In Depth: China’s Auto Industry Bets Self-Driving Cars Will Lead to Embodied AI
Listen to the full version
Self-driving cars may become the first large-scale application of embodied artificial intelligence (AI), as China’s auto industry accelerates toward higher levels of autonomous driving with a wave of new product launches and technical breakthroughs.
At the Beijing auto show on Friday, Geely Automobile Holdings Ltd., one of China’s largest automakers, unveiled its Eva Cab, a self-driving taxi powered by Nvidia Corp. and Qualcomm Inc. chips. The vehicle is designed to handle high-level autonomous driving tasks, boasting processing power of more than 3,000 TOPS (trillions of operations per second).
Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.
Save an extra $50. Introductory offer for new readers. Subscribe now.
- RELATED
- DIGEST HUB
Digest Hub
Back
- China's automakers advance autonomous driving: Geely unveils Eva Cab (>3,000 TOPS); Huawei trials Level 3 (commercial 2027), Level 4 by 2028.
- Cars positioned as embodied AI proving ground, extending to robots; Tesla ends Model S/X production, Li Auto launches L9 as first such product.
- Challenges: 100-150 EFLOPS compute for L4, high costs, competition, regulations after 2025 Xiaomi incident.
1. China's auto industry is accelerating autonomous driving as the first major application of embodied AI, with new launches and breakthroughs.[para. 1]
2. Geely unveiled the Eva Cab self-driving taxi at the Beijing auto show, powered by Nvidia and Qualcomm chips, with over 3,000 TOPS processing power for high-level autonomy.[para. 2]
3. Huawei introduced an upgraded assisted driving system, starting Level 3 trials this year, expecting large-scale use by 2027 and Level 4 urban vehicles by 2028.[para. 3]
4. China classifies autonomous driving into six levels (0-5), similar to SAE standards: Level 3 is conditional automation requiring human intervention in emergencies, Level 4 is nearly fully autonomous.[para. 4][para. 5]
5. These developments position cars as the initial testbed for physical AI, extendable to humanoid robots alongside large language models.[para. 6]
6. Horizon Robotics' Su Qing predicts vehicles will surpass human drivers as computing power grows.[para. 7]
7. Tesla plans to halt Model S/X production for humanoid robots, with Musk claiming leadership in AGI via humanoids.[para. 8]
8. Li Auto brands itself an embodied AI firm, launching the L9 as its first such product and expanding to robots.[para. 9]
9. Carmakers lead physical AI due to autonomous driving's structured environment as a stepping stone.[para. 10][para. 11]
10. QCraft CTO Li Dong highlights autonomous driving data's role in training world models for physical decision-making.[para. 12]
11. ZYT CEO Shen Shaojie sees physical AI as inevitable, with training costs in billions of yuan yearly; survival requires cross-scenario data loops.[para. 13]
12. Challenges include scarce robot data, variances across types, and video data optimization.[para. 14]
13. Immense computing power is needed; automakers build data centers and partner with clouds.[para. 15]
14. L4 requires 100-150 EFLOPS training compute; Tesla's 2026 capex is $25B.[para. 16][para. 17]
15. Chinese firms struggle with competition and profitability: Li Auto returned to 2025 losses, XPeng unprofitable, Horizon Robotics swung to loss.[para. 18]
16. Regulators tighten rules; China's MIIT issued UN-aligned safety draft; 2025 Xiaomi incident led to stricter approvals.[para. 19][para. 20]
(Word count: 498)
- MOST POPULAR
Sign in with Google
Sign in with Facebook
Sign in with 财新