The Robot Is the Sensor

Two companies, opposite answers to the same question: where does the value live in a physical robot? Tesla is converting its Fremont factory to build a million Optimus bodies a year, betting the whole vertical stack — the body, the autonomy software, the factory floor ^[1]. Piggy Robotics just walked away from humanoid hardware entirely. Its CEO called the body a strategic trap.

The single most valuable strategic decision you can make is deciding what not to build. — Qianyi Deng

The rest of the field scatters in every direction. Anthropic's Model Hardware Standard wants to be the interface between any model and any machine — robotic arms, microscopes, liquid handlers — cutting integration time from weeks to hours ^[2]. Google DeepMind wants to be the brain for all of them.

It's another milestone in our path towards really getting towards what we call like physical AGI, which means we get a robot to do anything that a human can. — Carolina Parada

Sanctuary AI is decoupling its software from its own Phoenix robot and selling the brain to run on whatever platform already exists ^[3]. Nobody's bet matches anyone else's. The digital layer is losing its moat. Token prices hit a record low of 97 cents per million this month ^[4], and analysts now describe frontier models as easy to copy ^[5]. But the price collapse is context, not proof of motive — the capability race is still running alongside it. ByteDance is training a 10-trillion-parameter model ^[6], and Chinese models still trail US rivals by months ^[7]. Nobody is abandoning the brain; they're just no longer sure it's where the money is. Underneath the scatter, the same instinct keeps surfacing at companies that otherwise share nothing. Weave sells a home robot for $8,000 and is explicit about what the deployment is really for.

This is one of the most challenging dexterous manipulation tasks for a humanoid robot. — Figure AI

Samsung built a dedicated data factory at Gumi to analyze real-world industrial data, and plans to deploy robots in its own plants before expanding anywhere else ^[8]. Spirit AI deploys in China to tap local supply chains and data ^[9]. In India, workers wear body cameras to record human motion that becomes training data for robotics ^[10]. Reimagine Robotics names the problem outright.

If they have to call the whiz team robotics guys from wherever they are to fly in and fix it, they just won't be able to do it. It will become an expensive paperweight. — Jonathan Scholz

Its answer is to have factory staff teach robots by physical demonstration, cutting task-teaching from a day to ten minutes ^[11]. Each of these is a company revealing, sometimes accidentally, that the robot is partly a data-gathering instrument. The mechanism runs one direction: deploy robots, collect real-world physical data, train better models, deploy better robots. The moat isn't the body — Piggy is right that bodies can be copied — and it isn't the brain, whose price is collapsing. It's the loop between them, and only companies that physically deploy can close it, because simulation can't substitute for the mess of the real world. The divide is real, and it will sort itself out. But the company that wins may be whichever one realizes first that the robot is not the product, and not the platform. It's the sensor.