XDOF has emerged from stealth with $70 million in funding to build the data and infrastructure layer underneath the next generation of robot foundation models, betting that the hardest part of general-purpose robotics is not the model or the hardware but the data that teaches machines to act in the physical world.
Building the picks-and-shovels for physical AI
Rather than building a robot for a single task, XDOF positions itself as a full-stack infrastructure partner, supplying datasets, robotic systems and software tools to AI labs and commercial developers. The company says it has already been working behind the scenes with leading robotics teams across hardware, operations and policy training. Founded in 2024 by Philipp Wu, Yide Shentu and Nemo Jin, the startup argues that while AI has transformed digital work, most everyday physical tasks remain stubbornly hard to automate, leaving a gap that better data pipelines can close.
ABC-130K: the largest open teleoperation dataset
As its first public release, XDOF open-sourced ABC-130K, which it describes as the largest open-source teleoperation dataset, with more than 130,000 demonstrations spanning 195 bimanual manipulation tasks. The dataset was developed alongside researchers from UC Berkeley, Carnegie Mellon, MIT and Amazon's Frontier AI & Robotics group, and is meant to give labs a head start on training general-purpose manipulation policies.
Heavyweight backers and a crowded race
The $70 million round drew a roster of prominent investors including Thrive Capital, Andreessen Horowitz, Spark Capital, Lux Capital and WndrCo. The raise lands amid a surge of capital into embodied AI infrastructure, echoing recent mega-rounds such as General Intuition's $320 million round for gameplay-derived robotics data and Mind Robotics' $400 million push into physical AI manufacturing. As demand for AI compute and training data climbs, infrastructure plays like Baseten's $1.5 billion inference round show investors are increasingly funding the layers beneath the models themselves.
XDOF's wager is that the next bottleneck in robotics will be the collection and feedback systems that teach robots to engage with the real world, and that whoever owns that layer will sit at the center of the physical-AI economy.
Reporting based on coverage from AI Insider, Pulse 2.0 and TechCrunch.