Tesla FSD is a bit “not smart”

Wen| Finance and economics are unscrupulous and rustless

In the opening year of 2025, the autonomous driving industry can be said to be constantly exploding.

After Wang Chuanfu ignited the popularity of national smart driving at the press conference two weeks ago, Musk, who followed closely, added fuel to the booming industry.

On February 25, Tesla China released a piece of information saying that it would push the 2024.45.32.12 software update for some users to upgrade Autopilot automatic assisted driving on urban roads, marking a substantial step in the localization of its FSD (Full Self-Driving) function.

As early as the end of last year, Tesla launched the new FSD V13, which attracted a number of domestic smart driving celebrities to check in across the ocean and received good reviews.

In contrast, what kind of scores can be retained by domestically implemented castrated products has become the focus of market attention.

After a day of high-intensity full-network testing, the entrance exam for the domestic version of FSD has basically passed. Can this product, which carries Tesla’s ambition to lock in the China smart driving market and new revenue expectations, let it do what it wants?

1. The castrated version of FSD only got a passing score”

The day that just passed, Tesla may have hit a new record for the number of cameras in the car.

From morning and evening peak tests to a leisurely stroll at noon, from the bustling downtown to the open and wild, the leading car and technology media completed an uninterrupted live broadcast of Tesla’s autonomous driving almost in a relay manner.

Among them, the number of kilometers on the dial of fellow cars has increased along with the endless stream of paragraphs in the live broadcast room.

For example, a leading car blogger likes to file seven fines, such as a community security guard who is at the gate, making the vehicle difficult to move. The most miserable team of car-savvy drivers has committed 34 violations in one go, so that there are not enough words……

Putting aside the details, from a macro perspective, it should be difficult for FSD to remove the title of driver’s license killer in a short period of time.

In terms of specific aspects, the performance of FSD is mixed.

At the end of last year, around the news that FSD might be implemented in China, the main curiosity points of domestic public opinion focused on the following issues:

·Compared with the US version of FSD, how successful can the domestic version of FSD retain?

·As the first official announcement of the P2P capability (Park to Park), it refers to the L5’s intelligent driving system. What is the level of driving anthropomorphism of FSD?

·FSD, which is accustomed to American road conditions, will it not adapt to the conditions in the face of China’s special traffic environment?

Regarding the first question, referring to the relevant information compiled by the media and the comprehensive performance of yesterday’s live broadcast, Caijing Wuji believes that the domestic version of FSD can basically score a 60-point pass score in terms of power restoration, which is in line with expectations, but the expectations are not high.

After all, subject to relevant restrictions of laws and regulations, no one really expects FSD to replicate L4+ level autonomous driving in China. The industry expected it to reach the level of driving assistance system (L2+) at the beginning.

Overall, compared with the original version, the castrated version of FSD does have an increase in takeover rate, and it does not have the ability to drive with one click to the end. The overall adaptation scenario is biased towards structured public roads, which is in line with the level of driving assistance systems.

The only pity is that one of the major advantages FSD promotes in the U.S. market: automatic parking space search does not seem to have been retained in the domestic version.

The second question, aside from road adaptability, Caijing Wuyi believes that the driving level of FSD can be awarded to old drivers with 80 points.

In the past, when measuring the level of autonomous driving in the industry, a set of adjectives was often used: conservative/radical.

But no matter whether the style of the smart driving system is conservative or radical, it is still essentially a driving tendency preset by the engineering team.

In contrast, a smarter and more anthropomorphic intelligent driving system requires neither conservatism nor radicalism, but flexibility between radicalism and conservatism.

In this regard, the performance of the castrated version of FSD is indeed amazing. During the morning and evening peak tests, many testers gave feedback on the characteristics of the driving system of this system:

Confidence.

This confidence is reflected in several aspects. The first is the smooth driving process, which can independently complete lane change, detour, follow, avoid and other operations. Especially in tunnels, elevated and other scenarios, driving without a takeover can be basically achieved. In the morning peak test conducted by some media, this system also demonstrated its strong continuous traffic capacity in the face of heavy traffic.

Secondly, it is the completion of difficult actions, such as large-curvature U-turns under the viaduct, narrow road crossing, emergency avoidance and other operations. FSD can smoothly handle these scenes where even old drivers have to respond mentally. Basically, it can be ranked in the first echelon in the country.

Finally, there is the level of game in complex scenarios. In fact, this was also a part that mainstream media was generally worried about before it was launched in China.

Compared with the easy model of the American driving environment, domestic complex driving scenarios place higher requirements on the game capabilities of the autonomous driving system. In this regard, FSD’s performance is also remarkable. It can not only actively detour to avoid jammed vehicles, but also proactively maintain a safe distance.

Compared with the good performance of driving skills, the last issue that is generally worried about in the industry: Although there are certain expectations in terms of localization capabilities, the actual performance of the castrated version of FSD is still surprising, and you can almost score 0 points directly.

Objectively, there are very large differences in the driving environment between China and the United States, including but not limited to:

Scene level: There are special areas such as traffic roundabouts, turning waiting areas, tidal lanes, and bus lanes;

Traffic rules level: There are rules prohibiting lane changes such as solid lines and yellow lines;

Road conditions: The number of non-motor vehicles is relatively high, there are a large number of sections where people and vehicles are mixed, and there are special scenarios such as temporary traffic lights and traffic control.

In this regard, due to regulatory policy and geopolitical issues, Tesla actually had difficulty obtaining complete domestic road data. At the worst time, according to Musk, Tesla could only go online to find videos for FSD to learn.

The final exam result was naturally a terrible one.

On the above three levels, this FSD system basically failed all of its tasks, compaction lines, ignorance of bus lanes, unlimited force to stop by electric vehicles, and random running red lights. This is also the origin of the title of driver’s license killer in the previous paragraph.

Overall, if you have to give a comment on the performance of FSD in the preliminary test, a more appropriate description may be:

“An old American driver who has never driven in China for a day.& rdquo;

2. Tesla can’t wait for the iPhone moment”

On this basis, the attitude of domestic car companies also reflects the implementation performance of this smart driving system to a certain extent.

So far, CEOs of many car companies, including Huawei, Xiaopeng, and Ideal, have expressed their views on this news, and everyone’s attitudes are basically the same:

“Welcome to visit and welcome competition.& rdquo;

Yu Chengdong, managing director of Huawei, chairman of Terminal BG, and chairman of Smart Car Solutions BU, was one of the first car circle tycoons to express their position in public channels. As early as June last year, he expressed his position on the news of FSD’s entry into China:

“Be willing to compete, and only after competition can you see who is doing better.& rdquo;

After Tesla’s official announcement, Xiaopeng CEO He Xiaopeng also said through personal Weibo: I heard that Tesla’s FSD has begun to be rolled out in China one after another. I warmly welcome it, and there will be more interesting exchanges and mutual promotion in the entire smart driving industry. rdquo;

Coincidentally, Li Xiang, CEO of Ideal Car, also issued a document saying that everyone is welcome to compare Ideal Car with Tesla FSD anywhere in the country.

Looking at China’s car circles where competition is fierce and friendly and business relationships are not harmonious, behind the rare enthusiasm of the bosses, there is undoubtedly two profound meanings:

On the product side, the castrated version of FSD does not put too much competitive pressure on domestic car companies, but on the traffic side, it naturally provides a stage for China car companies to demonstrate their smart driving skills.

Regarding the first point, unlike the world-shaking Model S when it came out, and unlike the cold price butcher of the Model 3, Tesla’s FSD system is currently unable to welcome the iPhone moment in terms of pricing, experience and technology.

Refer to Zhang Ning, vice president of Xiaoma Zhixing and head of Robotaxi business:

“Tesla FSD enters the market at this time point and cannot be a crushing existence. rdquo;

On the second point, compared with the technical pressure given by the complete FSD, the castrated version of FSD is undoubtedly the most suitable imaginary enemy for domestic car companies.

Previously, there has always been a misunderstanding at the cognitive level in China regarding autonomous driving technology: the larger the number, the more advanced it is.

But in fact, the so-called L2+ and L4 and L5 autonomous driving is not a simple advanced relationship, but completely belong to different technical routes. L2 pursues driver-assisted driving, while L4 and above systems pursue driverless driving.

Thanks to the application of various advanced technologies, domestic autonomous driving manufacturers have basically no disadvantages in the L2+ field. Instead, they are limited by policies and regulations, and there are some gaps in the field of autonomous driving at L4 and above.

On this basis, Tesla was forced to castrate FSD this time and violently change the L5 system to L2+ for implementation in China. This is undoubtedly giving up the strengths and taking advantage of the weaknesses, giving China car companies an opportunity to comprehensively benchmark.

This is also why Li Xiang had previously made a high-profile official announcement to launch a new ideal AD training model:

“Ideal AD Max V13 versus Tesla FSD V13, I am confident!& rdquo;

In addition, the performance upgrade of Tesla’s FSD also requires climbing time.

Different from overseas computing centers where H100 and H200 are freely circulated, Tesla wants to achieve further training on FSD in China and can only switch chips with relatively weak performance, which in turn provides China car companies with a longer development time window.

Unfortunately, in the face of increasingly competitive intelligent tracks, Tesla’s FSD does not seem to have enough time.

After the novelty in front of us, can the expensive smart driving system, which sells for as much as 64,000 yuan, really gain the recognition of users?

The answer is obviously questionable.

On the one hand, domestic users have not yet widely accepted the mentality of paying for smart driving. On the pricing side, other China car companies have also generally adopted lower pricing for their smart driving systems.

Compared with the buyout price, Tesla’s FSD price of 64,000 yuan is 1.7 times that of Xiaopeng XPILOT 3.0 (The price is 36,000 yuan, and the car purchase option is 20,000 yuan), which is 1.6 times that of the NIO Pilot full package (the price is 39,000 yuan). Some models of Ideal, Xiaomi and Ai ‘an do not even have an additional ADAS (Assisted Driving System) fee. Previously, BYD even made a high-profile official announcement at the press conference to get on the bus for free, which has undoubtedly further increased the pressure on Tesla.

On the other hand, even in North America, which has full-version performance, fewer competitors, and cheaper prices (US$8000, approximately 5.8w RMB), Tesla’s FSD penetration rate is actually not as strong as expected.

Referring to CITIC Securities estimates, in North America, Tesla’s FSD penetration rate on Model 3 is only 5% to 7% in single digits, and its penetration rate on Model Y is 12% to 13%. The data given by YipitData, a market research firm in New York, is even more pessimistic:

The proportion of Tesla users who tried FSD and then decided to buy was 2%.

Considering that the FSD released this time only supports hardware versions of HW4.0 and above, and this version of hardware will not be installed until early 2024, the number of vehicles in China that are eligible to upgrade this smart driving system so far is almost the total amount delivered by Tesla last year:

1.789 million vehicles.

We may wish to use the optimistic data of 8% to calculate that if 8% of these vehicles subscribe to a full set of FSD services, it is expected to only bring about US$1.26 billion in revenue to Tesla throughout the year. For reference, Tesla’s revenue in China in 2024 was US$20.944 billion, a decrease of 3.7% from US$21.745 billion in the previous year.

Even with this money added, Tesla’s China’s performance can only barely maintain growth.

This performance is undoubtedly far from Musk’s ambition.

From this perspective, the entry of FSD into China may further verify the old saying in China: You can’t eat hot tofu in a hurry.