Who Owns the Hours Inside the Robot's Mind?

In this episode we argue that the binding constraint on robotic intelligence has moved. It is no longer compute, and it was never the model architecture. It is the supply of human-generated demonstration data, the recorded hours of people driving real robots through real tasks. Unlike text or images, this data cannot be scraped from the internet. It has to be manufactured, one time-aligned teleoperation episode at a time, on instrumented hardware, at considerable cost.

Because the performance of vision-language-action models scales directly with the volume and diversity of these recorded hours, a durable advantage is forming for the few organizations wealthy enough to fund collection at scale. We walk through the economics of producing an hour of demonstration data, the skilled and largely invisible human labor inside that hour, and the deployment flywheels that let incumbents compound their lead.

What happens when the mind of every robot is distilled from a handful of private data silos?