Highlights

Less than three months after emerging from stealth, XDOF is reportedly in late-stage discussions for a Series B round valuing the company at about $1.2 billion. Rapid revenue growth — approaching $50 million annualized — has drawn investor interest soon after its $70 million Series A. The startup, founded by UC Berkeley researchers, focuses on collecting and annotating real-world teleoperation data to train general-purpose robots, positioning itself as a data-supply partner for robotics and AI labs.

Sentiment Analysis

• The tone of the piece is largely positive and optimistic, emphasizing strong growth metrics and high investor interest in XDOF. It highlights concrete achievements — a recent Series A, fast revenue acceleration, and partnerships with research labs — suggesting confidence from both customers and investors. Use of a potential $1.2 billion valuation frames the narrative as a breakout success in a niche that addresses a pressing industry need: real-world robotic training data. While the deal terms remain unconfirmed and negotiations could change, the overall sentiment is upbeat and forward-looking, focused on opportunity rather than risk.

Article Text

XDOF, a startup that builds data pipelines and collection systems for robotic training, is reportedly in advanced discussions to raise a Series B at an approximate $1.2 billion valuation, mere months after coming out of stealth. The company, cofounded in 2024 by UC Berkeley researchers Philipp Wu (CEO) and Fred Shentu (CTO), attracted attention earlier this year with a $70 million Series A that included investments from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital. Sources familiar with current negotiations say that XDOF’s swift revenue trajectory — nearing $50 million on an annualized basis — has prompted new investor outreach despite the company not planning to raise so soon after its previous round.

Details about the potential raise remain incomplete. TechCrunch and other outlets reported that the total amount sought and whether the reported valuation incorporates the new capital are unclear. Terms are still being finalized and could change before any agreement is reached. XDOF and the lead investor in the reported talks have not publicly commented on the negotiations.

XDOF’s core proposition is building the tooling, annotation systems, and collection processes that many robotics and frontier AI labs find difficult to develop in-house. By offering an outsourced data-supply chain tailored to physical robotics, the company aims to fill a gap that differs from the data model of large language models. Unlike LLMs, which were able to train on vast text corpora from the internet, robots require high-quality, real-world sensor data — and such datasets are scarce. This scarcity has been a major barrier to advancing general-purpose robotic systems.

The startup’s origins trace back to Wu’s doctoral research on how robots learn from large datasets. Facing a shortage of large-scale data, Wu and Shentu developed GELLO, an economical teleoperation system that enables human operators to control a robot arm remotely to gather training data. Their academic work resulted in a widely cited paper and laid the groundwork for the commercial venture that became XDOF.

Investors compare XDOF to the likes of Scale AI and Mercor in the context of physical robotics: companies that provide the data infrastructure enabling broader AI progress. To assemble what it calls ABC, a large collection of curated robot training data, XDOF is collaborating with UC Berkeley’s AI research lab. The company captures footage and task data by combining remote robot teleoperation with human data collectors who wear sensors to record everyday activities such as folding laundry or flattening cardboard boxes. This hands-on approach aims to create a high-quality dataset that robotics developers currently lack at scale.

Operationally, XDOF plans to recruit and train global teams of data collectors. Roles include teleoperators, who control robots from afar, and egocentric collectors, who use body-worn sensors to capture natural human motions. The startup has said it already serves around 20 customers, including several frontier AI labs working on robotic systems and embodied intelligence.

The move to pursue additional funding reflects both market demand and investor appetite for infrastructure that accelerates robotics development. Other ventures are attempting similar plays, including Mecka AI and established data-labeling platforms expanding into physical-data collection, such as Scale AI and Micro1. As interest in embodied AI grows, companies that can reliably supply volumes of annotated, real-world robotic data are becoming more central to the field’s progress.

For XDOF, the challenge will be scaling data operations while maintaining quality and managing costs. If the reported Series B proceeds at the rumored valuation, it would underscore investor belief in the commercial potential of curated robotic datasets. The outcome could also signal growing momentum for businesses that bridge the gap between lab research and production-ready data infrastructure for robots.

Key Insights Table