XDOF Enters Series B Talks at a Reported $1.2 Billion Valuation

XDOF is in late-stage talks to raise a Series B at an approximately $1.2 billion valuation, with 8VC expected to lead, according to people familiar with the deal. The proposed financing draws investor attention to XDOF’s approach of collecting and preparing demonstrations of physical tasks for robotics customers.

The company was founded in 2024 by UC Berkeley researchers Philipp Wu and Fred Shentu. Their earlier GELLO project used a low-cost system that let a person remotely control a robot arm to generate training data; Wu had said the lack of large-scale datasets had constrained his research. That work became XDOF’s foundation.

A second financing discussion follows quickly

XDOF raised a $70 million Series A in June, with participation from Thrive Capital, Andreessen Horowitz, Lux and Spark Capital. People familiar with the company said annualized revenue is approaching $50 million and that its growth prompted investors to approach it, despite XDOF not planning to raise again so soon.

$1.2 billionProposed valuation

People familiar with the deal described a late-stage Series B at this approximate valuation, with 8VC expected to lead.

Approaching $50 millionReported annualized revenue

The revenue figure was provided by people familiar with the company.

The data pipeline behind the pitch

XDOF develops teleoperation data, collection tools and annotation systems for frontier AI labs and robotics companies. It collects material through remote operators steering robots and people wearing sensors while doing everyday tasks, including folding clothes and flattening boxes. The company previously said it was working with 20 customers, including several frontier AI labs.

Two ways XDOF captures task data

• Remote teleoperators steer robots to create demonstrations for training.
• Sensor-wearing human collectors record their movements while carrying out everyday physical work.

An open dataset, and an unfinished deal

In June, XDOF said it had worked with the authors of the ABC behavior-cloning project and made its evaluation service available to them. The company says ABC-130K contains more than 130,000 episodes across 195 bimanual manipulation tasks, using station setups with YAM arms from I2RT. It says the listed ABC datasets are available under an Apache 2.0 license.

The financing has not closed. The amount being raised and whether the reported valuation includes the new funding remain unknown, and terms could still change. XDOF and 8VC did not respond to TechCrunch’s requests for comment.