Original content from New Energy Perspective (ID: xinnengyuanqianzhan)

Full text: 2,770 characters | Reading time: 7 minutes

For a condensed version, watch the video. For detailed content, continue reading.

When unveiling the second-generation Roadster flying supercar, Elon Musk exuded confidence and pride in his demeanor.

Figure/Second-generation Roadster flying supercar

Source/Internet, screenshot from New Energy Perspective

A few days later, during Tesla's Q3 2026 earnings call, Musk seemed to encounter 'everything going wrong.'

First, Tesla's profitability deteriorated, with GAAP net income at $787 million, a 43% year-over-year decline, and operating profit at $718 million, down 56% from the previous year.

Figure/Tesla's official website

Source/Internet, screenshot from New Energy Perspective

Additionally, Musk's 2030 sales target of 20 million vehicles annually, set in 2020, was officially revised down to around 5 million, a 75% reduction.

The most immediate 'setback' was the repeated delays in FSD's (Full Self-Driving) launch timeline in China, with no clear timeline provided during this earnings call. Musk admitted that Chinese domestic automakers are fiercely competitive in autonomous driving and that Tesla has not yet established a true leading advantage.

While FSD struggles to enter China, Chinese automakers like Changan and BAIC are testing, iterating technologies, and accumulating experience under real-world commercial operations and regulatory frameworks.

Thus, the issue with FSD in China is no longer just about 'when it will be approved' but whether it can even attract consumers once approved.

1. FSD is Already Falling Behind in the Chinese Market

Tesla's FSD is a classic case of 'starting early but arriving late.'

As early as 2019, Tesla began offering FSD as a paid option to Chinese owners, with the buyout price rising from RMB 56,000 to RMB 64,000. However, at the time, only basic features like autoparking and lane-keeping were available, with even automatic lane changes unavailable—essentially equivalent to the basic autopilot functions that come standard with the vehicle.

In other words, seven years ago, Tesla was already selling 'the future' in China, but that 'future' has yet to be fully delivered. The current FSD (Supervised) is legally and technically defined as Level 2 assisted driving, not fully autonomous driving.

Meanwhile, China's true autonomous driving (Level 3 and above) has been deployed on specific road segments in Beijing, Chongqing, and other locations since 2026. Vehicles like the Changan Deepal SL03 and BAIC Arcfox Alpha S, bearing dedicated 'Z'-ending license plates, are mass-produced models with 'legal road approval.'

Figure/Models with autonomous driving (Level 3 and above) on the road

Source/Internet, screenshot from New Energy Perspective

FSD's 'early start but late arrival' is also evident in volume. According to D1EV, in the first half of 2026, China's urban NOA (Navigation on Autopilot) installations reached 1.802 million units, with Huawei leading at over 300,000 units and Tesla in the second tier with over 200,000 units.

Figure/Urban NOA installation rankings in H1 2026

Source/Internet, screenshot from New Energy Perspective

While Tesla's ranking seems decent, the structure changes upon closer inspection. Nearly all of Tesla's 200,000+ units come from its existing owner base, while Huawei's volume is driven by multiple brands under its HarmonyOS Intelligent Mobility ecosystem. If Tesla is 'eating from its own bowl,' Huawei is 'constantly adding dishes to the table,' with far greater momentum.

Data on willingness to pay is equally direct. Since FSD's buyout option became available in 2019, only about 25,000 to 30,000 domestic consumers have paid RMB 64,000 for it, corresponding to a selection rate of just ~1.5% among Tesla's ~2.2 million vehicles in China. In North America, over 75% of new Teslas activate paid FSD upon delivery—a stark contrast for the same software in two markets.

The experience gap is also felt by users. In third-party tests, the Chinese version of FSD requires driver intervention every 3.8 times per 100 kilometers, with only a 70% success rate in unprotected left turns and less than 50% recognition accuracy for high-frequency domestic scenarios like food delivery vehicles cutting in and non-motorized vehicles running red lights.

Many interventions are not due to 'insufficient capability' but 'non-compliance with rules,' such as crossing solid lines, running yellow lights, or ignoring lane markings. FSD's logic, trained on U.S. driving habits, struggles to adapt to China's road conditions.

2. What's Holding Back Full-Fledged FSD in China?

Clearly, for Tesla's FSD to catch up or surpass domestic brands in China, real-world testing and computational iteration are not optional but mandatory. This is not a denial of Tesla's algorithmic capabilities but a requirement dictated by the fundamental logic of intelligent driving and China's unique market ecosystem.

Musk's urgency is evident from FSD's 'launch timeline' in China. A brief recap: In November 2025, Musk said, 'Probably February or March 2026'; in January 2026, he changed it to 'as early as next month'; in April, he pointed to Q3; now that Q3 has passed, no new date has been given; and during the latest earnings call, he didn't even dare to mention a timeline.

Why has FSD repeatedly missed its launch window? The core issue is that its vision-only approach fails to align with regulatory requirements for compliance logic. Regulators demand that autonomous driving systems provide a traceable, auditable logic chain for every steering and braking decision in specific scenarios. However, FSD's end-to-end large model architecture lacks the modular breakdown of perception, planning, and control found in traditional systems.

In plain terms, regulators don't just want 'good performance' but 'accountability in case of incidents.' FSD cannot clearly explain itself in this regard.

Compared to China's willingness to give Tesla time for rectification, France has already slammed the brakes on FSD. In July 2026, France's transport minister officially rejected the current version of FSD on safety grounds, making it the first major European economy to do so.

Figure/France's transport minister officially rejects the current version of FSD on safety grounds

Source/Internet, screenshot from New Energy Perspective

France cited two main flaws: First, FSD's traffic flow following function automatically speeds up to 70 km/h in 50 km/h zones when following faster vehicles, exceeding the limit by 50%. Second, the system fails to continuously monitor driver attention at roundabouts and intersections.

In other words, FSD bypasses the rigid rule constraints of traditional systems, with the model prioritizing 'smoother traffic flow' over 'strict speed limits.'

China's mandatory national standards for L3/L4 autonomous driving, set to take effect in July 2027, follow the same logic. The standards do not explicitly ban vision-only systems—they don't care whether you use cameras or radar—but they require one thing: if one sensor fails, others must compensate. Vision-only systems rely solely on cameras, which struggle in heavy rain, fog, or backlighting—a physical-layer common-mode risk that cannot be overcome by algorithm iterations alone.

Figure/China's mandatory national standards for L3/L4 autonomous driving, effective July 2027

Source/Internet, screenshot from New Energy Perspective

Domestic manufacturers generally adopt an 'end-to-end + rule-based fallback' approach, where the model handles normal driving, and a separate set of rigid rules (e.g., 'stop at red lights,' 'do not exceed speed limits') takes over when the model is uncertain. While the model can remain a 'black box,' the fallback rules must be transparent for regulatory review.

3. Does China Need FSD, or Does FSD Need China?

To be fair, Tesla has made efforts, such as storing data locally in Shanghai's data center and launching its Lingang AI training center in February 2026. However, local data storage addresses 'whether training can occur here,' not 'whether the vision-only architecture can pass L3 redundancy requirements.'

After July 2027, new L3/L4 vehicle models must comply with national standards, and already-approved models must meet them by August 2028. If FSD cannot align with L3 pathways, it may remain locked as an 'advanced L2 assisted driving' system.

As domestic brands grow stronger, Tesla's influence has waned compared to two or three years ago. If it falls further behind in autonomous driving, its market share will continue to erode.

In 2026, urban NOA penetration reached 88.1% for domestic models priced above RMB 200,000 and exceeded 52.8% for models in the RMB 100,000 range. Navigation-assisted driving in urban areas is shifting from a 'high-end option' to a 'standard feature.'

Figure/Urban NOA penetration rate

Source/Internet, screenshot from New Energy Perspective

Price pressure is even more direct: XPENG offers XNGP for free, Huawei's ADS buyout price is RMB 36,000 (subsidized to just RMB 12,000), and Li Auto's AD Max ranges from RMB 30,000 to RMB 36,000. FSD remains priced at a one-time buyout of RMB 64,000 in mainland China—5.3 times Huawei's subsidized price and two to five times XPENG's and Li Auto's offerings. Moreover, no subscription model for FSD has been announced in China yet.

Figure/Pricing for Huawei's ADS high-end feature package

Source/Internet, screenshot from New Energy Perspective

As domestic rivals make high-end intelligent driving 'standard' or even 'free,' FSD's insistence on a premium price based on 'global technological leadership' is undermining its commercial logic.

Thus, to achieve Musk's new goal of 'producing over 5 million vehicles annually by 2030,' China is undoubtedly a market FSD must conquer. After all, the world's most complex road scenarios, largest smart EV fleet, and fiercest intelligent driving competition are all here.

We've also seen Tesla's first safety report for its autonomous ride-hailing service, revealing that as of October 18, 2026, its fleet had accumulated 4.99 million kilometers with zero injuries, zero airbag deployments, and zero collisions with pedestrians or cyclists.

However, this occurred in limited operational areas in Texas, Florida, and California under a different regulatory framework and data closed loop (closed-loop system). It proves Tesla can achieve safety redundancy in autonomous driving under specific U.S. conditions but does not automatically translate to L3/L4 approval in China or consumer trust in a RMB 64,000 option package.

For FSD, China's market door remains open, but Tesla and Musk must act swiftly with substantive moves. The window for FSD to enter China calmly and establish a lead is indeed narrowing.

[Header image generated by AI]