Sakuya journal · Oct 22, 2026 · 1 min read

The five kinds of buyer for operational data

Frontier labs are only one part of the demand. Here is who else is buying.

When people hear that AI companies buy data, they usually picture one or two famous labs. In practice, demand for company records comes from five distinct groups, and each wants something slightly different.

Frontier labs

The companies training the largest models want records of real work that are not available publicly, with enough history to show outcomes.

Hyperscalers

The large cloud providers train models of their own and run the infrastructure other labs depend on. They look for the same depth of operating history.

Well‑funded new labs

A growing group of labs has raised enough capital to compete with the frontier. They buy to close the gap quickly, and they are often less focused on famous names than on useful records.

Agent training environments

Startups that build simulated workplaces, where AI agents practice real jobs, need to see how a support desk or a sales pipeline actually runs, system by system.

Data vendors

Companies that supply training sets and expert review to everyone above also acquire records to build their offerings.

Sakuya's job is to know which of these buyers fits your data and to get you in front of them.