Who buys a shut-down company's data? 36 AI practitioners vote to produce a list

A startup shuts down, and the data assets it leaves behind may still have buyers. AI companies need data to train models, and a shut-down company's data may include user behavior, code, conversations, and industry documents. The question is, who do you sell it to?

Chris Barber asked 36 AI practitioners on X one question: if you had a private dataset to sell, such as data from a shut-down company, who should you talk to?

The rules were straightforward: you could not vote for a company you work for or founded. Chris Barber did not vote. On the Frontier List page, he disclosed that he is a small investor in AfterQuery, Convergent, Fleet, Ionic, Nyne, Ooak Data, Patronus, SF Data, and Snorkel, and that he advises Zero Billion.

The results are as follows:

Companies with one vote include Appen, Eigen Research, Fleet, Ionic, OpenAI data team, Patronus, Tenkara, and Very Large Research.

The full list is on Frontier List: https://frontierlist.com/data-buyers

On this list, micro1 and Mercor are far ahead. Handshake and SF Data have 5 votes each, and Surge has 4. OpenAI's data team got only 1 vote. This result does not explain the whole market, but it at least reflects an intuition within a small circle: when selling private datasets, it is more common to go first to companies that specialize in data transactions and procurement than to go directly to AI labs.

There is more information beyond the list. The Frontier List page lists that some voters' own companies also buy data, including BenchFlow, CodeMerchant, Convergent, Proximal, Snorkel, and Zero Billion. The page also lists Aviro. After publication, the founders of SimpleClosure, SF Traces, Monterra, and Polyshares told Chris Barber that they also buy data.

Someone in the replies warned not to look only at the price range. micro1's public page has a track specifically for handling data from shut-down companies, with quotes of $100,000 to more than $2 million per approved data package, requiring 30 or more employees and English-language data. The process order is contract, data transfer, approval. For sellers, it is also important to ask clearly what happens if the files are rejected after handover. This is what the replier Muse said, not Chris Barber's original wording.

Grepped received one negative vote. It is the only one marked with an asterisk on the list. Someone in the replies joked that the person who cast the negative vote must have a very specific story.

There is also an image in the replies:

List sources: Chris Barber's X post https://x.com/chrisbarber/status/2115844044496752681 and Frontier List https://frontierlist.com/data-buyers .