Three Months Out of Stealth and Already Worth 1.2 Billion Dollars: The Company Does Not Build Robots, It Records What Robots Do Not Know

Three months. That is how long it has been since the startup XDOF came out of stealth, and that is how long it took to reach negotiations for a new round at an estimated valuation of around 1.2 billion dollars (about 1.1 billion euros). The round is being led by the fund 8VC.

The company was founded by two researchers from the University of California, Berkeley in 2024 - Philipp Wu as chief executive and Fred Shentu as chief technology officer. In June they raised 70 million dollars in a Series A, with Thrive Capital, Andreessen Horowitz, Lux and Spark Capital. They were not planning to raise again this quickly. The investors came to them, after annual revenue approached 50 million dollars.

What does XDOF actually sell? Not a robot. Data for robots.

The large language models got lucky - the entire internet lay there as learning material. Physical robots have no such pile. Nobody recorded billions of hours of a human hand folding a shirt or flattening a cardboard box. That is the bottleneck of the whole industry, and XDOF is building exactly that chain: collection tools, labelling systems, people wearing sensors and recording ordinary household tasks, and operators driving robotic arms remotely.

The foundation is a project called GELLO - a cheap system that lets a person control a robotic arm at a distance while generating learning material in the process. The company grew out of the paper that came from it. Investors today describe it as Scale AI, but for physical robots. It is already working with around twenty clients, among them several leading artificial intelligence labs.

Here is the part worth noticing. The plan is to hire and train teams of data collectors all over the world - people who will spend hours driving robots remotely or wearing sensors on their bodies while doing everyday work. That is a new kind of job that did not exist before, and that job will not be paid for in San Francisco. The data-labelling chains for language models have already passed through Kenya, India and the Philippines. There is no reason for this one to take a different route.

The terms of the deal are not final and may change. XDOF and 8VC did not respond to journalists' questions. But the figure has already been sent out into the world, and it is doing its work - a two-year-old company, three months out of stealth, worth as much as a mid-sized European firm with factories and employees. The question no investor asks out loud is what happens to that valuation if a competitor finds a cheaper way to collect the same data.