Why Is the XDOF Series B Deal Getting So Much Attention?

Question → What is happening with XDOF?

XDOF is reportedly in late-stage discussions to raise a Series B at a valuation of about $1.2 billion, with 8VC expected to lead the round, according to people familiar with the deal cited by TechCrunch.

What makes the situation unusual is the timing. XDOF had only recently emerged from stealth, and TechCrunch reported in June that the startup had raised a $70 million Series A from investors including Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital.

The company was not originally planning another financing round so quickly. According to the report, its rapid growth,including annualized revenue approaching $50 million,prompted venture capital firms to approach the startup about another round.

That puts XDOF in an unusual position. It is still a young company, but investors appear to see an opportunity to build infrastructure around one of the biggest bottlenecks in robotics: obtaining enough high-quality data to teach machines how to operate in the physical world.

What does the reported $1.2 billion valuation actually mean?

It means investors are reportedly discussing valuing XDOF at around $1.2 billion, but that figure should not be treated as a finalized valuation.

TechCrunch reported that the deal’s terms remain subject to change and that it could not confirm whether the $1.2 billion figure includes the new funding.

That distinction matters. Startup valuations reported during fundraising discussions are not necessarily the same as final transaction terms.

Still, even as a potential valuation, the figure shows how aggressively investors are pursuing infrastructure for physical AI,AI systems that interact with the real world through robots and other machines.

What Does XDOF Actually Do?

At first glance, XDOF might sound like another AI startup. It is not primarily building a chatbot, foundation model, or consumer robotics product.

Instead, XDOF is trying to solve a less glamorous but extremely important problem: collecting the data robots need for training.

Definition + Expansion , Robot training data: Robot training data is information collected from physical-world interactions that can help AI systems learn how to perceive environments, understand tasks, and control robotic systems.

Large language models can learn from enormous quantities of existing digital information, much of which is available online. Robots face a different challenge.

A robot cannot simply download the internet and learn how to reliably fold clothes, manipulate objects, navigate a kitchen, or operate tools. It needs information about physical movement, objects, environments, forces, actions, and consequences.

That information has to come from the physical world.

Question → Why is robot training data so difficult to collect?

Because physical data generally requires physical interaction.

Someone,or something,must perform the task, record the movements and environment, and then organize that information so researchers can use it to train AI systems.

That process can involve robots, sensors, human operators, annotation systems, specialized hardware, and large-scale data collection teams.

XDOF wants to provide much of that infrastructure as an outsourced service.

Why Physical AI Has a Data Problem

The AI industry has benefited enormously from the availability of digital data.

Large language models, for example, can be trained using huge collections of text and other digital material. The internet provides an enormous pool of information that can be processed computationally.

Robotics does not have an equivalent universal dataset.

A robot learning to manipulate a box needs information about how a human or machine approaches the box, grasps it, moves it, responds to resistance, and completes the task.

A robot learning to fold clothes needs entirely different physical interactions.

This means robotics companies face a fundamental challenge: how do you generate enough diverse, high-quality physical-world examples to train general-purpose robots?

Question → Is robot data more important than robot hardware?

Hardware remains essential, but high-quality data is increasingly viewed as a critical component of robotics development.

A sophisticated robot still needs an AI system capable of interpreting its environment and deciding what to do. Without sufficient training examples, researchers can struggle to develop models that generalize across different objects, environments, and tasks.

This is why companies like XDOF are attracting attention. They are targeting the infrastructure behind the robots rather than only the robots themselves.

How Teleoperation Helps Train Robots

One of XDOF’s core approaches involves teleoperation.

Teleoperation means that a human operator remotely controls a robot or robotic system. The operator’s actions can then be recorded as training data.

Imagine a robotic arm sitting in a laboratory. Instead of asking an AI model to figure out how to manipulate an unfamiliar object from scratch, a human can remotely control the arm while sensors capture the interaction.

That creates an example of how the task can be performed.

XDOF is attempting to turn this concept into a scalable data-collection pipeline.

Question → What is robotic teleoperation?

Robotic teleoperation is the remote control of a robot by a human operator. The operator’s movements can be captured and transformed into data that helps AI systems learn physical tasks.

Teleoperation is particularly useful because it allows humans to demonstrate behaviors that robots can later learn to reproduce.

The approach also creates a bridge between human expertise and machine learning. Rather than requiring engineers to manually program every possible movement, researchers can collect examples of people performing tasks.

How XDOF’s GELLO Research Led to the Startup

XDOF’s origins go back to research conducted by its co-founders at UC Berkeley.

The company was founded in 2024 by researchers Philipp Wu, who serves as CEO, and Fred Shentu, who serves as CTO.

Wu was studying how robots learn from large datasets during his PhD research. According to his comments to TechCrunch, one major obstacle was a lack of large-scale data to work with.

That problem led Wu and Shentu to work on GELLO, a low-cost teleoperation system designed to let a human remotely control a robotic arm to generate training data.

Their research resulted in an influential robotics paper and eventually provided the foundation for XDOF.

Question → Why is GELLO important to XDOF?

GELLO demonstrated a practical way to collect robot-learning data through human-controlled robotic systems.

The concept helped establish the technical foundation for XDOF’s broader business: creating systems that can collect large quantities of physical-world training data for companies developing robots and AI models.

This is an important example of academic research turning into startup infrastructure.

Why Investors Compare XDOF With Scale AI and Mercor

One of the most interesting descriptions of XDOF is that it could become the Scale AI or Mercor for physical robotics.

The comparison is not saying XDOF is identical to either company. Instead, it describes the role XDOF hopes to play in the robotics ecosystem.

Definition + Expansion , AI data infrastructure: AI data infrastructure consists of the systems, workers, tools, and processes used to collect, label, organize, and prepare data for machine-learning models.

Scale AI became an important part of the AI ecosystem by helping companies obtain and prepare data for machine-learning systems.

Mercor has also built a business around human expertise and data-related work for AI development.

XDOF is pursuing a similar infrastructure position, but with an important difference: its data is generated from physical-world activities.

That makes the problem considerably different from labeling text, images, or other existing digital information.

Question → Why does robotics need its own version of Scale AI?

Because physical robots cannot rely entirely on the same datasets used to train language or vision models.

Robotics requires information about actions and interactions in the real world. XDOF is trying to create a specialized supply chain for collecting that information at scale.

If general-purpose robots become widespread, the companies supplying their training data could become strategically important.

XDOF’s $70M Series A Came Before the Reported Series B

The speed of XDOF’s fundraising is one of the strongest signals in the story.

TechCrunch reported in June that XDOF raised a $70 million Series A.

The round included participation from:

Thrive Capital
Andreessen Horowitz
Lux
Spark Capital

At that point, XDOF was building its business around collecting data for robotics companies and frontier AI labs.

The startup reportedly told TechCrunch that it was already working with 20 customers, including several frontier AI labs.

That customer traction helps explain why investors could become interested in another financing round so soon after the Series A.

Question → Why would a startup raise another round so quickly?

Normally, a company might use a recent funding round to finance growth for a longer period before returning to investors.

In XDOF’s case, TechCrunch reported that rapid business growth and annualized revenue approaching $50 million attracted unsolicited investor interest.

The company therefore appears to be responding to market demand rather than following its original fundraising timeline.

What Is XDOF’s ABC Robot Training Data Project?

XDOF is also working with UC Berkeley’s AI Research lab on a project called ABC.

The project is intended to release what XDOF believes will be the largest collection of high-quality robot training data assembled to date.

The significance of ABC goes beyond the size of a dataset.

A useful robotics dataset needs to contain diverse examples that help AI systems understand how humans interact with the physical world. The broader the collection of tasks and environments, the more opportunities researchers may have to train systems capable of generalizing.

Question → What is the ABC project designed to accomplish?

ABC is designed to create a large collection of high-quality robot training data that researchers and robotics developers can use to advance physical AI.

XDOF’s broader strategy is to make data collection more systematic and scalable rather than leaving every robotics company to build its own collection infrastructure.

How XDOF Collects Real-World Robot Data

XDOF’s approach combines several forms of human-generated data collection.

One method involves remote teleoperation, where human operators control robots from a distance.

Another involves egocentric operators who wear sensors while performing everyday activities. Egocentric data refers to information captured from the perspective of the person performing the action.

The company has discussed collecting data from ordinary tasks such as:

Folding clothes
Flattening boxes
Manipulating everyday objects
Controlling robotic systems remotely
Recording human movement using body sensors

The objective is to capture information about how physical tasks are performed.

Question → Why use humans to generate robot data?

Humans already possess broad physical intelligence. We can manipulate unfamiliar objects, adjust our movements when something goes wrong, and complete everyday tasks without explicitly programming each movement.

Recording these behaviors gives AI researchers examples that can potentially be used to train robots.

The challenge is scaling that process to thousands or millions of examples.

XDOF Wants to Build a Global Data-Collection Workforce

Collecting a handful of demonstrations is not enough to create a massive robotics dataset.

XDOF therefore plans to hire and train teams of data collectors around the world.

These workers could include different categories of operators, including:

Teleoperators: People who remotely control robots to generate demonstrations of physical tasks.

Egocentric operators: People who wear body sensors while performing tasks so that their movements and interactions can be recorded.

This workforce would effectively become part of the physical-AI data supply chain.

Question → Why does XDOF need people around the world?

Robots need to operate in diverse environments and encounter different objects, tasks, and human behaviors.

A dataset collected in one laboratory or one type of environment may not represent the complexity of the real world.

A distributed data-collection workforce could help produce more diverse training examples.

XDOF vs Traditional AI Data Companies

XDOF’s business becomes easier to understand when compared with traditional data companies.

Factor	Traditional AI Data Companies	XDOF
Main focus	Digital AI data	Physical robotics data
Typical data	Text, images, human expertise	Robot interactions and human movement
Collection environment	Primarily digital or controlled workflows	Physical-world environments and robots
Human involvement	Data labeling or expert contribution	Teleoperation and sensor-based collection
Main customers	AI and software companies	Robotics companies and frontier AI labs
Key challenge	Scaling high-quality digital data	Scaling high-quality physical-world data
Strategic opportunity	AI model development	General-purpose robotics and physical AI

The comparison reveals why investors may see XDOF as part of a new category.

The company is not simply selling robots. It is building the data infrastructure underneath robotic intelligence.

Why XDOF’s Business Could Become Important for General-Purpose Robots

General-purpose robots are designed to perform a broad range of tasks rather than one narrowly defined function.

A warehouse robot might be optimized for moving packages. A general-purpose machine could eventually be expected to manipulate different objects, perform household tasks, navigate unfamiliar environments, and adapt to new situations.

That ambition creates an enormous data requirement.

Question → Why do general-purpose robots need so much training data?

They need to encounter many different situations if they are expected to operate reliably outside controlled environments.

A robot trained only on one table, one type of box, and one movement pattern may struggle when the environment changes.

Diverse data can expose AI systems to more variations and potentially improve their ability to generalize.

This is why XDOF’s business thesis is larger than simply collecting demonstrations for individual robots. It is attempting to create infrastructure for the broader development of adaptable physical AI.

Who Else Is Collecting Real-World Data for Robots?

XDOF is entering an emerging market rather than operating without competition.

The supplied TechCrunch report identifies Mecka AI as another startup working on real-world data collection for robot training.

Meanwhile, established human-data companies are also moving beyond traditional AI workloads.

Scale AI and Micro1, for example, have been expanding their activities around human-generated data and AI development.

The competitive environment matters because data collection could become a major layer of the robotics economy.

Question → Does XDOF have the market to itself?

No. Other startups and established AI-data companies are also exploring real-world data for robotics.

XDOF’s potential advantage lies in its focus on physical robotics, its research origins, its customer base, and its effort to build specialized infrastructure for collecting robot demonstrations.

Whether those advantages translate into a durable market position remains to be seen.

What Could Go Wrong for XDOF?

A billion-dollar valuation is not a guarantee of success.

The robotics-data market still has major uncertainties. The cost of collecting physical data can be significant, and it may be difficult to determine which datasets will ultimately provide the most value to robotics companies.

There is also a technological question.

If robot-learning models become much more efficient, they might require less human-generated data than currently expected. Alternatively, robotics companies could decide to build proprietary data-collection systems instead of outsourcing the work.

Question → What is XDOF’s biggest business risk?

One major risk is whether XDOF can turn today’s demand for robot training data into a durable, scalable infrastructure business.

The company needs customers to continue paying for its data and collection services while robotics models and training techniques evolve.

Competition is another challenge. If robotics becomes a major AI market, other data companies are likely to pursue the same opportunity.

Why the XDOF Series B Could Signal a Shift in AI Investing

For years, much of the AI investment boom centered on language models, GPUs, cloud infrastructure, and software.

Robotics introduces another layer: physical-world infrastructure.

Investors increasingly have to think about how AI systems will learn to interact with the real world.

That includes robots, sensors, teleoperation platforms, simulation environments, data pipelines, annotation systems, and human workers who generate physical demonstrations.

Question → What does the XDOF Series B say about investor priorities?

If the reported deal closes near the proposed valuation, it would demonstrate strong investor confidence in the idea that robot training data can become a major infrastructure category.

The reported valuation also suggests that investors may see data collection as more than a supporting service. They may view it as a strategic layer of the emerging physical-AI economy.

Why This Matters to Students and Young AI Professionals

The XDOF story is useful for anyone learning about AI because it demonstrates that AI careers extend far beyond building language models.

Modern AI systems depend on multiple layers of work.

Those layers can include:

Data collection
Data engineering
Data annotation
Machine learning
Robotics
Computer vision
Sensor technology
Human-computer interaction
AI infrastructure
Model evaluation
Hardware-software integration

For students, this is an important lesson.

You do not necessarily need to build the next foundation model to participate in the AI economy. Companies such as XDOF are creating businesses around the infrastructure required to make advanced AI systems work in the physical world.

Question → What skills could become valuable as physical AI grows?

Knowledge of robotics, computer vision, machine learning, data engineering, sensor systems, human-computer interaction, and AI evaluation could all become increasingly relevant.

Understanding how data moves from the physical world into an AI training pipeline is particularly useful for anyone interested in robotics.

What Happens Next for XDOF?

The immediate question is whether the reported Series B actually closes.

TechCrunch reported that XDOF and 8VC did not respond to its request for comment, and that the terms were not final.

If completed, the round would give XDOF additional capital to expand its data-collection infrastructure, hire and train more operators, develop its technology, and work with additional robotics customers.

The larger question is whether XDOF can establish itself as a foundational supplier for physical AI.

Question → Could XDOF become a major robotics infrastructure company?

It could, if the market for general-purpose robots grows rapidly and robotics companies continue to need outsourced, high-quality physical-world data.

Its potential success depends on several factors: customer demand, dataset quality, collection costs, model-training requirements, competition, and the speed at which general-purpose robotics develops.

The Bigger Picture: Robots Need More Than Better Hardware

It is tempting to think of robotics progress primarily as a hardware problem.

Build a better actuator. Add better sensors. Improve the battery. Make the robot stronger or more precise.

But intelligence is another challenge.

A robot needs to understand what it sees and determine what action to take. That requires models trained on relevant examples of physical interaction.

This is where XDOF’s business becomes interesting.

The company’s core thesis is simple: if general-purpose robots need huge amounts of real-world training data, someone has to build the infrastructure for collecting that data.

XDOF wants to be that infrastructure provider.

Its potential $1.2 billion valuation therefore reflects more than excitement around one startup. It reflects investor interest in the possibility that physical AI will require an entirely new data economy.

XDOF Series B: Key Takeaways

The most important facts from the reported deal are:

XDOF is reportedly in late-stage talks for a Series B at a valuation of about $1.2 billion.
8VC is reportedly leading the potential round.
The company emerged from stealth less than three months before the report.
XDOF previously raised a $70 million Series A.
Its annualized revenue was reportedly approaching $50 million.
XDOF was already working with 20 customers, including several frontier AI labs.
Its technology is focused on collecting real-world data for robot training.
Its founders, Philipp Wu and Fred Shentu, are researchers from UC Berkeley.
Its GELLO project provided a foundation for the company’s teleoperation approach.
XDOF is partnering with UC Berkeley’s AI Research lab on the ABC robot-data project.
The company plans to build global teams of teleoperators and egocentric data collectors.
Competition includes robotics-data startups and established AI-data companies.
The reported financing terms are not final and could change.

The bigger lesson is straightforward: the AI data problem is expanding from the digital world into the physical world.

As robots become more capable, the companies that can efficiently collect and organize real-world demonstrations could become just as important as the companies building the models themselves.

Frequently Asked Questions About XDOF Series B

What is XDOF?

XDOF is a robotics-data startup founded in 2024 by UC Berkeley researchers Philipp Wu and Fred Shentu. The company collects real-world teleoperation and human movement data intended for training general-purpose robots.

What is the reported XDOF Series B valuation?

XDOF is reportedly in late-stage discussions for a Series B at a valuation of approximately $1.2 billion. The terms are not final and could change.

Who is reportedly leading XDOF’s Series B?

According to the supplied TechCrunch report, 8VC is reportedly leading the potential Series B.

How much has XDOF previously raised?

TechCrunch reported that XDOF raised a $70 million Series A in June 2026. Investors included Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital.

Why does XDOF collect robot training data?

XDOF collects real-world data because physical robots do not have an equivalent to the enormous internet-scale datasets available to many AI models. The company aims to provide robotics companies with data pipelines, collection tools, and annotation infrastructure.

What is GELLO?

GELLO is a low-cost teleoperation system developed by XDOF co-founders Philipp Wu and Fred Shentu during their research. It allows a human operator to remotely control a robotic arm to generate training data.

Final Takeaway

XDOF’s reported Series B is a reminder that the next phase of AI may depend on something surprisingly old-fashioned: people doing things in the real world.

Every folded shirt, flattened box, robot demonstration, and recorded human movement can become a potential training example. XDOF is attempting to turn those physical interactions into a scalable data supply chain for the robotics industry.

Whether the company ultimately reaches the reported $1.2 billion valuation remains uncertain because the financing has not been finalized. But the underlying problem is real and increasingly important: if AI is going to move from screens into the physical world, it needs data from the physical world too.

For more explainers on robotics, physical AI, startup funding, and the technologies shaping India’s future workforce, keep exploring Kalinga.ai.