Radar – by nexxworks: China's Robotic Economy 🤖

Nvidia says this is "the ChatGPT moment of robots." A Unitree humanoid just ran a Beijing half marathon in under 50 minutes, faster than any human alive, and a year ago it needed 2h40. So that's another thing you're now worse at than a machine.

In this special robots edition of Radar, Pascal Coppens and Dado Van Peteghem dig into how fast China is pulling ahead, ahead of the nexxworks China Robotic Economy tour they'll co-moderate this November through Beijing, Shenzhen and Guangzhou.

The short version: China owns 90% of the magnets every robot needs, Unitree's price crashed from €85,000 to €23,000 in two years, and JD.com is paying 500,000 people €3 an hour to wear cameras all day so robots can learn to do the dishes. Meanwhile Europe is still deciding whether all this is "deep future." We did the same thing with EVs. It went great.

Pascal and Dado also get into the AI actually running these machines (world models and vision-language-action systems), Galbot's underground robotic pharmacies, DeepSeek's fresh $7 billion, and Asimov's three laws. The takeaway from Dado: stop reading about the future and go shake its very dexterous hand.

‍

‍

Takeaways:

• 🤖 Nvidia founder Jensen Huang says "the ChatGPT moment of robots" is here, with robotics already contributing $6 billion to Nvidia's own P&L.
• 🏃 A Unitree humanoid ran a Beijing half marathon in under 50 minutes, down from 2h40 a year earlier, faster than any human on the planet.
• 🧲 China holds a ~90% monopoly on the magnets every robot needs; without Chinese components, Musk's Optimus would cost roughly 3x as much.
• 🔋 The four bridges to scale are uptime, dexterity, safety and cost, with battery-swapping stations already chipping away at the uptime problem.
• 🧠 Robots now run on two new AI layers beyond LLMs: world models (Fei-Fei Li's World Labs, Yann LeCun's lab) to train them, and vision-language-action (VLA) models to operate them, as with the Tsinghua-trained Unitree robot that can play tennis.
• 📸 JD.com pays 1,000 of its own employees plus 500,000 externals €3 an hour to wear cameras and sensors, harvesting real-world household data at scale.
• 💸 Chinese AI runs 5–10x cheaper per watt than in the US, and around 80% of US startups already build on Chinese open-source models like DeepSeek and Kimi.
• 📉 Prices are collapsing: Unitree's robot fell from €85,000 to €23,000 in two years, and Galbot's box-picking robot dropped from €150,000 to €100,000 in just two months.
• 🏭 China already accounts for 54% of all industrial robots sold worldwide; the government wants 10,000 robots deployed across 100 use cases by the end of 2026.
• 💊 Real deployment is already here: Galbot's underground robotic pharmacies, elderly and healthcare, and dangerous rescue work lead the first wave.
• 💶 Europe finally has a level playing field, but after sleeping through EVs, it must move before robots reach factories (2027) and homes (2030).

‍

‍🎧 Listen to Radar by nexxworks here:

‍

Or listen on Apple Podcasts or Spotify

‍

‍