Just three months after coming out of stealth mode, XDOF is in advanced discussions to secure Series B funding at roughly $1.2 billion, according to a TechCrunch report published September 4, 2026. The startup, which gathers real-world teleoperation information to train general-purpose robots, has drawn interest from venture capitalists led by 8VC. The company's meteoric rise reflects investor appetite for infrastructure powering the next wave of AI-driven robotics.
XDOF raised $70 million in Series A funding just three months earlier in June, with backing from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. The company didn't originally plan to pursue another round so quickly, but its accelerating expansion prompted venture firms to initiate discussions, according to multiple sources familiar with the negotiations. XDOF's annualized revenue is nearing $50 million, the report states. The startup already counts 20 customers among its client base, including several frontier AI labs. The deal's terms haven't been finalized and remain subject to change, and it's unclear how much total capital the round will bring in or whether the $1.2 billion figure accounts for the new investment.
UC Berkeley researchers Philipp Wu and Fred Shentu launched XDOF in 2024 after their academic work exposed a critical gap in robotics development. Wu told TechCrunch in June that during his PhD studies on how robots learn from massive datasets, the absence of "large-scale data to work with" became a major obstacle. The pair created GELLO, an affordable teleoperation system enabling human operators to remotely guide robotic arms and produce training information, which became the subject of an influential robotics paper. Investors now characterize the company as the Scale AI or Mercor equivalent for physical robotics, referencing the data-labeling firms that powered the recent AI surge.
The report explains that unlike large language models, which drew on the entire internet for initial training, physical robots lack a comparable real-world dataset, making information gathering a crucial constraint for building multipurpose machines. XDOF constructs the data infrastructure, collection mechanisms, and annotation frameworks that frontier AI labs and robotics firms struggle to develop internally, functioning essentially as an outsourced data supply chain for the robotics sector. The startup blends remote robot teleoperation with human collectors wearing sensors to document routine activities such as folding laundry and flattening cardboard. XDOF is collaborating with UC Berkeley's AI Research lab to publish what it considers the most extensive archive of premium robot training information ever compiled, called ABC.
XDOF intends to recruit and prepare teams of data collectors around the globe, including remote teleoperators who pilot robots from a distance and egocentric operators who wear body-mounted sensors to capture motion information. Other startups pursuing similar real-world data for robot training include Mecka AI, while human-data platforms like Scale AI and Micro1 are expanding beyond language models into this space. For companies building the next generation of autonomous machines, access to diverse, high-quality movement data may determine who leads and who follows. The robotics industry's infrastructure layer is taking shape faster than many anticipated, and the firms supplying the raw material for robot intelligence are attracting capital at a pace that mirrors the early LLM era.