Guanghe Capital partner: we bet on Unitree as R&D infrastructure, embodied AI bar has risen
创业邦 · wechat · 2026-11-17
At the 2026 DEMOCHINA podcast in Hangzhou, Guanghe Capital partner Zhu Jia explained his embodied AI investment logic. He backed Unitree early as a contrarian bet, viewing it not as a robot maker but as R&D infrastructure — analogizing to Nvidia selling GPUs to researchers before ChatGPT triggered inference demand.
For new embodied AI startups, he says the bar has risen:
- Hardware must show strong locomotion plus fine manipulation (e.g., picking a bottle off a human-sized convenience-store shelf);
- Models must demonstrate real generalization — deep RL on single tasks like folding clothes has limited value. He tested Ziz Var by grabbing a dozen bottles off the street: wobbly, trial-and-error like a 3-year-old, but promising;
- Architecture will be cloud-local hybrid: cloud LLMs handle reasoning and task decomposition, local "big/small brains" execute each step. VLA brains can go into non-humanoid forms, such as portfolio company Weifen Zhifei's palm-sized autonomous drone.
He also warns much of today's hype is unfalsifiable over-optimism, with landing timelines widely overestimated.
More from Venture
- Indie dev shares inbound DM: buyer asking to acquire his tool Steercamp — pramodk73 · 2026-11-22
- Anthropic's record ~$81B IPO, GPT-6.2 pricing, Qwen4-Max and Copilot Autopilot: week in review — ksraj1001 · 2026-11-22
- Users want frontier AI at bargain prices while complaining labs aren't profitable — VraserX · 2026-11-22
- MCP server checks if a pay-per-call x402 API is worth paying, from 24,766 real paid calls — biryaniAssasin · 2026-11-22
- Plausible took 324 days to hit $400 MRR, then a HN post fueled $500K ARR — sven_ai · 2026-11-22
- Token economics model projects $1.14T global API spend by 2030, but only 1% of users pay beyond coding — AccBalanced · 2026-11-22