Wei Xiaoli Built Cores, Can The Market Still Tolerate A Tesla?

TraditionCar industryThe upstream and downstream supply chain of the division of labor has ushered in a reconstruction in the smart car era.The most obvious manifestation is the intelligent driving track.Several hills that constitute this track, with upstream chip manufacturers, asTier 2 exists, with midstream software suppliers, exist as Tier 1, downstream car companies, facing users as customers.

However, what we see now is that some players with clear division of labor in the original chain are building their own research and development capabilities from soft to hard.Auto companies such as Wei Xiaoli and some traditional OEMs of the head, algorithm suppliers such asMomenta has its own chip R & D team and recently coming from streaming news; chip manufacturer's autonomous driving algorithm is also in full swing, such as Nvidia, Horizon, and so on.Is this vertical industrial chain feasible?Once the smartphone industry has given some examples. Some of the manufacturers such as Apple successfully construct their own hard and hard moat, but after all, they are a minority, and more are integrators based on Qualcomm and Android.

How will Chen Tao Capital hold the Soft and Hard Starring Report Conference? How will the intelligent driving track develop in the future?Not long ago, Chen Tao Capital was held in ShanghaiAt the press conference of the self -driving software and hard -hard coordination development forum and the Xingxing Research Report, many industry experts gave reference answers.First of all, it is certain thatIt is reasonable to study from soft to hard, because the entire industry is in oneAt the stage of KNow-HOW, the chip dealer must design a chip from the software cognition; algorithm suppliers need to have a deeper understanding of the chip and better deploy the software on different platforms.

As for car companies, just like the smartphone industry, under the premise of large shipments, building a closed -loop chain from soft to hard is the most economical and moat. Tesla has given some answers.It's just that there is only one Apple. Tesla may have but not too much. How should most of the industrial chain players play?

01. Soft and hard, gradually becoming a trend

Before answering the above questions, we need to make a simple explanation of the so -called soft and hard.In intelligent driving, or in the entire upstream and downstream industrial chain of autonomous driving, it can be simply divided into several parts: the upper -layer application algorithm, the middle link software algorithm and the computing chip's middleware, operating system and other soft bottoms, and development toolsChain and so on, the bottom floor is the calculation platform, the core isSOC computing chip.The division of labor of the traditional automotive industry, from OEM to Tier 1 to Tier 2, the level of division of labor in the entire industry chain is relatively clear, but in the intelligent driving track, a very obvious trend in the recent period is that more and more players have begun to startPlay the way from soft to hard vertical research and development.Specifically, there are several types of representative players, and in the process of in -depth layout of the soft and hard integration, several fixed paradigms have been evolved.At the level of car companies, it is mainly based on the new forces of car manufacturing, such as Xiaopeng, Weilai, ideal, etc. At the beginning of its establishment, autonomous driving ability is used as one of the competitiveness of traditional car companies.At the algorithm level, the leader -in layout of self -research and leading the industry.

Xiaopeng released a self -developed chipAt the chip level, Turing is also represented by Wei Xiaoli. It has begun to include chip self -research into the planning plan earlier and the results are intensive this year.For example, at a press conference not long ago, Xiaopeng took out Xiaopeng Tu Ling chip; Weilai’sHeavenThe NX9031 computing chip will also be carried on ET9 in 2025; the ideal self -developed chip is still ongoing, I believe that there will be good news soon.

In addition, traditional car companies, such as BYD, Geely, have also included self -developed computing chips into the plan.This kind of software and hard -hardness dominated by car companies can refer to Tesla with the evolution to the final template.Heavy soft and hard -in -one mode.In the era of smartphones, the reference that can be found is Apple.

The second type of player is mainly algorithm suppliers. The mainstream practices of such players are more inclined to at the level of software. Based on the understanding of the chip, the algorithm of different computing platforms is efficiently migrated and customized to meet different car companiesCustomer needs.This is a kind ofThe soft and hard -in -one model is deeply bound by algorithm companies to deeply bind some chip platforms for research and development of production and mass production.

butMomenta, was exposed in July last year. If this road is smooth, the ultimate direction will be a deeper and deeper software and hard -and -hard delivery model.Finally, it is represented by chip suppliers and covers from hard and soft to the top.Except for the earliest software and hard -and hard all -round black box modeIn addition to Mobileye, including Nvidia and Horizon, they all work hard at the level of autonomous driving algorithms.In the research report of Chen Tao Capital, this model is defined as an intermediate state derived from the heavy soft and hard integration, that is, the chip manufacturer provides from the chip to some sensory algorithm (such as the horizon recruitmentProcedure5 The supporting facilities have perception algorithms).

The horizon is based on the journey6P releases the SuperDrive solution. However, at present, chip manufacturers are not limited to reference perception solutions at the software level. For example, the horizon is based on the latest journey 6P.At the fixed point on the car company, the estimated production time is expected to be in the third quarter of next year.Nvidia recruits Wu Xinzhou's behavior itself, that is, to build its own algorithm team and ability.In other words, from car companies to algorithm suppliers, to chip solution providers that should have appeared in Tier 2 characters, representative players in the entire industry are building their own hard and hard -and -hard industry chain.The trend of hard and hard one is today, and it is accelerating from light to heavy.But in fact, this is based on the long -term publicity of the industry -level propaganda and a clear maturity model with a clear division of labor in the automotive industry (similar to the smartphone industry. In addition to Apple and HuaweiAndroid cooperation) is a contradictory trend.

02, heavy soft and hard, motivation and opportunity

From the perspective of Chen Capital, the motivation behind the heavy soft and hard one is summarized to summarize2 points can be said.One is cost -driving. On the one hand, algorithm companies develop a custom chip or bind a chip in depth, which can use the chip capacity to avoid the waste of platform chip design; on the other handThe cost of chip reduces the overall expenses on the premise of satisfying sufficient shipments.

Compared with the cost of Nvidia and Tesla's self -developed chip, the second is that in terms of development efficiency. From the perspective of software to chip coordination, software vendors and hardware vendors need to cooperate closely with each other, but universal platforms (such as Nvidia) may need to take the test.Think of more needs and cannot quickly meet the needs of downstream manufacturers.In contrast, the benefits of soft and hard stacks are obvious.

In addition, whether it is an OEM or a chip dealer, or a middle software in the middleTier 1 requires sufficient full stack understanding of software and hardware to be presented at the product level.In particular, intelligent driving has begun to go to complex and changeable urban scenes, and a critical moment when you break through the L3 autonomous driving breakthrough, this is even more important.For example, Lu Peng, vice president of Horizon, said that SOC chips are not as good as power chips or power chips. The complexity of its design requires a strong software Know-How to design well.

There is a motivation to have opportunities. After all, from soft to hard or hard and soft, the middle discipline and business are not small.At present, the opportunity of the industry's opportunities to move towards deep hard and hard parts are mainly3 o'clock.

First of all, the software level technical solution convergence, whether it was in the past few yearsThe BEV+Transformer solution, or the recent end -to -end architecture, the industry can basically form a consensus in a short time;

The second is the difficulty of chip design. Thanks to the mature design of the semiconductor industry,The mature ecological links such as IP and Toolchain, the difficulty of the chip design itself is declining;

Finally, the input -output ratio. We mentioned above that the cost of self -developing chips is lower than purchasing in the cost of a single chip, but this premise is a single.R & D investment will be relatively stressful for the enterprise itself.

For this critical point, a data that is more consensus in the industry is1 million shipments.Regardless of whether it is a car company or a software Tier 1 now, the pressure of this shipping volume is not small.Chen Tao Capital believes that the above three standards or opportunities meet one of the companies to consider the conditions of the hard and hard layout, and meet the two companies to have the motivation to promote the layout of the hard and hard and hard.A company's choice strategy.

After HW 3.0, Tesla uses full stack of self -developed hardware that there is no doubt that Tesla meets these three standards and has already released the results on this road, and gives a reference answer to the outside world, that is, the severe ones, that is, the severe onesThe integrated layout of the software and hardware can indeed make the ceiling of the scheme higher, and it shows a huge advantage in cost control.

03. Will the next Tesla still appear?

It is also based on the previous article3 opportunities (or standards),Algorithm convergence and chip design are difficult to designFor most players, it is an objective condition that has been realized.This can also explain that whether it is car company or chip dealer, or software solution providers, in the industry has developed vertically, seeking its own general trend of hard and hard -and -hard all -stack capabilities.But whether the next Tesla can be born, it seems unclear.An important factor in front of many players is the problem of volume.

Wang Xiaohui, the person in charge of Tianzhong Technology Domain Control, Wang Xiaohui, the person in charge of Tianzhun Technology Realm Control, said that, such as Wei Xiaoli to build cores, a problem faced with a long -term share of the market. Only a certain proportion of the market for a long time will it meet the business of business.Logical input -output ratio.Therefore, he believes that the model of the OEMs can exist but not too many, at most, at most1-2.Yu Chendi, a partner of technology and technology, believes that referring to the PC industry and the smartphone industry, vertical division of labor solves the problem of no matter, and the level division of labor solves the problem of popularity.

The partner of the technical and technical and technical technical in the feet fell to the intelligent driving itself, high -speedNoa has reached the stage of popularization, so we can see that there are more division of labor models or light and hard -and -integrated models in this feature.

CityNoA's user experience has not reached the stage of qualitative change. In this transitional stage, someone will try to do vertical full stack soft and hard.However, the end of the industry will eventually be a model of level division of labor. With reference to the development trajectory of the automotive industry, the vehicle manufacturers did everything at first, and then split the Tier 1 giants such as Bosch and the mainland.So there are about two consensus formed here:

First of all, the full stack of hard and hard integrated trend is indeed in high -end intelligent driving and even autonomous driving.The Know-HOW stage will provide a high ceiling, and in the process, ecological closed giants such as Tesla and Apple are born;

Secondly, for most players in the industry, based on the maturity and convergence of technology, at the cost of cost, the level of division of labor is expected, andIntelligent driving, that is, the development mode of hard and hard decoupled decoupled.

Of course, this needs to meet a condition. Lu Peng believes that as the computing power supply of the chip is getting larger and larger, it is far more than the state of software development demand, and hard and hard decouples will become easier.Just as the era of smartphones, it is also the condition for the creation of Qualcomm Snapdragon processor and Android ecology. Xiaomi has made personalized development through this ecology, and eventually occupied the market to divide cakes with apples.