A 22-Year-Old CEO, Five Rounds, Over $100 Million: The First Term Sheet Decides Everything
In a year and a half, 22-year-old Huang Yi has grown RoboParty to nearly 150 people: five rounds totalling over $100 million, and RPO shipping around 100 units a month. His calls: the first round's terms decide whether you live or die, open source is a moat because it lowers the friction of working with you, and the robot body is the entry point to embodied AI.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
What you should fear is not a board seat but personal liability
At the seed round you want as many cards in hand as you can get, but what matters more are the terms of that first agreement — they will carry over essentially unchanged all the way to Pre-IPO. Huang Yi stresses that the company still has no redemption rights and no VAM clauses, and that this comes down to negotiating the first round on friendly enough terms. He is not afraid of giving up a board seat: as long as the shareholder genuinely helps (Xiaomi, for instance), handing over a seat actually makes the collaboration run smoother. What you really have to guard against are the buried traps — personal joint liability, redemption with simple or compound interest — which is why you must find a lawyer who has negotiated a large volume of investment terms.
— Huang YiDon't dodge the bubble; it is the industry's momentum
Huang Yi flatly concedes ‘of course there's a bubble’, but he treats the bubble as the industry's momentum: suppliers give you a second look because you are in embodied AI, and talent pours in from other sectors. Shunwei taught him to ride the trend — no company can change its era, you can only build your company while the heat is on; the counterexamples are photovoltaics and real estate, where nobody is looking. He also explains where the money comes from: the secondary market has capital, another bag of money becomes an LP inside a GP, and that turns into ammunition for startups. The sentiment cycle is the fundraising window.
— Huang YiTake everything you can raise, but not every check is worth taking
RoboParty closes a round every one to two months, which looks insane, but behind it sits a clear set of milestones: shipping the RP1 launch plus volume production is the Series A bar, and closing the commercial loop is the Series B bar. Huang Yi's logic is that on the industry's way up you take as much as you can take, because missing the window may cost you another year of waiting. But he also insists that not all money should be taken: dollar funds weigh technical leadership, RMB funds care more about commercialization, and the later the stage, the less professional the investors he tends to run into.
— Huang YiThe value isn't the robot body, it's the empty layer between model and body
Asked ‘how do you make money if the whole stack is open source’, Huang Yi offers two concepts: a threshold is how much fixed assets and headcount you put in, while a moat is where the money is. He positions RoboParty in the missing layer between the model companies and the vendors that have commercial robot bodies — the model companies have no body, and existing body makers won't open up the low levels, so the motor layer, URDF snapshots and data-collection interfaces are all absent. Open source lowers the friction of working with you and widens the brand; selling bodies is the base business, but the compounding from users is bigger: one CMU professor buying a body effectively brings in two PhD engineers at million-level annual salaries who keep publishing papers.
— Huang YiThe honeycomb only suits the Lab; manufacturing has to have process
A company of nearly 150 people runs on a ‘honeycomb structure’: each person has at most six effective connection surfaces, driven by talent density rather than hierarchy — an idea taken from Anthropic. But Huang Yi says plainly that the honeycomb only suits the Lab; if the whole-machine division also went talent-driven, you get the chaos of ‘nobody even knows who to buy parts from’. So the company carves out a pure Lab that runs on honeycomb logic, while the manufacturing division must have process, and the two sides absorb each other's knowhow through a communication-style meeting system.
— Huang YiEmbodied gross margins are double the carmakers', so the giants will come
Huang Yi offers a read on the landscape: there are roughly 300 embodied companies today, and by 2030 perhaps only 10 to 15 will make it into the next round. He uses gross margin as the yardstick: embodied runs around 40% while carmakers are at only 20%, so the next wave of players will be companies at the hundreds-of-billions scale — NIO, XPeng and Li Auto, Honor, Geely. Only when BYD, Huawei, CATL, BAT and Xiaomi finally step in does ‘the big era’ truly begin. He is also bullish on model companies like OpenAI and Zhipu, arguing that once they fill in their robotics understanding, their scaling ability far exceeds a startup's.
— Huang YiChina doesn't reward explorers, it rewards the people who pull it off
Asked which embodied companies in China and the US he admires most, Huang Yi names Tesla, Figure, and Shuran Song's team Generalyst — the ones willing to explore paradigms nobody has attempted, which in the US is what attracts the most money. But he says outright that Chinese culture does not reward the explorer; it rewards the person who ‘can get a company off the ground’. He has deliberately chosen the second path, and he gives an extremely concrete standard for success: that his parents, without knowing he built this robot, would be willing to buy one and take it home.
— Huang YiThe body is halfway through; the brain doesn't even know where the finish is
Using a marathon as the metaphor for embodied progress, Huang Yi splits it into segments: the body is already at the halfway mark, with iteration and convergence left; the cerebellum (whole-body coordination) is also about halfway, and the behavior foundation model can already handle interactions like ‘carrying something back to its original position’. On the brain he is most candid — ‘I don't know what number it is’ — because nobody has seen the endgame. He thinks the endgame is ‘an agent that has accumulated your entire history and reacts exactly the way you would’, and by that definition we have run only one or two kilometers today. This is the one place in the whole conversation where he admits a gap in his own understanding.
— Huang YiIn their own words · checked verbatim
It was the first-generation robot, and once we'd finished it we felt it wasn't much use, so we open-sourced it.
我是第一代机器人 然后就做完了感觉没啥用 然后把它开源了
Huang Yi19:18
I think by 2030, maybe 10 to 15 companies will come through.
我觉得2030年吧 可能会有10家到15家走出来
Huang Yi46:43
Figures
| Rounds and amount raised | 5 rounds within one year, over $100 million | 1:00 |
| Team size | about 140 people | 1:00 |
| RPO monthly shipments | about 100 units (2025) | 1:00 |
| RP1 sales target | about 2,000 units for all of next year | 1:00 |
| Time with zero voluntary departures | one and a half years | 43:41 |
| Number of embodied companies and the shakeout | about 300 today, an expected 10-15 left by 2030 | 46:43 |
| Gross margin comparison | embodied about 40%, autos about 20% | 47:43 |
| RPO customer mix | about 70% research and education, 10% factories, 20% companionship and entertainment | 53:45 |
| Large overseas order | 500 units, only about twenty or thirty delivered | 31:33 |
Glossary
- missing layer
- The open platform layer absent between model companies and robot-body makers, reaching down into the motor level, URDF and other low-level pieces.
- deep pocket
- A well-capitalized institutional investor that can keep writing checks in later rounds.
- BFM (behavior foundation model)
- A behavior foundation model on the cerebellum side, used for whole-body motion generation and object interaction.
- sim2real
- The technical path of transferring policies trained in simulation onto real robots.
- honeycomb structure
- Huang Yi's own org form: each person manages at most about six others, driven by talent density rather than hierarchy.
How to listen
AI and robotics founders about to raise, primary-market investors covering the embodied space, and engineers curious about how a young CEO handles open source and org design.
You can skip the opening rapid-fire round and the AI-tools chat starting at [59:50]; do not miss the ‘marathon splits’ at the end.