As Tesla's Full Self-Driving (FSD) system rolls out on a large scale in the Chinese market for the first time, a multifaceted game of autonomous driving technology, localization adaptability, and business strategies quietly unfolds. From North America to China, FSD embodies Tesla's ambition to revolutionize traditional driving experiences, yet it has exposed numerous discrepancies between technology and reality in a series of real-world tests.

The core competitiveness of Tesla's FSD lies in its pure vision technology path. By capturing environmental information through 8 cameras, combined with end-to-end neural network algorithms, FSD has achieved autonomous driving performance in the North American market with an average of only 0.3 human interventions per 1,000 miles. This technology path, heavily reliant on data training, thrives in the American environment with clear road rules and relatively standardized driving behavior.

However, when this system is transplanted to China's complex urban road conditions, its technological prowess is swiftly challenged by reality. A test video shot by a Tesla owner in Shanghai is highly illustrative: at an intersection with ground traffic lights, FSD misinterpreted the left-turn green light as a straight-through signal, causing the vehicle to run a red light; ten minutes later, the system failed to recognize tidal lane markings at another intersection, repeatedly crossing solid lines four times while changing lanes.

These errors expose the inherent limitations of the pure vision approach—when there are discrepancies between the traffic sign system and the training data, the system lacks dynamic adjustment capabilities. Common non-standard traffic lights on Chinese roads (such as horizontally arranged left-turn/straight lights), unexpectedly appearing electric bicycles, and the dynamic activation times of bus-only lanes are all "cognitive blind spots" for FSD.

More alarmingly, Musk publicly acknowledged that the Chinese version of FSD is trained solely using publicly available video data, meaning the system lacks a "three-dimensional space" understanding of the local road environment, making it difficult to respond to emergencies.

Tesla has set a price of RMB 64,000 for the FSD option, equivalent to the price of an entire Wuling Hongguang MINI EV. Behind this aggressive pricing strategy lies Tesla's confidence in the premium pricing power of software services.

However, in the Chinese market, this business logic faces triple challenges: Firstly, local automakers have adopted a competitive strategy of "standard hardware + free software." For example, Xiaopeng's XNGP system covers 243 cities nationwide without additional charges, and BYD has even extended advanced intelligent driving features to models priced at just RMB 100,000.

Secondly, consumer acceptance of paying for software is still limited to basic functions. The McKinsey report "McKinsey Insights on Chinese Auto Consumers 2024" shows that the proportion of respondents willing to pay separately for autonomous driving features has decreased by 8 percentage points compared to last year. Furthermore, consumers' willingness to pay for autonomous driving technology has generally declined across different payment models (one-time purchase of option packages, annual subscriptions, monthly subscriptions, and usage fees per 100 kilometers), except for a 25% increase in one-time purchase amounts for urban road scenarios.

In terms of payment method preferences, nationwide, 32% of consumers prefer to purchase autonomous driving option packages in one-time payments, 25% choose annual subscriptions, 9% choose monthly subscriptions, and 35% prefer to pay based on actual usage (such as mileage or number of times). However, these preferences vary across cities of different tiers. For instance, in first-tier cities, up to 49% of consumers prefer one-time purchases, while this proportion drops to 20% in second-tier cities and similarly 20% in third-tier and lower cities. In contrast, the proportion of payments based on actual usage is highest in third-tier and lower cities, reaching 51%, indicating that consumers in lower-tier cities place more emphasis on cost control and flexible payment options.

More crucially, when the actual performance of FSD cannot fully replace human driving, it is difficult for users to justify paying a high premium for a "half-baked product."

The deeper challenge faced by Tesla stems from the unique competitive landscape of China's intelligent driving market. While the American team iterates algorithms based on North American road data, local automakers are building moats through "data loops."

According to a November 19, 2022, report on Sohu.com, Xiaopeng has deployed over 300 triggers on nearly 100,000 Xiaopeng models, enabling scene data collection and upload anytime, anywhere. Huawei has also cooperated with various governments to establish dynamic update mechanisms for high-precision maps. According to an IT Home report from April 19, 2021, Huawei obtained Class A surveying and mapping qualifications for navigation electronic map production in July 2019. Its high-precision map system, Roadcode, is divided into two parts: Roadcode HD and Roadcode RT. Roadcode HD is equivalent to a high-precision map drawn by a professional surveying and mapping fleet and is offline, while Roadcode RT is equivalent to a self-learning map. Huawei plans to make high-precision maps commercially available for national highways and high-speed roads in Beijing, Shanghai, Guangzhou, and Shenzhen by 2021 and gradually expand coverage.

In this way, these companies not only grasp the micro-characteristics of local driving behaviors (such as electric vehicles suddenly changing lanes, pedestrians crossing green belts, etc.) but can also optimize algorithms several times a week through OTA updates. In contrast, due to restrictions on cross-border data transmission, the hundreds of millions of kilometers of driving data generated by Tesla vehicles in China are difficult to directly use for training, resulting in a lag in system evolution.

This gap is further amplified in technology paths. While Tesla adheres to a pure vision solution, mainstream local solutions have shifted to multi-sensor fusion: NIO's ET9 is equipped with Innovusion's Falcon W lidar, boasting flagship performance with a wide-angle lidar and a maximum detection distance of 150 meters, with nearly 10 times the angular resolution of similar products; Zeekr 007 even stacks the computing power of Orin-X chips to 508 TOPS. Although these hardware additions increase costs, they provide higher safety redundancy in China's complex urban environment—when cameras are interfered with by strong light, lidars can still accurately identify obstacles; when neural networks misjudge traffic lights, high-precision maps can provide lane positioning. The strategy of "combining hardware and software" is eroding Tesla's algorithmic advantages.

Despite facing numerous challenges, the entry of Tesla's FSD into China still holds far-reaching industrial significance.

Firstly, it forces local automakers to accelerate technological breakthroughs. A senior executive of an automaker stated that 2025 will be the year of intelligent driving for automakers. Facing Tesla's FSD, Li Xiang, founder of Ideal Auto, said, "Welcome to compare Ideal Auto with Tesla's FSD anywhere in the country. I am confident." The "catfish effect" of face-to-face competition will undoubtedly accelerate the pace of innovation throughout the industry.

Secondly, Tesla's end-to-end technology path offers another possibility for the industry. While competitors rely on high-precision maps, FSD demonstrates the potential of a pure data-driven solution. An article on Sohu.com on February 26, 2025, mentioned that the FSD system performs remarkably well at multiple complex intersections, smoothly executing basic operations such as lane changes, following vehicles, and parking, and also demonstrating self-decision-making capabilities based on environmental understanding.

Additionally, extensive real-world testing of Tesla's FSD by AMCI Testing also revealed that FSD can exhibit quite complex driving behaviors, such as navigating through a gap between two parked cars to let an oncoming vehicle pass or moving to the left to give space to pedestrians waiting for the green light at a crosswalk. Relevant test results indicate that Tesla's FSD's pure data-driven solution has the potential to adapt to complex road environments to a certain extent.

More noteworthy is the potential trend of technological integration. It is reported that Tesla has actively engaged with domestic automakers such as SAIC and Geely, with two rounds of in-depth communication already conducted with SAIC. If the collaboration is successful, Tesla can transplant its mature FSD technology into China's complex and ever-changing road environment, while leveraging SAIC's deep understanding of local road conditions and driving habits to accelerate the localization adaptation and optimization of the technology.

Musk's retweet of a test video by a Chinese blogger hints at Tesla's open attitude towards localization improvements—perhaps soon, we will see Tesla's FSD integrated with a certain domestic high-precision map or a lane-changing strategy specifically tuned for Chinese road conditions.

Standing on the cusp of the explosion of intelligent driving technology, Tesla's FSD journey in China is far from over. In the short term, its technical shortcomings and pricing strategy may indeed limit market share, and some extreme cases (such as owners being deducted 12 points due to FSD errors) may even trigger a crisis of trust. However, in the long run, Tesla's true competitiveness lies in the scale effect of its global data network—when road data from North America, Europe, and China are eventually pooled into the same training pool, the system's generalization ability will increase exponentially.

The competitive landscape in 2025 may hinge on three key variables: who can achieve data-driven localization iteration faster, who can find the best commercialization path for software services, and who can break through the technical bottleneck of L3 autonomous driving while ensuring safety.

For Tesla, FSD is not only a weapon to maintain its high-end brand premium but also a crucial stepping stone towards the "Apple model" of the automotive industry (hardware profits + software subscriptions). The outcome of this game may not be a zero-sum game of either-or but rather a redefinition of the value chain of smart cars through collision and fusion.

When a Tesla on the streets of Shanghai abruptly brakes due to misreading traffic lights again, it reminds us: the ultimate battle of autonomous driving is not about parameter comparisons in the lab but rather in every mixed human-vehicle intersection, every emergency construction section, and every subtle handover of control between humans and machines. This localization test in the Chinese market may be the ultimate touchstone to test whether FSD can truly become "Full Self-Driving."

More recommendations:

• Making American civil servants write weekly reports: How far can Musk's government efficiency revolution go?
• The rise of AI traffic police in Changsha vs. the ban on AI prescriptions in Hunan: Our love-hate relationship with AI
• The speculators lost RMB 10.8 billion: Mankind lost another utopia

About the Author