming a new category of monetizable resource.
For founders winding down their companies, this shift opens an unexpected avenue for recovering some value, one that could reshape how liquidators and investors appraise the remaining assets of a failed business. It also raises questions about the provenance and consent attached to such data, which was typically collected for purposes far removed from the AI training it may ultimately serve.