XDOF is in Series B talks at a .2B valuation just months after exiting stealth, positioning itself as the essential data supply chain for general-purpose robots.

The trajectory of generative AI was paved by the availability of massive, unstructured datasets from the open web. However, the quest for general-purpose robotics has hit a physical wall: there is no internet equivalent for how a robotic arm folds a shirt or flattens a cardboard box. Enter XDOF, a San Mateo-based startup that is rapidly becoming the infrastructure layer for embodied AI.

Less than three months after emerging from stealth mode on June 17, 2026, XDOF is reportedly in advanced discussions to secure Series B funding. Sources indicate the round is being led by 8VC and could propel the company to a valuation of approximately .2 billion. This meteoric rise follows a million Series A closed in June, which saw backing from heavyweights including Thrive Capital, Andreessen Horowitz, Lux Capital, and Spark Capital.

Solving the robotics data bottleneck

The fundamental challenge in robotics is the scarcity of high-quality, real-world training data. While Large Language Models (LLMs) could ingest trillions of tokens from websites, robots require precise, multimodal demonstrations of physical tasks. XDOF addresses this by acting as an outsourced data-supply chain, building the pipelines, annotation systems, and collection tools that most AI labs lack the resources to develop internally.

The company's technical foundation stems from the academic work of co-founders Philipp Wu (CEO) and Fred Shentu (CTO), both researchers from UC Berkeley. During his PhD, Wu identified the lack of large-scale datasets as the primary impediment to robotic learning. This led to the creation of GELLO, a low-cost teleoperation system that allows human operators to control robotic arms remotely. This system enables the generation of the precise training data necessary for robots to mimic human dexterity and decision-making in physical spaces.

A business model built for scale

XDOF is not merely a research project; it has scaled into a commercial powerhouse with surprising speed. The company currently employs roughly 60 people and serves approximately 20 customers, primarily consisting of frontier AI labs and specialized robotics firms. This focused client base has driven annualized revenue toward the million mark, a figure that prompted venture capitalists to approach the company for a new round of funding far sooner than the founders had originally planned.

To maintain this growth, XDOF utilizes a structured approach it calls the data pyramid. This framework allows the company to scale the volume and quality of data it provides to clients, ensuring that the transition from a laboratory prototype to a general-purpose machine is supported by a robust evidentiary base of physical interactions.

The ABC-130K dataset and open-source strategy

Beyond its proprietary services, XDOF is positioning itself as a pillar of the broader robotics community. In partnership with MIT and UC Berkeley's AI Research lab, the startup has created and released ABC-130K. This is described as the largest open-source dataset for bimanual robot manipulations, containing 130,000 trajectories.

By releasing such a massive dataset, XDOF achieves two goals: it establishes its technical authority in the field and creates a baseline for the industry that encourages more companies to adopt the data-centric approach to robotics. This strategy mirrors the early days of the AI boom, where open-source contributions often drove the adoption of the underlying tools provided by the contributors.

How XDOF captures physical intelligence

The process of gathering this data is a blend of high-tech remote operation and human labor. XDOF combines remote robot teleoperation with a global workforce of human collectors. These collectors wear specialized sensors to record the nuances of everyday tasks, which are then translated into training data for AI models.

The startup plans to hire and train teams of data collectors worldwide, including teleoperators who can guide robots through complex environments from a distance.

This approach allows XDOF to capture the edge cases of physical movement—the slight adjustments a human makes when a grip slips or a surface is uneven—which are nearly impossible to simulate in a purely synthetic environment. This real-world grounding is what makes their data highly coveted by firms racing to build humanoid robots.

The Scale AI of embodied AI

Investors are increasingly viewing XDOF as the robotics equivalent of Scale AI or Mercor. Just as those companies provided the labeling and data curation that made LLMs viable, XDOF is providing the physical-world labels for the next generation of machines. The total funding to date, including an early seed round and the million Series A, stands at approximately million, but the pending Series B suggests a massive shift in how the market values data infrastructure for the physical world.

As the industry moves toward general-purpose robots capable of working in warehouses, hospitals, and homes, the demand for high-fidelity teleoperation data will only increase. XDOF's ability to scale its workforce of global data collectors will likely be the deciding factor in whether it can maintain its lead over internal data teams at larger AI labs.

Strategic implications for US and UK enterprises

For business leaders in the USA and UK, the rise of XDOF signals a shift in the robotics procurement landscape. Companies no longer need to be robotics experts to integrate advanced automation; they can now leverage the outputs of a specialized data supply chain. In the US, where the race for humanoid robotics is intensifying among tech giants, XDOF provides a critical shortcut for mid-sized firms to compete without building their own massive data collection infrastructure.

In the UK, where the robotics sector is heavily integrated into advanced manufacturing and healthcare, the availability of datasets like ABC-130K could lower the barrier to entry for SMEs developing niche robotic applications. From a regulatory perspective, the move toward outsourced data collection brings questions of data provenance and labor standards for the global workforce of teleoperators. While the EU's AI Act focuses heavily on the risk profiles of AI systems, US and UK firms will likely focus on the intellectual property rights of the training data and the efficiency of the data pipelines provided by vendors like XDOF. The ability to outsource the most tedious part of AI development—data collection—allows enterprises to focus on the application layer, accelerating the deployment of embodied AI in commercial settings.

FAQ

---

Sources: TechCrunch, Cryptobriefing, Robottoday ·