XDOF, a startup that supplies the human-demonstrated action data robotics firms use to train general-purpose machines, is closing in on a Series B that would value it at about $1.2B, with 8VC in line to lead, TechCrunch reported. Terms remain fluid and no agreement is final.
The company effectively sells the raw material of robot intelligence: collection rigs, data pipelines and annotation tools that frontier labs would rather buy than build themselves. Co-founder Philipp Wu has said the scarcity of large-scale training data once stalled his own academic research, and XDOF commercializes the workaround he developed.
Founded in 2024 by Wu and Fred Shentu, both UC Berkeley researchers, XDOF left stealth about three months ago after banking a $70M Series A in June from Thrive Capital, Andreessen Horowitz, Lux and Spark Capital. An earlier project called GELLO showed how a low-cost rig could let a single operator remotely drive a robot arm to capture demonstrations.
Growth has outpaced the founders’ plan. Annualized revenue is approaching $50M, and investors approached them about a new round rather than the reverse, according to people familiar with the talks. The size of the raise has not been disclosed.
