An XDOF Series B valuation of roughly $1.2 billion is taking shape in late-stage talks, with venture firm 8VC set to lead the round, according to people familiar with the matter. The terms are not final and could still change. Neither XDOF nor 8VC responded to requests for comment.
What makes the deal unusual is its timing. XDOF only emerged from stealth in June, closing a $70 million Series A backed by Thrive Capital, Andreessen Horowitz, Lux Capital, Spark Capital, and WndrCo. The company was not planning to raise again so soon. Investors, it seems, had other ideas.
From a Berkeley Lab to a Billion-Dollar Bet
XDOF was founded in October 2024 by UC Berkeley researchers Philipp Wu, who serves as CEO, and Fred Shentu, who serves as CTO, alongside a third co-founder, Nemo Jin, according to AI Weekly. The company’s origin runs directly through a piece of academic work: a low-cost teleoperation system called GELLO, which Wu and Shentu developed as PhD students.
GELLO was submitted to arXiv in September 2023, well before XDOF existed as a company. The system uses 3D-printed parts and off-the-shelf motors as joint encoders to let a human operator control a robotic arm remotely, generating training data as it goes. The paper was published at the IEEE/RSJ International Conference on Intelligent Robots and Systems in Abu Dhabi in October 2024, with UC Berkeley professor Pieter Abbeel among the listed co-authors. It has since accumulated 386 citations, per Semantic Scholar.
Wu has described the bottleneck he encountered during his research plainly: a lack of ‘large-scale data to work with.’ That problem became the company’s entire business model.
XDOF positions itself as the outsourced data-supply chain for the robotics industry, building pipelines, collection tools, and annotation systems that AI labs and robotics companies would struggle to assemble internally. Investors have taken to comparing it to Scale AI or Mercor, the data-labelling platforms that fed the large language model boom. The analogy holds insofar as physical robots face a data problem LLMs never did: the internet provided a ready corpus for training text models, but no equivalent real-world dataset exists for teaching a robot to fold clothes or load a small case.
The XDOF Series B Valuation in Context
What appears to have accelerated investor appetite is revenue. Annualised sales are approaching $50 million, and the company already counts 20 customers, including several frontier AI labs. That trajectory, sources say, prompted VCs to seek out XDOF rather than the reverse.
The total capital being raised in the new round is not yet known, and it is also unclear whether the $1.2 billion figure is pre- or post-money. The XDOF Series B valuation is being watched closely enough that the lead investor’s own positioning matters. 8VC closed its seventh fund at $1.5 billion in July 2026, its largest to date, built around what the firm describes as an ‘American AI Industrial Revolution’ thesis covering AI, defence, advanced manufacturing, and healthcare, according to F4 Fund’s firm profile. A robotics data infrastructure company sits squarely within that mandate.
On the data side, XDOF has been building its collection capacity across three tiers: robots operated by remote teleoperators using the GELLO device, robots steered through standard deployment setups, and egocentric operators who wear body sensors to record everyday human movements. Tasks in the dataset include folding shirts and loading AirPods cases.
In partnership with UC Berkeley’s AI Research lab, the company has released what it calls the ABC-130K dataset: 130,000 bimanual robot manipulation trajectories, described as the world’s largest open-source collection of its kind, along with 300 hours of simulation data and 100 hours of evaluations, according to SiliconANGLE. Releasing that dataset publicly is partly a recruitment tool: it signals the quality of the pipeline to potential customers and draws researchers into the orbit of the company’s standards.
XDOF is not alone in targeting this gap. Mecka AI is pursuing a similar strategy, and established data platforms such as Scale AI are expanding into physical robotics from their LLM-era bases. The competitive window, as 8VC’s interest suggests, may be narrower than the $1.2 billion price tag implies: the company that locks in the frontier labs as anchor customers earliest will be hardest to displace. Whether the Series B closes at that valuation, or the terms shift before signing, is the immediate question.
