XDOF Robotics in talks for $1.2B Series B funding round
The startup is building data pipelines for training general-purpose robots.
1Rapid funding progression
Just three months after emerging from stealth and only months after a $70M Series A, XDOF Robotics is already in late-stage talks for a Series B at a $1.2B valuation. The round is being led by 8VC, with the startup's annualized revenue approaching $50 million driving investor interest. This rapid funding progression highlights the intense VC appetite for AI infrastructure plays in robotics. The company wasn't initially planning to raise again so soon after its June funding round.
2Founders' academic roots
XDOF was co-founded in 2024 by UC Berkeley researchers Philipp Wu (CEO) and Fred Shentu (CTO). Wu's PhD research focused on how robots learn from large datasets, where he encountered the critical problem of insufficient training data. Shentu collaborated with Wu on GELLO, a low-cost teleoperation system that became the foundation for XDOF. Their academic work resulted in an influential robotics paper that demonstrated the potential of human-generated training data for machine learning in physical robots.
3Data collection methodology
XDOF combines two innovative approaches to collect robot training data: remote teleoperation where humans control robotic arms, and egocentric data collection where operators wear body sensors. This dual methodology captures diverse real-world tasks like folding clothes and flattening boxes. The company is partnering with UC Berkeley's AI Research lab to release ABC, what it claims will be the largest collection of high-quality robot training data ever assembled.
4Market positioning
Investors describe XDOF as the Scale AI or Mercor for physical robotics - a reference to data-labeling giants that fueled the AI boom. Unlike LLMs that trained on internet data, physical robots lack equivalent real-world datasets. XDOF aims to solve this by building the data pipelines, collection tools, and annotation systems that AI labs and robotics companies can't easily develop themselves, positioning as an outsourced data-supply chain for the industry.
5Current traction
The startup already counts 20 customers, including several frontier AI labs, demonstrating early market validation. XDOF plans to scale its operations by hiring and training global teams of data collectors. This includes both teleoperators who remotely control robots and egocentric operators who capture movement data through body sensors. The company's rapid revenue growth to nearly $50 million annualized shows strong demand for its services.
6Competitive landscape
XDOF operates in a emerging but competitive space. Other players include Mecka AI in robot-specific data collection, plus human-data platforms like Scale AI and Micro1 that are expanding beyond LLMs into physical robotics. The $1.2B valuation talks suggest investors see XDOF's specialized approach - focusing exclusively on robotics data - as particularly valuable compared to more generalized data labeling platforms.
7Technical differentiation
XDOF's technology builds on Wu and Shentu's academic work with GELLO, which demonstrated how low-cost teleoperation could generate quality training data. Their system addresses the fundamental challenge that while internet-scale data exists for digital AI, physical robots require different kinds of real-world interaction data. XDOF's annotation systems and data pipelines are specifically optimized for robotics applications rather than being adapted from other AI domains.
8Industry implications
XDOF's rapid rise reflects the critical importance of training data in advancing robotics AI. As the field moves toward general-purpose robots, high-quality, diverse datasets become the key bottleneck. The startup's $1.2B potential valuation after just three months suggests investors believe solving the data problem could unlock the next phase of robotics innovation, similar to how data labeling enabled the current AI boom.
Original article: TechCrunch — XDOF Robotics in talks for $1.2B Series B funding round