Evolution of China's Autonomous Driving Technology and Intensified Industry Competition
AI summary card
Evolution of China's Autonomous Driving Technology and Intensified Industry Competition
Bernstein, through test drives and industry research, points out that China's L2++ autonomous driving capabilities have improved significantly, with intense competition among players in software-hardware integration; L4 Robotaxi bottlenecks shift to cost and operations, with regulation being the biggest uncertainty.
- Full Enhancement of L2++ Capabilities: XPeng VLA 2.0 Performance Human-like, Momenta and Xiaomi Significantly Improved
- Fierce Urban NOA Competition: Momenta Leads Market Share, Horizon HSD Solution Adopted by Multiple Auto Makers
- Auto Makers Accelerate In-house Chip R&D: XPeng Has Delivered Over 200k Turing Chips, NIO and Li Auto Follow Closely
- L4 Robotaxi Focuses on Cost Reduction: Pony.ai Targets Vehicle Cost Below 230k CNY by 2027, WeRide Aims to Reduce BOM to 150k CNY Within Five Years
- Sector Ratings Show Divergence: Bullish on BYD, Xiaomi, Horizon; Bearish on Black Sesame
Report interpretation
Overview
Bernstein held the third 'China Autonomous Driving Tour' in Beijing, testing XPeng, Xiaomi, and Momenta L2++ vehicles as well as Pony.ai and WeRide L4 Robotaxis, and visiting supply chain companies like Horizon and Black Sesame, summarizing the latest developments in China's autonomous driving industry. Overall, L2++ autonomous driving technology capabilities are significantly improving and accelerating in scale application, with software-hardware integration becoming a core competitiveness for enterprises. The bottleneck for L4 Robotaxis has shifted from technical breakthroughs to cost reduction and operational optimization, while regulation becomes the biggest uncertain factor. The report shows structural divergence in ratings for different companies.
Core views
Technology Capability and Competitive Landscape: L2++ autonomous driving capabilities are making continuous substantial progress. XPeng's VLA 2.0 system driving performance is confident and close to human habits; Momenta shows significant improvement over the previous generation; Xiaomi's driving confidence is also increasing. Broader adoption of world models for simulation and scenario generation is accelerating capability expansion and reducing development costs, supporting evolution towards higher levels of autonomous driving. Some L2++ players (including Horizon) plan to leap to L4 levels; XPeng plans to conduct L4 Robotaxi pilots in the second half of 2026. Software and Chip Competition Heats Up: In terms of city NOA software, Momenta still leads in market adoption, but Horizon is adopting HSD full-stack solutions with more auto makers such as Chery, Volkswagen, and Changan. Qianli Technology also delivered intelligent driving algorithms to Geely/Zeekr, while WeRide collaborated with GAC to sell L4 downgraded algorithms as L2++. In chip terms, Horizon is the only local enterprise capable of challenging Nvidia in the 500+ TOPS chip field in 2026, receiving design awards from Chery, Volkswagen, BYD, etc. Black Sesame's L2++ chips have also begun to receive auto maker awards. In-house Chip Development by Auto Makers Creates Disruption: XPeng has shipped over 200k self-developed Turing chips, targeting 1 million in 2026 (including supplies to VW); NIO Shenji chips are equipped on ET9 and other models with external licensing opened; Ideal M100 chips launched with L9. This raises competition barriers for third-party chip suppliers, but Horizon maintains differentiation advantages through product innovation and cost reduction via the newly launched Cockpit-Integrated SoC Starling chip. L4 Robotaxi Shifts to Cost and Operational Optimization: Constraints on L4 Robotaxis are no longer technical breakthroughs, but depend on cost reduction, operational scale, utilization density, and regulatory trust. Pony.ai targets vehicle costs below 230k RMB before 2027; WeRide estimates BOM costs can drop to about 150k RMB within five years. China has basically verified technical feasibility, but overseas markets with higher ARPU and faster break-even points are expected to become profit drivers.
Analysis framework
This report adopts a method of primary research and supply chain cross-validation. The institution obtained sensory cognition and technical progress evidence for each enterprise system in natural driving and confidence through real-life test drive experiences (covering mainstream L2++ passenger cars and L4 Robotaxis). In the industry verification layer, the report splits autonomous driving into four dimensions: software algorithms, chip hardware, in-house car development, and Robotaxi operations, analyzing the competitive landscape and core barriers in each link one by one, such as horizontally comparing chip players and car maker in-house development processes, refining the core logic of 'software-hardware integration as a moat'. For L4 Robotaxi analysis, the report starts from the economic model, shifting focus from 'technical breakthroughs' to operational indicators such as 'BOM cost, fleet scale, utilization density', deriving the conclusion that autonomous taxi economics converges towards manufacturing and operational optimization.
Methodology notes
Software-Hardware Integration as a Moat
Refers to enterprises forming insurmountable barriers for competitors by deeply binding self-developed chips with software algorithms. The report uses this to analyze how car makers (like XPeng, NIO) self-developing chips creates disruption to third-party suppliers, and how Horizon maintains differential advantages through cockpit-integrated chips.
BOM Cost and Economies of Scale
BOM (Bill of Materials) cost refers to the total parts cost required for product manufacturing. The report applies this method to analyze the profitability path of L4 Robotaxis, pointing out that through expanding fleet scale, parts price drops (especially LiDAR), and domestic substitution, single-vehicle BOM costs are expected to decrease significantly, thereby promoting Robotaxi commercial closure.
EV/Sales (Price-to-Sales) Valuation Method
Ratio of Enterprise Value to Sales Revenue, often used for growth enterprises that have not yet achieved stable profitability. The report mentions in the valuation table that EV/Sales valuation was used instead of traditional P/E for some names (such as relevant semiconductor and tech companies).
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Horizon Robotics (9660.HK)Benefiting from growth in urban NOA software and high-compute chip demand
- Strengths
- The only local enterprise in 2026 500+ TOPS chip field capable of challenging Nvidia; Launched Starling Cockpit-Integrated SoC, reducing costs through product innovation to form differentiated competitive advantage
- Weaknesses
- Facing competitive pressure from car maker in-house chips
- Comparison
- In the chip track, its differentiated innovation capability is superior to third-party suppliers like Black Sesame; in the software track, its HSD solution is narrowing the market share gap with Momenta
- XPeng Inc. (9868.HK)Leading autonomous driving technology and advancing in-house chips and Robotaxi
- Strengths
- VLA 2.0 system driving performance is human-like; Has shipped over 200k Turing chips, targeting 1 million in 2026; Plans L4 Robotaxi pilot in second half of 2026
- Comparison
- Leads many traditional auto makers in software-hardware integration layout
- Black Sesame Technologies (2533.HK)Faces commercialization pressure in the highly competitive chip market
- Strengths
- L2++ chips have begun to receive OEM design awards, achieving initial commercial progress
- Weaknesses
- Rating downgraded to Underperform, facing fierce competition from Horizon and car maker in-house chips
- Comparison
- Compared to Horizon's status in 500+ TOPS field and cockpit-integrated innovation, Black Sesame's competitiveness is relatively weaker
- BYD (1211.HK)Beneficiary target in intelligent driving and chip adoption
- Xiaomi (1810.HK)Beneficiary target with significantly improved autonomous driving capabilities
- Strengths
- Autonomous driving confidence is improving, rated Outperform
Key data
- Pony.ai Target Vehicle Cost< 230k RMBTarget realized by 2027, driven by fleet scale, parts price drops (especially LiDAR), domestic substitution, and manufacturing optimization
- WeRide BOM Cost PathApproximately 150k RMBExpected to achieve Robotaxi BOM cost at this level within five years
- XPeng Turing Chip ShipmentsOver 200k unitsShipped quantity, targeting 1 million units in 2026 (including supplies to VW)
- Horizon Chip Compute Positioning500+ TOPSIn 2026, Horizon is the only local enterprise in the 500+ TOPS chip field capable of challenging Nvidia
- BYD Target Price124.00 CNY / 136.00 HKDRated as Outperform
- Xiaomi Target Price43.00 HKDRated as Outperform
- Horizon Robotics Target Price15.00 HKDRated as Outperform
- Black Sesame Technologies Target Price16.00 HKDRated as Underperform
Impact & implications
Competition in the autonomous driving software and chip fields is intensifying, bringing better technical experiences to consumers but also pressure on all players. Third-party chip suppliers face severe challenges from car maker in-house chips, and only maintain competitiveness through product innovation (such as cockpit-integration) rather than pure price wars. For the Robotaxi industry, technology is no longer the biggest threshold; enterprises capable of achieving extreme cost reduction and high-density operations will win in the commercialization process. Overseas markets, due to higher ARPU and faster break-even points, are expected to become profit sources for Robotaxi enterprises.
Risks
- Regulation is the biggest uncertain factor for L4 autonomous driving; government approval prioritizes safety records, operational reliability, and public trust
- Industry-wide safety accidents or setbacks may quickly trigger regulatory backlash (e.g., Baidu Apollo Go event caused license issuance suspension)
- Intensified competition in autonomous driving software and chip fields creates pressure on all players
What to watch
- Progress of XPeng Inc. L4 Robotaxi pilot in second half of 2026
- Path of decreasing vehicle costs for Robotaxis and realization of BOM cost goals (e.g., Pony.ai 2027 target, WeRide 5-year goal)
- Shipments progress of car maker in-house chips (e.g., XPeng 2026 1 million unit target) and impact on third-party suppliers
- Changes in regulatory policies and impact of safety events on autonomous driving license issuance