Major AI laboratories are acquiring training data from defunct startups to improve their models, a practice exemplified by the case of Keelan Miskell, whose cryptocurrency startup was being wound down. Miskell initially assumed the company's accumulated data had minimal value, but discovered that AI development firms were interested in purchasing the dataset. This trend reveals a secondary market emerging around data assets from failed ventures—companies that accumulated valuable user interaction logs, transaction records, or domain-specific datasets during their operations now represent a revenue opportunity for liquidating founders.

The practice highlights how AI model development has become data-constrained, with labs actively seeking alternative sources of training material beyond web scraping and public datasets. For startups, even failed ones, accumulated data can represent residual asset value.

What This Means for Your Business

Startup founders should evaluate whether accumulated user data or transaction records represent monetizable assets during wind-down processes. If your startup has collected domain-specific datasets, consulting with data brokers or AI development firms could offset closure costs. Conversely, if your organization uses third-party AI services, be aware that training data origins may include data from acquired defunct companies, potentially introducing legal or privacy risks depending on original user consent. Request transparency from AI vendors about training data provenance, particularly for models used in regulated industries.