Every humanoid robot you will eventually see on a factory floor has to learn like an apprentice first — by watching someone do the job. The unglamorous bottleneck is that someone has to record, clean, and deliver those lessons at industrial volume. That is the business that just pulled in nine figures.
GenRobot (简智新创（北京）机器人科技有限公司, known as 简智机器人), a Beijing data and evaluation infrastructure company founded in 2025, disclosed a US$100 million (≈ RMB 710M / HK$780M) Series A+ round. The financing closed in the third quarter of 2026 and was made public on November 2. The investor list splits cleanly into two camps: strategic and corporate backers — SAIC Motor, Malaysia's Eastworlds, Baidu Group, Shoucheng Holdings, and embodied robotics developer Galaxy General (brand name GALBOT) — and financial institutions including Guanghe Ventures, SUNHELM, CITIC Goldstone, Zhongzheng Investment, CCB International, and Huatai Zijin.
The company does not build robots. It builds the data pipeline robots learn from — and that distinction is exactly where the money is flowing this year.
From human demonstrations to training-ready data
GenRobot's premise comes straight from its founders' history. CEO Chen Jianxing was a senior algorithm director at Momenta, where he planned the data flywheel (数据飞轮) mass-production strategy for autonomous driving; co-founder Zhu Yanming spent years on mass-production development in the same industry. Hardware team members came from DJI and Huawei, the software team from autonomous driving algorithms and multimodal model backgrounds. They are applying driving's data engineering playbook to a domain with no fleet of cameras already on the road.
A robot learning to pick up a folded towel, stock a supermarket shelf, or assemble a component needs records of vision, hand and body motion, and contact — synchronized and labeled. GenRobot's product stack covers the full arc:
- Gen DAS — wearable capture hardware that records a person's viewpoint, hand movements, and manipulation actions in real settings such as homes, supermarkets, and factories.
- Gen Matrix — a data platform that synchronizes inputs across devices, cleans and labels recordings, and turns raw captures into model-ready training material.
- Gen ADP — a crowdsourced production line that organizes large-scale collection through an app, where contributors receive tasks, wear the devices, perform everyday operations, and upload the results for settlement.
The company also trains its own Data Foundation Model to process and reconstruct captured information. For customers with proprietary environments — factory lines, closed labs — GenRobot sells hardware and local deployment so data stays inside; for open, everyday scenarios it sells processed data produced through crowdsourcing.
The scale metric investors are watching
Delivery volume is the number that anchors this round. The company says effective data delivered has grown from roughly 100,000 hours per quarter to the million-hour-per-quarter level, verified across multiple physical AI model projects worldwide. Its stated target is ten million hours per quarter by 2027. Those are company figures rather than audited statistics, but the trajectory is the core of the pitch: data supply, not model architecture, is what most embodied teams cite as their binding constraint.
Why strategics, not just funds, wrote the checks
- Carmakers are hedging the physical AI transition. SAIC's participation mirrors a broader pattern of automakers investing upstream in embodied data, the same way they once invested in autonomous driving stacks. Malaysia's Eastworlds adds a cross-border angle unusual for this stage.
- Robot makers are funding their own supply chain. Galaxy General investing in a data vendor it presumably buys from looks circular at first glance — but it locks in capacity at the layer both parties agree is scarcest, similar to how AI labs pre-purchase computing capacity.
- The "picks and shovels" rotation reached embodied intelligence (具身智能). After 2026's wave of large rounds for robot makers themselves, capital has moved one layer down to whoever instruments the learning process. GenRobot's prior financing shows the cadence: several rounds totaling several hundred million yuan announced in June, led by Ant Group, DiDi, and Delian Capital, then an A round led by Momenta in August with an undisclosed amount, and now this US$100 million step.
What the round does not resolve is pricing power. Data infrastructure companies sit between contributors who supply demonstrations and model developers who have alternatives, including building collection in-house. Whether GenRobot's pipeline becomes a durable toll road or a compressible line item is the question the next two years will answer.
Key takeaways
- GenRobot raised a US$100 million (≈ RMB 710M / HK$780M) Series A+ closed in Q3 2026 and disclosed on November 2, with SAIC Motor, Eastworlds, Baidu, Shoucheng Holdings, Galaxy General and financial investors including CITIC Goldstone and CCB International.
- Founded in 2025 and headquartered in Beijing, the company runs a capture-to-evaluation stack — Gen DAS hardware, Gen Matrix data platform, Gen ADP crowdsourced collection — for embodied and world-model developers.
- Reported delivery volume has reached the million-hours-per-quarter level, with a company target of ten million hours per quarter by 2027.
Honest limitations
- The US$100 million figure, the investor list, and the delivery-volume metrics all originate from company announcements relayed by Chinese financial media and the Beijing municipal science and technology commission; no independent audit or filing verifies them, and the amount of the earlier August round was never disclosed.
- Valuation was not disclosed, so the round's pricing and dilution cannot be assessed from public information.
- The 2027 ten-million-hour target is a company projection contingent on crowdsource supply and customer demand; the company publishes no methodology for how "effective hours" are counted, which limits cross-vendor comparability.
What readers can do now
- If you develop embodied or world models, price the build-versus-buy decision explicitly: compare third-party data hours against the fully loaded cost of your own capture fleet before defaulting to either.
- If you invest in or watch the robotics supply chain, track whether data vendors begin publishing standardized delivery and quality metrics — standardization is usually the precursor to industry-wide procurement.
- If you are considering new career directions, demonstration collection, data curation, and robot evaluation are emerging roles that sit upstream of model training and require domain skill rather than deep learning research credentials.