Baidu, Galbot back robot-data startup Genrobot AI as "picks-and-shovels" bet heats up
What's the deal? Genrobot AI, a Chinese startup building training-data infrastructure for embodied AI, has added Baidu and Galbot subsidiary Suzhou Gaibote TechnologyDealroom has a profile for this one. Try Dealroom → as shareholders, according to business registry Tianyancha. The round lifted its registered capital from about US$26.5K to roughly US$1.45M.
What does it do? Genrobot doesn't build robots. It uses wearable devices to capture how real people perform tasks, then turns that "bodiless data" into training material for robot AI models.
By decoupling data collection from hardware, the data can transfer across robot types and scenarios — lifting efficiency over costly, limited teleoperation-based capture.
Why now? Capital in embodied AI is shifting from robot bodies and "brains" to the upstream data layer. Simulated data can't replicate reality's uncertainty, and real-machine collection stays expensive — making high-quality physical data the sector's bottleneck.
What's the track record? Founded in May 2025, Genrobot has raised five rounds in 18 months. Baidu VenturesDealroom has a profile for this one. Try Dealroom → entered at the seed stage, backed it through Pre-A, with Baidu then taking over. Ant Group, Didi, Sinovation Ventures, and others have also invested, totalling several hundred million yuan.
Founder Chen Jianxing spent a decade in autonomous driving, including as a senior algorithm director at MomentaDealroom has a profile for this one. Try Dealroom →, where he led data-flywheel production. Genrobot applies that proven methodology to embodied-AI data collection.
What's the traction? Early this year the company released an open-source dataset of more than 13,000 hours of bodiless embodied-AI data. It reports monthly downloads above one million, ranking first globally among open embodied datasets by downloads and academic citations.
Commercially, it spans home, industrial, logistics, and medical uses, with thousands of skills and over one million hours of data, and has partnered with more than 30 AI companies.
The investors' logic: Baidu, which builds large models and embodied-AI products, needs real-world data to refine them. Galbot makes humanoid robots and gains training fuel by backing an upstream supplier — a classic supply-chain play. Ant and Didi are positioning for the next generation of general-purpose robots.
The signal: Genrobot's run reflects a broader pivot toward infrastructure. In the first half of 2026, 25 Chinese embodied-AI data startups raised more than US$2.34B combined, according to IT Juzi — evidence that real, high-quality physical data is becoming the industry's scarcest asset.
Read more: sohu.com
Image credit: jurvetson
