When the VR and metaverse bubbles burst, taking his company from 300 employees down to 70, Noitom Robotics CEO Dai Ruoli started running. Not from his problems, but around a track. In two months, he lost 40 pounds. He wasn’t obsessed with his physique; he was trying to survive. When the market goes to hell and your company’s future is out of your hands, controlling the number on the scale is the only way to remind your brain that you still have agency.
In the AI gold rush, capital is the adrenaline, and the founder is just the heartbeat strapped to the machine.
Now, Dai is standing in the middle of his third hype cycle: Embodied AI. And if you think the robotics revolution is about who can build the coolest humanoid hardware, you’re looking at the wrong battlefield. The hardware is just the shiny wrapper. The real war—the one that will dictate who owns the next decade of physical AI—is entirely about data infrastructure.
Right now, everyone is obsessed with teleoperation. Engineers are strapping on VR headsets and controlling robot arms to teach them how to open doors or fold laundry. It works, but it’s a trap. The data you collect on a Unitree G1 robot barely transfers to a Unitree H1, let alone a competitor’s machine. The sensors, actuators, and body proportions are too different. It is bespoke, incredibly expensive, and fundamentally unscalable. You’re paying $100 an hour to teach a machine a trick it can only perform in one specific body.
Everyone is screaming about robot hardware, but the real bottleneck isn’t steel and servos—it’s the raw, scalable data of human interaction.
Dai Ruoli’s bet is radically different. He is building a ‘World Compiler.’ Instead of teleoperating robots, Noitom captures high-fidelity motion data from humans. Why? Because human data is cross-embodiment reusable. You capture a human picking up a cup, and that data can be compiled and transferred to a humanoid robot, a dog robot, or a factory arm. It turns bespoke data collection into a standardized, sellable commodity. It’s the ultimate moat in an industry desperate for scale.
But here is where the market is currently losing its mind. Drunk on the Scaling Law, companies are racing to the bottom to capture EGO (egocentric) data. They’re handing cheap head-mounted cameras to anyone who will wear them, paying rural laborers pennies on the dollar to record mundane tasks. They think volume will save them. They are wrong.
In AI, volume isn’t a magic cure-all. Feed your model 10,000 hours of garbage, and you don’t get a smarter robot—you get a degraded one.
Research has already proven that adding 10,000 hours of flawed EGO data to 10,000 hours of high-quality data actually degrades the model’s performance. The low-cost EGO data race isn’t a breakthrough; it’s poison. Quality control, not sheer volume, is the hidden battlefield. The company that can guarantee high-fidelity, cross-embodiment data will own the bottleneck. The companies chasing cheap volume will drown in their own noise.
What makes Dai Ruoli fascinating isn’t just his technical foresight. It’s that he is a ‘classical entrepreneur’ trapped in a modern meat grinder. He is a man who writes thousand-word essays on the texture of pineapple buns. A man who quietly mourned his 16-year-old cat, Mocha, on a flight home. A man who, upon discovering a former colleague had been laid off and was delivering food to survive, hired him back immediately without a second thought.
In the brutal meat grinder of startups, gentleness and kindness aren’t the most celebrated traits—but they are the ones that build lasting moats.
Yet, even a founder who just wants to build durable tech is forced to play the game. Dai openly admits he feels a ‘strong sense of being hijacked’ by the current funding environment. His company has enough money in the bank, but the FOMO of the embodied AI market forces him to keep raising, keep inflating valuations, and keep feeding the capital machine. He is a humane founder forced to play an inhumane game.
When his partners insisted on removing ‘gentle and firm’ from the company’s core values because it sounded too weak for a aggressive startup, Dai relented. But every time he presents his pitch deck, he quietly sneaks the phrase back onto the slides. It’s a small act of rebellion, but a deeply revealing one.
If the embodied AI revolution is really going to succeed, we don’t just need the most ruthless algorithms or the cheapest data pipelines. We need people who actually love the physical world enough to translate it for the machines. In a robotics war defined by cheap poison and hype, the most human founder might be our safest bet.
FAQ
Q: What is the key takeaway?
A: See the article.