99%+ data gap: Sophia Capital partner on embodied-AI investing

创业邦 · wechat · 2026-10-14

Shangshu Guoxiang Capital partner Yu Jun shared his robotics investing framework at DEMOCHINA, favoring full-stack teams that integrate models, data, hardware and product over pure world-model startups.
Data is the hard constraint: global high-quality real-world interaction data is 500K hours vs 10M+ hours needed for generalist embodied models — a gap exceeding 99%; OpenAI has already procured over 10M hours, more than any Chinese embodied-AI firm.
Commercialization must pencil out: he suggests benchmarking robot hardware cost against 1–2 years of a worker's wages; demo demand (traffic-driving robots) counts as real demand today.
His rule: only two chances to invest — seed and Series A; he also backs young founders while pushing them to learn competitive strategy.

More from Venture