In 2024, a group of UC Berkeley researchers launched XDOF, a robotics data company now nearing the close of a Series B fundraising round that could see its valuation rise to about $1.2 billion, according to individuals familiar with ongoing discussions.
Specializing in large-scale teleoperation data to help accelerate the evolution of general-purpose robots, XDOF is drawing new investment interest led by 8VC. Sources close to the negotiations caution that deal specifics remain unsettled and the final structure of the round could shift. There is still no public detail on the round’s exact size or whether the newly reported valuation includes incoming capital.
XDOF Builds Momentum Following Series A
Emerging from stealth less than three months ago, XDOF has made waves thanks to swift commercial progress, now reporting annualized revenues near $50 million, which is exceptional for a company in such early scaling stages. This impressive momentum followed its June $70 million Series A, which attracted investments from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. Although XDOF had initially not planned to raise again so soon, escalating proposals from venture capital firms have prompted these new negotiations.
Neither XDOF nor 8VC have issued statements on the developments, but those familiar with their plans suggest a strong competitive position within the robotics data sector going forward.
Tackling the Robotics Data Challenge
Philipp Wu (CEO) and Fred Shentu (CTO) established XDOF, driven by a mission to address the scarcity of robust data that AI robotics systems need. Wu, inspired during his PhD by the absence of “large-scale data” for effective machine learning in robotics, teamed up with Shentu to create GELLO, a low-cost teleoperation system enabling remote control of robotic arms. Their efforts resulted in a respected robotics paper, laying the groundwork for XDOF’s inception.
At the core of XDOF’s business, the company works as an external data infrastructure provider for roboticists and AI labs, offering data collection pipelines, annotation frameworks, and custom datasets. This function has become crucial since large language models leverage vast internet data, yet real-world data for physical robots remains elusive. XDOF hopes to bridge this gap for the global robotics sector. Some investors see their approach as analogous to leading AI data-labeling companies like Scale AI and Mercor.
A highlight of their partnerships is XDOF’s work with UC Berkeley’s AI Research lab on the ABC dataset project, purportedly the largest and best-quality robot training set to date. Their methodology combines remote robot teleoperation with egocentric human data: operators wearing body-mounted sensors record everyday activities such as folding laundry and disassembling boxes.
Scaling Data Gathering Operations Globally
To keep pace with the rising demand for robust training data, XDOF intends to assemble an international cohort of data contributors, including remote teleoperators and individuals leveraging wearable sensors to capture genuine movement patterns. Previously, the company announced it has brought on 20 customers, several of which are leading names in the AI research community.
Competition for real-world robotics data is heating up, with entrants like Mecka AI developing similar data platforms and established players such as Scale AI and Micro1 diversifying beyond their traditional focus on large language model data.
XDOF’s quick ascent and active fundraising highlight the crucial role of expansive, high-quality real-world training datasets as the industry chases the goal of versatile, general-purpose robotics—potentially placing XDOF at the center of infrastructure for the next era in AI and robotics.