Research · 18 Oct 2026
Company shutdowns as a source of AI training data: a register of 149 closures and what their records sell for
Preprint
Yuma Heymans, AIData.Broker
Abstract
When a software company closes, its code, tickets, documents and chat outlive it, and since 2026 AI data vendors and wind-down services have offered to buy them. This paper describes a public register of technology companies that are closing, sold or have gone dark, compiled each morning from public signals, and sets it beside the public record of what such companies' records sell for. As read on 18 October 2026 the register holds 149 companies: 69 listed as acquired, 64 as closing or closed, 11 in insolvency and 5 whose sites went dark. 132 are in the United States and 17 in the United Kingdom. Only 15 entries rest on a stated public record (a notice, a filing, a news report or an announcement); 134 rest on a startup directory's status confirmed by the company's site no longer working, and of those only 57 sites were last seen working in 2025 or 2026. Of the 11 insolvencies, 6 are UK administrations, 4 creditors' voluntary liquidations and 1 a US bankruptcy filing. None of the 149 appears as a seller in the compiled records of deals, prices or market events. Those records show sales by closing companies mostly in the tens of thousands of dollars: one wind-down service reported about $1 million across nearly 100 deals in its first year. The middle half of published offers for one company's records runs from $174,000 to $693,000. We describe the sources, the publication rule and the limits of both records.
Keywords: company closures, insolvency, startup shutdowns, AI training data, data licensing, corporate records, wind-down, data markets
Citing this paper
Yuma Heymans (2026). Company shutdowns as a source of AI training data: a register of 149 closures and what their records sell for. AIData.Broker, 18 Oct 2026. Licensed under CC BY 4.0.
Also available from Zenodo (DOI 10.5281/zenodo.23453396).
Every figure in the paper comes from the open datasets on the datasets page, where each row names its source.