XDOF raises $70M to build the data pipelines for training robots
What's the deal? XDOF, a startup pronounced "ecks-doff," has emerged from stealth with $70M in funding to tackle what it sees as the biggest bottleneck in robotics: training data. The round was backed by a16z, Lux, Spark Capital, Thrive Capital, and WndrCo.
The company, founded in October 2024, builds data pipelines, collection tools, and annotation systems for frontier AI labs and robotics companies. It has about 60 employees and is already working with 20 customers, including several major AI labs.
Co-founder and chief executive officer Philippe Wu spotted the problem during his PhD at UC Berkeley, where he focused on teaching robots to learn from large-scale datasets. "We didn't have large-scale data to work with," he said. "There was this chicken-and-egg problem — we first needed to actually collect data before we could even ask how to train a foundation model for robotics."
Why now? In Q2 2026, OpenAI announced it would relaunch its robotics programme, shuttered in 2021 — the latest sign that the biggest AI labs are racing to teach machines to operate in the physical world. Unlike large language models trained on publicly available text, robots need data capturing physical interaction, and that kind of data barely exists.
"All of the top labs are trying to pursue robotics," Wu said. "You don't want to be in this type of situation where you pursue this technology too late."
As a starting point, XDOF is partnering with UC Berkeley's AI Research lab to release what it calls the largest collection of high-quality robot training data ever assembled — dubbed ABC. It includes 130,000 trajectories of robot manipulation data, 300 hours of simulation, and 100 hours of evaluations. Researchers have already used the data to train robots on tasks like folding T-shirts and loading AirPods into their cases.
What could go wrong? Data provision alone can be a commodity business with thin margins. XDOF is betting that layering in data cleaning, tooling, and annotation will create a self-reinforcing feedback loop that's harder to replicate. But frontier labs are well-resourced and could build similar capabilities in-house if the market proves valuable enough.
Wu and co-founder Fred Shentu previously built GELLO, a low-cost teleoperation system that lets a human operator control a robotic arm to generate training data. It became an influential paper in robotics — but turning academic influence into a durable business is a different challenge entirely.
The signal: This funding reflects a broader conviction that physical AI is entering its infrastructure phase. Just as the language model era spawned billion-dollar businesses in data labelling and compute, the robotics wave is creating demand for entirely new data supply chains. XDOF is positioning itself as the picks-and-shovels play for that transition — betting that whoever controls the data pipelines will shape how quickly robots learn to navigate the real world.
Read more: TechCrunch
