Robotics data company XDOF is discussing a Series B financing at a valuation of roughly $1.2 billion, less than three months after it emerged from stealth. The talks are being led by 8VC, according to sources cited by TechCrunch, and follow a $70 million Series A announced in June with participation from Thrive Capital, Andreessen Horowitz, Lux Capital and Spark Capital.
The speed of the potential new round was not part of XDOF’s original plan. Rapid commercial growth changed the conversation: its annualized revenue is approaching $50 million, which reportedly drew venture investors back to the company. The total amount of capital being discussed is not known, the valuation may be pre- or post-money, and the terms are not final. XDOF and 8VC did not comment.
A young company targets robotics’ data shortage
XDOF was founded in 2024 by former University of California, Berkeley researchers Philipp Wu, now chief executive, and Fred Shentu, now chief technology officer. Its business is built around a bottleneck Wu encountered in research: physical robots do not have an internet-scale store of real-world examples comparable to the text and images available for training language models.
The company builds data pipelines, collection tools and annotation services for frontier laboratories and robotics developers. In effect, it is trying to create an outsourced supply chain for physical AI data. XDOF says it already serves 20 customers, including several frontier labs, although TechCrunch’s report does not identify those customers or independently quantify their spending.
That market position helps explain why investors might consider another round so quickly. Revenue momentum is a meaningful signal, but an annualized figure is not the same as audited full-year revenue, and a proposed valuation is not a completed financing. The discussions could still change.
Teleoperation turns human actions into robot examples
XDOF grew from GELLO, a low-cost teleoperation system developed by its founders. GELLO lets a person remotely control a robot arm, producing demonstrations that can be used as training data. The founders’ paper became influential in robotics research and supplied the technical foundation for the company.
The collection process extends beyond operators moving robotic arms. XDOF also employs people wearing body sensors while they perform ordinary physical tasks, including folding clothes and flattening cardboard boxes. Those recordings capture first-person movement data that models can use to learn how actions unfold in the real world.
The company plans to assemble global teams of both remote teleoperators and sensor-equipped workers. It has also partnered with the Berkeley Artificial Intelligence Research lab to release ABC, which it describes as the largest high-quality robot training dataset. That superlative is XDOF’s claim; the source does not provide an independent comparison with other datasets.
Together, these methods address a practical constraint. A robot needs examples grounded in contact, motion and changing physical environments, but collecting those examples is slower and more labor-intensive than gathering digital text. XDOF is betting that standardized tools and distributed human collection can make the process large enough for the next generation of robotics models.
Competition is forming around the physical data layer
XDOF is not alone in pursuing this opportunity. TechCrunch identifies Mecka AI as a competitor, while data-labeling companies including Scale AI and Micro1 are expanding into related work. The contest is therefore not only about building robots; it is also about owning the systems that produce, organize and annotate the demonstrations those robots need.
For XDOF, the reported financing talks would provide a large valuation marker at an unusually early point in the company’s public life. They would also show investor interest shifting toward the less visible infrastructure behind physical AI: human operators, sensors, annotation and repeatable collection pipelines.
The central wager is that robot data can become a scalable business before any single robotics application dominates. XDOF’s reported revenue pace and customer count suggest demand, but the proposed Series B remains under discussion and its financial terms remain uncertain. What is already clear from the company’s model is where it sees scarcity: not simply in better robot hardware, but in the volume and quality of real-world experience available to train it.
