Samsung's real robot advantage isn't the hardware

Samsung launched a dedicated robotics unit — RX, short for Robotics eXperience — on July 21. But the organizational move is not the point. The company's factories double as the robot training ground. Every task a Samsung robot completes on a real production line becomes data that makes the next robot better.

The AI at the centre is Shallow-π, a model built by Samsung Research. It runs on small, cheap edge computers and takes roughly ten decisions per second using about half the computing power of older systems. In tests, it hit a 95% success rate inserting a cable into a socket with sub-millimetre precision. This works because it is lean AI that runs without a data centre — something you can actually put in a factory today.

Samsung is also spending seriously. Nineteen trillion won — roughly $14 billion — is earmarked just for a new physical AI hub in Gumi, South Korea. Lee Dongkun, who previously oversaw Boston Dynamics during its time at Hyundai, will lead the new division. That hire signals Samsung wants credibility, not headlines. Samsung shares jumped nearly 7% on the day of the announcement — markets see the data moat, even if the headlines focus on the robot.

For anyone building in or around manufacturing, the lesson is uncomfortable. Companies hoping to sell robots to factory owners will be beaten by the factory owners who build their own. Samsung's advantage isn't the machine itself — it's decades of real production data that no startup can buy. If your business touches industrial operations, now is the time to think about what proprietary data you already hold, before someone turns your operations into their training set.