Part 1: Robotaxi – Most High-Profile, Most Difficult to Profit
1. Current Status: Operations in Multiple Cities, But Not Yet Profitable
As of 2026, more than ten cities in China including Beijing, Shanghai, Guangzhou, Shenzhen, and Wuhan have permitted commercial paid Robotaxi operations. Leading players such as Baidu Apollo Go, Pony.ai, WeRide, and AutoX have each accumulated over 10 million kilometers of testing. However, except for certain limited zones, the vast majority of Robotaxi projects have yet to break even.
2. Cost Structure Breakdown
Taking the operating cost of one Robotaxi as an example: vehicle hardware modification (sensors, compute units) approximately RMB 150,000-300,000; safety operator costs (still mandatory in some regions) approximately RMB 100,000/year; remote monitoring and maintenance team allocation approximately RMB 30,000-50,000/year per vehicle; insurance, charging, parking, etc., approximately RMB 50,000/year. Total annual cost approximately RMB 350,000-500,000. Based on RMB 2.5 per kilometer, 15 paid rides per day at 8 kilometers each, annual revenue is approximately RMB 220,000 – leaving a significant gap.
3. Key to Breakthrough: Removing the Safety Operator
Industry consensus is that Robotaxi profitability requires the removal of safety operators, reducing annual per-vehicle costs to RMB 150,000-200,000. Waymo and Baidu have already piloted fully driverless (no safety operator) operations in certain limited zones. However, long-tail challenges including extreme weather, complex intersections, and traffic police gesture recognition still need breakthroughs.
4. Competitive Landscape: Tech Companies and Automakers Progress on Dual Fronts
· Baidu Apollo Go: Operates in the most cities, plans to achieve regional profitability by 2027.
· Pony.ai: Partnering with Toyota and GAC, focusing on breakthroughs in rainy and nighttime scenarios.
· Tesla Robotaxi: Based on pure vision approach, plans to launch dedicated Cybercab model in 2026 with targeted cost below $50,000.
---
Part 2: Unmanned Delivery – Low-Speed Scenarios Achieve Commercialization First
1. Advantage Analysis: Simpler Scenarios, More Lenient Regulations
Unmanned delivery vehicles typically operate on closed or semi-closed roads such as campuses, industrial parks, and residential communities, with speeds below 25km/h, presenting significantly lower risk levels than highway autonomous driving. Consequently, regulatory barriers are lower, and multiple Chinese localities already permit operations without safety operators.
2. Business Model: Clear Cost Reduction Logic
Taking JD.com's unmanned delivery vehicle as an example: hardware cost approximately RMB 80,000-100,000 per vehicle, no safety operator, electricity and maintenance approximately RMB 20,000 per year, average 80-120 deliveries per day, per-delivery cost approximately RMB 1-1.5 – a clear advantage over traditional human delivery (RMB 2-3 per delivery). Meituan, Neolix, BAI (White Rhino), and others have completed hundreds of thousands of real deliveries.
3. Main Scenarios and Players
· Last-mile parcel delivery: JD.com, Cainiao, SF Express
· Instant food delivery: Meituan, Ele.me (Alibaba)
· Retail and grocery: Yonghui, Hema (partnering with third parties)
· Cleaning and security: Kuwa Robot, Gaussian Robot
4. Scale Forecast
According to McKinsey, by 2028, China's unmanned delivery vehicle fleet will exceed 500,000 units, with daily deliveries exceeding 50 million and market size reaching RMB 30-40 billion.
---
Part 3: Long-Haul Trucking – High-Value, High-Difficulty Track
1. Industry Pain Points: Driver Shortage and Cost Pressure
Long-haul freight faces severe driver shortages (average age over 45, low willingness of younger generations to enter the profession). Labor costs account for 25-30% of total long-haul logistics costs. If autonomous driving can replace one driver, annual savings per vehicle would be approximately RMB 150,000-200,000.
2. Technology Routes: Platooning vs. Single-Vehicle L4
· Platooning: A lead vehicle with a human driver, followed by autonomous following vehicles. Can save 10-15% in fuel consumption while reducing reliance on high-performance perception. TuSimple and Inceptio Technology have launched commercial pilot operations.
· Single-Vehicle L4: Fully unmanned heavy trucks requiring extremely high reliability and extended perception distance (over 500 meters). Still in testing phases, with complex highway conditions (road construction, toll stations, non-standard lanes) as major obstacles.
3. Business Model: Pay-Per-Kilometer Freight
The mainstream model is Autonomous-driving-as-a-Service: logistics companies pay technology providers on a per-kilometer basis (e.g., RMB 0.5-1/km), without purchasing hardware. Inceptio Technology, in partnership with Deppon, ZTO, and others, has achieved transition from two drivers to one (one safety operator plus autonomous system) on routes such as Beijing-Shanghai and Chengdu-Chongqing, with cumulative mileage exceeding 30 million kilometers.
4. Regulatory Breakthroughs: Unmanned Testing Permitted on Select Sections
Zhejiang, Jiangsu, Sichuan and other provinces have opened certain highway sections for fully unmanned (no safety operator) L4 heavy truck testing. However, issues such as cross-province operations and adverse weather response remain to be resolved.
---
Industry Perspective
"Robotaxi proves technology, unmanned delivery proves commerce, long-haul trucking proves scale." - Chen Tao Capital's 2026 autonomous driving investment white paper notes that while the three tracks have different commercialization rhythms, all will ultimately move toward driverless operations. Over the next 3-5 years, the track most likely to achieve profitable models first is unmanned delivery, followed by long-haul trucking, with Robotaxi taking longer.
---
Knowledge Card · One Minute to Understand the Difference Between L4 and L3
L3 (Conditional Automated Driving) : The system performs all dynamic driving tasks under specific conditions, but the human driver must respond promptly when requested to take over. L4 (High Automation) : The system performs all driving tasks within a limited operational design domain, with no human takeover required. The key difference: L3 requires hands-on-wheel, eyes-on-road readiness; L4 does not. Consequently, Robotaxi, unmanned delivery, and long-haul trucking all fall under L4 and above.