Robotics deployments are now about data versus territory, not just flashy demos

broodsugar · x · 2026-07-23

Robots are moving from demos to deployment, but the business model is still unsettled

This long post argues that after months of viral demos, robots are finally starting to enter productive environments. The author frames the early market as a split between two strategies:

The post then digs into the economics of deployment. It cites Ken Goldberg’s estimate that robotics is still missing 100,000 years of training data, but also points out that the enthusiasm for data collection has cooled. Kyle Vedder argues that once a robot is good enough to deploy, additional site-specific data may have diminishing returns. Animesh Garg is cited for a similar take in his essay Moneyball for Physical AI, which tries to quantify the value of each hour of collected data.

The core question is how to judge whether a deployment is worth it: are you optimizing for data, territory, or long-term robot business viability?

Original post →

More from Embodied

Embodied channel →