XDOF in Series B Talks at $1.2 Billion Valuation for Robot Learning Data

According to reporting by Marina Temkin of TechCrunch AI, XDOF, which collects training data for general-purpose robots, is in talks to raise a Series B at a valuation of around $1.2 billion. The talks are said to be in the final stages, with 8VC expected to lead. This was reported citing multiple people familiar with the matter. Terms have not been finalized and are subject to change. It is unclear what the total raise would be or whether the valuation includes new money. XDOF and 8VC reportedly did not respond to requests for comment. The large-scale talks come less than three months after the company emerged from stealth. Rapid revenue growth is behind the talks.

XDOF’s Rapid Rise Three Months After Emerging

From Stealth

XDOF is a startup founded in 2024. Its co-founders are UC Berkeley researchers Philipp Wu and Fred Shentu. Wu serves as chief executive officer, and Shentu serves as chief technology officer. The company collects real-world teleoperation data and supplies it for training general-purpose robots. It handles the parts that leading AI labs and robotics companies find difficult to build in-house. It provides data pipelines, collection tools, and annotation systems. Its distinctive feature is that it functions as a contracted data supply network. It reportedly raised a $70 million Series A in June. Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital participated. The current talks would mark an additional raise soon after that. The company initially had no plans for an early re-raise, according to reports. Its pace of growth drew investors to approach it.

Behind the Growth Toward $50M in

Annualized Revenue

According to people familiar with the matter, annualized revenue is approaching $50 million. That level was reached in a short time after its public launch. The talks are said to have begun at the initiative of investors. The company said it already has 20 customers. The customers are said to include several leading AI labs. Strong demand has translated directly into revenue. In general-purpose robot development, training data is in short supply. Willingness to pay to resolve that shortage is seen as high. That is because data supply determines development speed. The company’s hiring plans also focus on expanding its collection capacity. It plans to hire and train collection workers around the world. It deploys operators who control robots remotely. It also uses workers who wear body-mounted sensors to record movements. Its supply capacity, combining people and equipment, is the source of its revenue.

Training Data as the Bottleneck for

General-Purpose Robots

Large language models initially used the entire web as a training resource. No equivalent resource exists for physical robots. Real-world motion records are critically scarce. In building general-purpose machines, collection becomes the bottleneck. XDOF aims to resolve this bottleneck. Investors describe the company as the Scale AI or Mercor for physical robots. The label likens it to the annotation giants that supported the rise of language AI. It reflects an assessment as an outsourced supply network for the robotics industry. Robots’ spatial understanding requires continuous measurement and reconstruction. This shares challenges with the continuous reconstruction discussed in Streaming 3D Reconstruction Achieves SOTA with Geometric Context Transformer. It requires not only vision and language, but also records of contact and force. Synchronizing diverse senses is a difficult challenge. Establishing quality-control standards is also essential.

GELLO Low-Cost Teleoperation as the Technical

Foundation

The founding traces back to Wu’s doctoral research. He was studying how robots learn from large volumes of records. A lack of records hampered the research, he said.

“large-scale data to work with”

He explained this in an interview in June. He then worked with Shentu on GELLO. GELLO is a low-cost teleoperation system. It allows a human operator to move an arm-type robot from a distance. It leaves a trace of the operations as training records. The results were published as an influential paper. This research became the foundation for XDOF. Inexpensive operating devices broaden the base for collection. That is because they allow scaling up units without relying on expensive equipment. Training procedures to maintain operation quality are also important. Both tools and personnel are central to the technology. Distributing collection tools and building a verification system are challenges. This operation requires management close to the test-distribution framework discussed in GNOME OS Test Center, Inspired by Apple TestFlight.

The Structure of the Outsourced Data

Supply Network

XDOF is collaborating with UC Berkeley’s AI research lab. It plans to release ABC, a large-scale collection of high-quality robot training records. It aims to make it one of the largest collections to date. The records cover everyday tasks. Examples include folding clothes and flattening boxes. It collects records of machine movements via teleoperation. It also collects motion records obtained from humans wearing sensors. It combines the two to ensure diversity. High-definition displays are essential for visual verification. Consumer display trends overlap in part with the quality standards shown in LG OLED evo Brings Creator Original Picture Mode to Prime Video. The annotation process is also included in the supply network. It consists of three layers: collection tools, pipelines, and annotation. It carves out the layers that leading labs find difficult to build in-house. Outsourcing allows development labs to focus on core development. The stability of the supply network will determine the speed of the entire industry.

Competition Points to an Emerging

Physical-Data Market

XDOF is not the only startup seeking to capture real-world records. Mecka AI operates in the same area. Human-data platforms such as Scale AI and Micro1 are also expanding. Expansion from language-model applications into the physical domain has begun. Moves by major data-annotation companies are seen as signaling the formation of a market. Specifications for general-purpose robot records are not yet established. The very definition of quality is becoming a competitive area. Diversity of operations, diversity of environments, and granularity of annotation are key. The track record of 20 customers serves as a leading indicator. Adoption by leading labs serves as validation of quality. The $1.2 billion valuation reflects expectations for the supply network. The caveat that terms are not finalized remains. Even so, the talks themselves are assessed as showing strong demand.

Editorial Opinion

On short-term impact. We see this negotiation affecting unit prices and delivery times for physical data procurement. Procurement budgets at leading labs will expand, and demand for mass production of collection devices and training of operators will increase. Competitor Mecka AI and existing majors will also be forced to raise their terms. We expect competition for talent to intensify in three to six months. On the long-term view. In one to three years, we assess that standardization of record specifications will become the focus. Public collections like ABC will become benchmarks, and quality metrics will be developed. The establishment of outsourced supply networks will lower barriers to entry for small and mid-sized firms and encourage diversification in hardware development. We assess this will affect the speed of real-world deployment. Questions from the editors. Is verification of bias in records and safety sufficient? It needs to be verified whether everyday tasks such as folding clothes and flattening boxes are representative of general-purpose actions. Is consent management and compensation design for body-worn records appropriate? Will concentration of the supply network create an imbalance in pricing power?

References

• “XDOF, just three months out of stealth, is in talks for a Series B at a $1.2B valuation”, by Marina Temkin — TechCrunch AI, 2026-09-04T23:36:14.000Z (ARR)
• Source URL: https://techcrunch.com/2026/09/04/xdof-just-three-months-out-of-stealth-is-in-talks-for-a-series-b-at-a-1-2b-valuation/