14 billion for a robot brain: Who is building robotics' operating system

When the iPhone arrived, dozens of companies built mobile phones. In the end, two operating systems made the money. Now two American startups are running exactly this bet for robotics: Skild AI from Pittsburgh and Physical Intelligence from San Francisco don't build their own robots—they build the brain that's supposed to be embedded in as many other machines as possible. Strictly speaking, it's not an operating system like Android, but an AI foundational model that combines vision, language, and movement. But its role in the market would be the same: the layer that no one can get around.

The numbers show how seriously investors take this. Skild AI raised around 1.4 billion dollars in January, led by SoftBank, with Nvidia and Jeff Bezos backing it. Valuation: over 14 billion dollars, three times more than seven months earlier. Physical Intelligence raised 600 million in November, led by Alphabet's growth fund CapitalG, and has been negotiating since spring, according to media reports, over another billion at a valuation exceeding 11 billion dollars. For comparison: Skild's revenue in 2025 was around 30 million dollars. What's being paid for here is a future, not a present.

The two companies come from the same background but are taking different paths. Skild, founded in 2023 by Carnegie Mellon professors Deepak Pathak and Abhinav Gupta, calls its model Skild Brain and markets it with the term omni-bodied: the same brain controls quadrupeds, humanoids, robotic arms, and mobile platforms without needing to be retrained for each form factor. If a motor fails, the model adapts instead of stopping. Skild is pushing aggressively into industry. Partnerships with ABB, Universal Robots, and its sister company MiR are meant to bring the brain into established industrial robots; at Foxconn in Houston, it controls assembly lines where Nvidia's Blackwell servers are made. In April, Skild also acquired the robotics business of Zebra Technologies, the former Fetch Robotics, securing access to warehouses and fleet software.

Physical Intelligence, known in the scene simply as Pi, is the researcher supergroup of the two. Among its founders are Karol Hausman, who led robot manipulation at Google Brain, Berkeley professor Sergey Levine, and Stanford professor Chelsea Finn. Their model Pi Zero initially learned to fold laundry and clear tables; version 0.5 generalized to apartments it had never seen. In November came the biggest step yet: with a training method called Recap, the model continues learning from its own mistakes, like an apprentice who no longer needs every step shown. In the demos, a robot pulls espressos from half past five in the morning until evening, folds fifty unfamiliar clothing items, and assembles praline boxes in a real factory, including labels. On the hardest tasks, the method doubled throughput and halved the error rate, the company says.

The business models also differ. Pi released its first model as open source, playing out the script that Meta pioneered with its Llama language models: first your own tool becomes the standard in labs, then corporate customers license the more powerful versions. Skild keeps its model closed and sells directly to industrial partners. Which path succeeds is open. What's certain is that hardware manufacturers are watching the development closely, because the smartphone story also knows its losers: whoever just built the devices back then handed over the margin to the platform operator.

For Switzerland, this is no distant debate. ABB, headquartered in Zurich, is bringing Skild Brain into its own robots, positioning itself early on one of the two platforms. For operations in Switzerland that today are procuring a robot, the question about software becomes as important as the question about load capacity: a device whose brain learns through updates ages differently than one that only knows its factory programs. Whether in the end Skild, Pi, Google, or Nvidia will occupy the role of Android will be decided in the coming years. That robotics is just experiencing its smartphone moment can hardly be overlooked anymore.