XDOF, just three months out of stealth, is in talks for a Series B at a $1.2B valuation
Original reporting by TechCrunch

XDOF is a startup specializing in the collection of real-world teleoperation data to train general-purpose robots. Less than three months after officially emerging from stealth, the Berkeley-born company is reportedly in late-stage discussions to secure a Series B funding round, valuing it at approximately $1.2 billion. This rapid ascent, led by investor 8VC, follows a $70 million Series A just last June, with significant backing from firms like Thrive Capital and Andreessen Horowitz. The move to raise capital again so quickly was reportedly driven by XDOF’s explosive growth, with annualized revenue nearing $50 million, prompting venture capitalists to initiate the new funding talks.
Addressing the Data Gap
Co-founded by UC Berkeley researchers Philipp Wu and Fred Shentu, XDOF addresses a critical challenge in robotics: the severe lack of large-scale, real-world training data. Unlike large language models, which leveraged the vastness of the internet for training, physical robots lack an equivalent foundational dataset. XDOF aims to fill this void by providing an outsourced data-supply chain for the robotics industry, building sophisticated data pipelines, collection tools, and annotation systems that frontier AI labs and robotics companies struggle to develop independently. Leveraging methods like remote robot teleoperation and human-worn sensors to capture everyday tasks, XDOF is positioning itself as the "Scale AI" for physical robotics, a crucial enabler for the next generation of autonomous machines.
XDOF's rapid ascent to a multi-billion dollar valuation, mere months after its Series A, powerfully underscores the acute and unfulfilled demand for real-world teleoperation data. This swift Series B, driven by its impressive revenue growth, solidifies XDOF's position as a crucial infrastructure provider, akin to a "Scale AI for physical robotics," tackling the fundamental data bottleneck that has long impeded the development of truly general-purpose machines. The investment isn't just a win for XDOF; it's a resounding validation of the strategic imperative to build comprehensive, high-quality datasets for physical AI.
A Foundational Shift
The implications of XDOF's trajectory extend far beyond a single startup's success. This funding round signals a critical maturation in the broader robotics and AI landscape, indicating a shift towards a data-centric paradigm for physical intelligence, mirroring the breakthroughs seen in large language models. The availability of large-scale, diverse real-world data, painstakingly collected and curated by companies like XDOF, is poised to accelerate the development of adaptable robots capable of navigating and performing complex tasks in unpredictable environments. This trend will likely catalyze further investment in data infrastructure, annotation, and teleoperation technologies, fostering an entirely new sub-industry dedicated to feeding the insatiable data appetites of future robots. Ultimately, XDOF's rise foreshadows a future where robust data pipelines enable the widespread deployment of intelligent machines, fundamentally reshaping industries from manufacturing and logistics to healthcare and daily life.
Frequently asked questions
- What is XDOF and what service does it provide to the robotics industry?
- XDOF is a startup that acts as an outsourced data supply chain for the robotics industry. It specializes in collecting real-world teleoperation data to train general-purpose robots. Many frontier AI labs and robotics companies struggle to build the necessary data pipelines and annotation systems themselves, and XDOF provides these critical tools and services, addressing a significant bottleneck in robotics development.
- How does XDOF gather real-world data for training advanced robots?
- XDOF employs a multi-faceted approach to gather high-quality robot training data. It utilizes remote robot teleoperation, where human operators control robotic arms from afar. Additionally, it uses "egocentric operators" who wear sensors to record human actions performing everyday tasks like folding clothes. This combination creates large-scale, real-world datasets essential for teaching robots diverse physical manipulations and behaviors.
- Why is collecting real-world data crucial for developing general-purpose robots?
- Collecting extensive real-world data is critical for developing general-purpose robots because, unlike large language models trained on the vast internet, physical robots lack an equivalent pre-existing dataset. Robots need to learn how to interact with the physical world through vast amounts of observed and teleoperated actions. This data gap is a major impediment, making specialized data collection essential for advancing robotic capabilities and fostering more versatile machines.