What Frontier Labs Actually Buy: Inside Handshake AI's Data Operation

Three years ago, the market for frontier training data was a billion dollars. It is now a $10 billion market. "Three years from now," says Handshake AI CEO Garrett, "I don't see how it's not a $100 billion market." The reason is that data has become the primary bottleneck for model progress, and the nature of that data has transformed completely—from college students drawing boxes around cars at $10 an hour to domain experts authoring entire simulated worlds.

The Shift from Labeling to Authoring

When large language models first emerged in late 2022, they produced gibberish, meaning a high school student could easily train them at scale through RLHF. That era is over. Models can now produce coherent language, reason through problems, and call tools. The bottleneck has shifted from volume to expertise and complexity.