NTU-Born Ropedia Raises US$38.7M to Build the Data Pipeline Physical AI Depends On
Source: Singapore Business Review
Language models learned everything they know from the internet. Robots can't — there is no equivalent corpus of real-world movement, depth and interaction for them to train on. That gap between what machines can read and what they can physically do is the quiet bottleneck holding back physical AI...
Language models learned everything they know from the internet. Robots can't — there is no equivalent corpus of real-world movement, depth and interaction for them to train on. That gap between what machines can read and what they can physically do is the quiet bottleneck holding back physical AI, and a Singapore startup barely a year old is betting tens of millions that it can fill it.
Ropedia, founded in 2025 by CEO Chen Zhaoxi, Hong Fangzhou and NTU associate professor Liu Ziwei, announced a US$28.4 million pre-Series A in July that brings its total funding to US$38.7 million. The company grew out of Nanyang Technological University, where the founders worked on methods and patents for collecting and processing human-motion data. Its Homie device captures egocentric video, depth, motion and audio from everyday human activity, which the platform synchronises and processes into datasets sold through the Xperience-10M dataset and custom data-as-a-service offerings.
Ropedia is targeting 1 million hours of high-quality data by the end of 2026, plus a proof-of-concept model validation run. The new capital will expand its R&D team, develop wrist and full-body data-capture devices, strengthen its supply chain and push into North America and Southeast Asia. Chen told Singapore Business Review that technology companies are already spending heavily on data infrastructure to train AI models — and that deploying those models in physical environments will create its own demand for real-world data.
The bet is that whoever owns the data layer becomes indispensable. Chen's five-year ambition is for Ropedia to play the role for physical and spatial AI that Cloudflare, Databricks and AWS play for internet and cloud infrastructure. It is a lofty comparison for a startup that began as a university project, but it speaks to how early this market is: there is no dominant standard yet for physical AI training data, and Singapore's robotics and AI research base gives local players a credible starting point.
Why it matters for Singapore: Ropedia is the kind of company Singapore's AI strategy is designed to produce — research spun out of a local university, funded by regional venture capital, and aimed at a global infrastructure layer rather than a consumer app. With NTU and A*STAR pushing robotics research and government funding flowing into physical AI applications from healthcare to port operations, a homegrown data infrastructure player gives the ecosystem a stake in how the next generation of AI gets trained, not just how it gets used.
