XDOF Is Betting That Robots Need Their Own Data Boom

The next valuable AI dataset may not come from the internet. It may come from humans folding clothes and moving boxes.

XDOF emerged from stealth only a few months ago.

It is already reportedly discussing a new funding round that could value the robotics data startup at around $1.2 billion.

The potential Series B is reportedly being led by 8VC. XDOF had already raised a $70 million Series A in June, with investors including Thrive Capital, Andreessen Horowitz, Lux and Spark Capital.

What changed so quickly?

Growth.

XDOF's annualized revenue is reportedly approaching $50 million, prompting investors to approach the company sooner than it originally expected to raise again.

But the bigger story is what XDOF sells.

Robots have a data problem

Large language models had a useful starting point: the internet.

Billions of webpages, documents, books, images and other digital resources offered enormous datasets for training AI systems.

Robots face a different problem.

There isn't an internet-sized database showing machines exactly how humans physically manipulate objects in the real world.

Opening drawers.

Folding shirts.

Packing boxes.

Picking up unfamiliar objects.

Handling tools.

Those actions contain huge amounts of physical information that models need if general-purpose robots are going to become reliable.

XDOF wants to build that data layer.

Humans are training the machines

The startup uses teleoperation and sensor-based collection systems to record real-world actions.

Human operators can remotely control robotic arms, while other data collectors wear sensors that capture physical movements while performing everyday tasks.

That information can then be used to help train robotic AI systems.

XDOF says it is already working with around 20 customers, including several frontier AI labs. It is also working with UC Berkeley's AI Research lab on a large robot-training dataset.

In other words, the startup isn't primarily trying to build the robot.

It wants to build the data infrastructure robots need.

We've seen this playbook before

There is a familiar pattern here.

Scale AI became a major company by providing the data infrastructure necessary for AI models.

Companies such as Mercor and AfterQuery pushed the same idea toward expert and professional knowledge.

Robotics could create another version of that market.

Except this time, the scarce resource isn't text or professional reasoning.

It's physical experience.

That could create an entirely new category of data businesses.

Why investors are moving early

The robotics industry is having its own AI moment.

Advances in multimodal models are giving machines a better ability to understand images, instructions and physical environments.

But software intelligence alone isn't enough.

Robots need enormous amounts of high-quality real-world interaction data.

If XDOF can become one of the main suppliers of that data, it could occupy an important position regardless of which robotics manufacturer ultimately wins.

That is likely part of what investors are pricing into a billion-dollar valuation.

What happens next?

The most interesting startups in robotics may not all manufacture humanoids.

Some will provide simulation.

Others will develop sensors, foundation models, teleoperation systems or data pipelines.

XDOF is betting that the data layer will become one of the most valuable.

It's a familiar startup strategy: don't necessarily compete in the most visible part of a new market.

Find the bottleneck underneath it.

For physical AI, data may be exactly that bottleneck.