1. China's L3 National Standard: High-Level Autonomous Driving Gets "Traffic Rules"
On May 20, MIIT officially released national standard GB/T 44468-2025. This is China's first legally binding mandatory national standard for L3 and higher automated driving systems, hailed by the industry as the "birth certificate" for high-level autonomous driving on Chinese roads.
The standard's core provisions can be summarized in four aspects:
First, liability allocation. When an L3 system is activated, the vehicle assumes dynamic driving responsibility. The driver may divert their gaze from the road, remove their hands from the steering wheel, and engage in activities unrelated to driving, but must remain "capable of taking over" - resuming control within 5-10 seconds after the system issues a takeover request. This means accident liability during system activation falls on the vehicle manufacturer or system supplier, not the driver. This is one of the clearest definitions of autonomous driving liability worldwide.
Second, safety redundancy. The standard requires dual or multiple redundancy in critical subsystems - braking, steering, perception, and power supply - to ensure single-point failures do not lead to hazardous situations. While redundancy increases hardware costs (an estimated 5,000-10,000 RMB per vehicle), it is a necessary price for safety.
Third, data recording. The standard mandates a DSSAD "black box" that continuously records critical data before, during, and after collisions, including system status, driver presence, vehicle speed, and environmental perception data. Data must be retained for at least 90 days in tamper-proof form and will serve as the core basis for accident liability determination.
Fourth, human-machine interface. Clear visual, auditory, and even tactile prompts are required for system activation and takeover requests. In-cabin displays that may excessively distract the driver are prohibited.
The standard takes effect July 1, 2026. After this date, all vehicles sold in China with L3 or higher functionality must pass compliance certification. Systems such as Huawei ADS, Xpeng XNGP, Li AD Max, and Nio NAD can theoretically be upgraded to meet L3 requirements. It is foreseeable that starting in the second half of 2026, China will see its first truly mass-produced L3 vehicles.
2. Huawei ADS 3.0: End-to-End Architecture Fully Deployed
One week before the standard was released, Huawei launched ADS 3.0 at its annual spring event. This is the latest achievement from Huawei's full-stack, self-developed capabilities following US sanctions.
The biggest highlight of ADS 3.0 is "parking-to-parking" driverless navigation without HD maps. Through the fusion of real-time perception and navigation maps, it achieves full-process autonomous driving from a garage parking space to a destination parking space - exiting the garage, navigating city roads and highways, entering the destination parking lot, and completing parking. Previously, this capability was only partially实现的 by Tesla FSD in North America.
At the architectural level, ADS 3.0 adopts an end-to-end neural network, merging perception, prediction, planning, and control into a single large neural network. Compared to modular architectures, end-to-end better handles long-tail scenarios such as debris on the road, irregularly behaving vehicles, and complex roundabouts.
In terms of computing power, ADS 3.0 is powered by Huawei's MDC 810 platform, delivering 400 TOPS on a 7nm process, supporting ASIL-D functional safety. The sensor configuration includes one 192-line semi-solid-state LiDAR, three millimeter-wave radars, 11 high-definition cameras, and 12 ultrasonic radars.
Launch vehicles include the AITO M9, Avatr 12, and the Stelato S9 - a new luxury sedan developed by BAIC in partnership with Huawei. After the launch, Jin Yuzhi, CEO of Huawei's Intelligent Automotive Solutions BU, revealed that the ADS 3.0 uptake rate on the AITO M9 has exceeded 70%. Despite the not-insignificant 36,000 RMB option price, consumer recognition of high-level intelligent driving is increasing rapidly.
3. Tesla FSD V13: North America Rollout, China Entry Imminent
Tesla began pushing FSD V13 to North American users in early May. This represents another major upgrade following V12.
V13 completely removes approximately 300,000 lines of C++ control code inherited from V11 and earlier, achieving a truly pure end-to-end neural network. Average city road intervention distance has increased from approximately 200 miles (V12) to approximately 400 miles, with notable improvements in complex scenarios such as roundabouts, unprotected left turns, and construction zones.
Regarding FSD's China entry, there are new developments. According to sources, the China Association of Automobile Manufacturers is leading the development of testing standards and准入 rules for foreign autonomous driving systems, with Tesla as a participating company. This suggests that FSD's China entry is no longer a question of "whether" but "when and how" to achieve compliance.
Current obstacles include data security (can perception data be transmitted overseas? Is Tesla willing to build a supercomputing center in China?), mapping qualifications (partnership with qualified Chinese firms required), and regulatory approval processes. Industry expects that Tesla FSD V13 could enter small-scale public testing in China by late 2025 or early 2026.
4. Software Subscriptions: Automakers' Second Growth Curve
First-quarter 2025 earnings reports show strong growth in feature subscription revenue across multiple automakers. BMW's Connected Drive covers over 6 million active users with 38% YoY growth; Mercedes-Benz saw 35% growth; Nio's NIO Pilot full package achieved a 45% subscription rate.
The most popular subscriptions include heated seats (€15/month), heated steering wheel (€10/month), advanced driver assistance, remote unlocking, and acceleration enhancement (reducing 0-100 km/h time by 0.5-1 seconds via OTA, €50/month).
Why is the subscription model gaining traction? From automakers' perspective, an EV with hardware gross margins of only 15-20% can generate additional revenue equivalent to 10-15% of the vehicle price over a 5-8 year usage cycle through software subscriptions, while the distribution cost of software is near zero. From consumers' perspective, subscriptions lower the purchase barrier - users need not pay high option prices for functions they use only a few times per year.
McKinsey's May report, Automotive Software Monetization, forecasts that by 2030, software and services will contribute 15% to 20% of automakers' gross profit, becoming the second-largest revenue source after hardware sales. New business models - hardware on demand, in-vehicle app stores, data monetization - are accelerating.
---
Summary: In the intelligent driving sector in May 2025, institutional development and commercial deployment are accelerating in tandem. The L3 national standard clears institutional obstacles for high-level autonomous driving; Huawei and Tesla have achieved breakthroughs along their technology pathways; software subscriptions are becoming a new profit growth engine for automakers. When the value of a vehicle comes increasingly from code rather than steel, the era of "software-defined vehicles" will have truly arrived.