Thirteen years ago, Facebook engineer Calos Bueno wrote "The Full Stack," making the term "full stack" known to people for the first time. Thirteen years later, the automotive industry is moving towards intelligent electrification at a high speed, and "full stack self research capability" seems to have become a synonym for enterprise technological strength.

　　Especially in the summer of 2021, after Chen Hong, Chairman of SAIC Motor Corporation, stated the "soul theory", building a stack of self developed ecosystems in core technology fields such as autonomous driving, intelligent internet connection, and intelligent cockpit became one of the core means for automakers to declare discourse leadership to the market overnight.

　　The wind has changed direction

　　Two years later, the wind is still blowing, but the direction of the wind seems to have changed.

　　At the recent China Electric Vehicle 100 Person Conference, Huawei Yu Chengdong half ridiculed and half helplessly said, "The new forces in China, Li Xiang and Li Bin, are both here today. I believe they are unlikely to choose Huawei in intelligence. International giants will not choose us, and traditional car companies will not choose us if they are afraid of losing their soul.".

　　When the main engine manufacturers have played the strategic card of self research in the whole stack, the traditional vertical and linear supply chain system has also undergone changes, and the parts suppliers and the main engine manufacturers have fallen into a protracted supply and demand game.

　　Because of this, Yu Chengdong made a judgment that in the era of intelligence, Huawei cannot achieve Tier 1 like Bosch and Continental, because Huawei provides more software, algorithms, clouds, and chips, and cannot achieve standardization like traditional components. It must be more involved with automotive companies, deeply customized, and empowered.

　　When the perspective was switched to the host factory, Ideal CEO Li Xiang also reiterated the need for full stack self research at the meeting. "Previously, traditional head tier suppliers could develop the same system for many manufacturers, but using the supplier's software solutions, the iteration speed could not keep up with competition, so they had to self research.".

　　Yes, the design of the entire vehicle manufacturing, especially at the autonomous driving level, is quite complex, involving the integration of code algorithms, autonomous driving chips, and various sensor hardware. If you do not adopt full stack self research and bring the project into a controllable range, it is easy to encounter passive situations that require supplier support in subsequent iterations.

　　Previously, the MobileEye system was too closed to keep pace with Tesla's development, which ultimately led to their separation. This is a typical example. At that time, everyone found that it seemed that only by researching the entire stack, could they hold the initiative in their hands and make differentiated intelligent driving solutions in the field of intelligence.

　　Interestingly, two days ago, General Motors announced that it would completely abandon its cooperation with Apple, phase out Apple's Carplay, and instead use the native automotive system developed in collaboration with Google. BMW has previously stated that it may no longer be equipped with a carplay on future new models.

　　The pros and cons behind this are inexplicable, but there is an obvious starting point: Carplay allows users to mirror mobile phone data onto the vehicle's on-board screen, which largely invades the native space of the vehicle's computer system. Once the main engine factory cedes the vehicle and machine system to each other, it means losing the most important medium and position for human-computer interaction, which may be the direct reason why GE chose to develop itself.

　　In fact, in the era of traditional fuel vehicles, top players such as Toyota and Volkswagen also mostly adopt the form of self research in the fields of core components such as vehicle engine systems, power systems, and steering systems. After all, only by firmly holding the supply chain in hand through vertical integration can they ensure platformization, scale, and achieve cost and reliability advantages.

　　In the era of smart electric vehicles, the emergence of Tesla has also shown car companies the possibility and allure of "self research in a whole stack". Zero Run is one of our loyal "supporters". "Our comprehensive self-development is a comprehensive integration of software and hardware, with circuit boards, structures, software, and operating systems developed by ourselves.

　　Unfortunately, like in the era of traditional fuel vehicles, the true "self development of the entire stack" is a capability that only a few players have.

　　Li Peng, CEO of Youpao, is also reflecting: "The repeated business model of building cars behind closed doors in the automotive industry is necessary in the era of internal combustion engines, but can it continue in the era of smart cars? Just like in the era of smart phones, everyone uses Qualcomm chips, affecting the differentiation of brand building?"

　　Yes, look at today's BYD, which has such excellent vertical integration capabilities, and rarely boasts of "full stack self research". Therefore, it can be seen that in terms of intelligent layout, BYD will still choose to cooperate with companies such as Momenta, Sato Juchuang, Baidu, and Horizon to enter the field of autonomous driving.

　　Why is even BYD so cautious? In fact, from the moment the industry set off a "full stack of self research" craze, it was reminded that enterprises wanting to self research everything inevitably lead to huge waste in the industry, inefficient industrial development, and lagging product experience.

　　However, at that time, the hot investment atmosphere in the capital market was accompanied by so-called disruptive innovation, which made the entire stack of self research loud and sound, and the industry lost its rational soil. Now, as the capital market gradually cools down, there is a strong trend towards returning to the commercial nature of the entire stack of self research topics.

　　It must be acknowledged that the ideal of auto companies wanting to achieve full stack self research is worthy of respect, but the reality is brutal. This cruelty is not only reflected in the high intensity and long cycle of resource investment, but also puts forward high requirements for internal organizational management mechanisms.

　　Although Li Xiang supports self research, he also frankly said, "The automotive industry chain is so high and difficult, and the biggest challenge lies in organizational ability.". Whether an enterprise has an effective innovation mechanism within it, whether it can form an appropriate and reasonable talent echelon model, and whether it can build an orderly and efficient organizational structure is the key to building a stable and effective self research system.

　　It seems simple, but McKinsey has a very heartfelt survey data: only about 3% of enterprises have truly and completely achieved the established goal of Digital transformation. After analyzing cases, it is generally found that technical issues account for only a small proportion, with the main sticking points concentrated on the organization and business.

　　Is the entire stack self-developed or controllable?

　　In addition, we have been sparing no effort to explore, "The first half of new energy vehicles is electric, and the second half is intelligent." So what is the second half of intelligent vehicles? It should be scale, and the core element to achieve scale is to reduce costs and increase efficiency.

　　The reality is that if the entire intelligent ecosystem is truly vertically integrated from the underlying gateway, to sensor hardware, algorithm software, and operating systems, achieving full stack self research, the cost of silence can be significant from the perspective of maintaining leadership. This cost will naturally be transferred to the market, becoming a "roadblock" in front of scale.

　　Currently, a significant feature is that the L2/L2+function has achieved large-scale mass production and gradually penetrated into vehicles below 200000 yuan as standard equipment, and advanced assisted driving is also further exploring to around 300000 yuan.

　　According to data, in 2021, the number of new cars with standard L0-L2 level ADAS installed in China exceeded 8 million, with a penetration rate of nearly 40%. In 2022, this proportion increased to more than 45%. Some institutions predict that in 2025, the market penetration of domestic L0-L2 ADAS will exceed 91%, L2 penetration will exceed 45%, and the penetration of L3 and L4/L5 is expected to reach 8% and 1% respectively.

　　In other words, for a period of time, in the mid to low end intelligent driving market, Tier1's technology maturity has been quite high, with the advantage of scale, and there is no need for host manufacturers to work hard to develop independently.

　　Former Tesla Vice President of Production, Gray Rachel, once said, "If you are producing a product that fully utilizes other commercialized components and will not undergo significant innovation for a certain period of time, there is no point in producing it yourself.".

　　In the field of advanced assisted driving, most enterprises in the current industry are still unable to build a suitable research and development environment. Instead of choosing to "build a car behind closed doors," it is better to choose to develop in conjunction with Tier1, which means that the host manufacturer uses the supplier's basic platform base to create its own customized and differentiated application functions. This is clearly a more pragmatic, economical, and efficient technical route.

　　What's more, in terms of core software, with the rapid development of 5G communication, OTA, and cloud computing, complex requirements have been put forward for information technology such as bandwidth and storage for major enterprises. At the same time, around regulatory regulations, functional security, network security, privacy protection, and other fields will face significant challenges, and collaborative development is bound to be one of the trends in this process.

　　In the words of Chu Ruisong, vice president of Baidu Group, it is difficult to achieve "self development of the entire stack", and "controllable control of the entire stack" should be completely achieved.

　　However, from a commercial perspective, there is still that issue - with more partners, the supply chain system will change from the original vertical linear model to a multi-party parallel network model, which undoubtedly poses a serious challenge to cooperation efficiency.

　　In fact, relevant supply chain enterprises are aware of this issue. Therefore, in order to dispel the concerns of host manufacturers, suppliers such as Cambrian and Black Sesame are consciously building a complete set of easy-to-use, friendly, and promotable development tool chains to ensure the efficiency of the development optimization process.

　　In addition, OEMs like Great Wall Motor are also consciously promoting the platformization and universalization within the enterprise, establishing a unified hardware platform plan and a unified software baseline to lay the foundation for ecological partner interaction and in-depth collaborative development.

　　A single tree does not make a forest, and the main engine factory is rushing towards suppliers in both directions. As Wang Yuanli, CTO of Great Wall Motor, said, "Many people would say that mastering the ability of self research in the whole stack is equivalent to building a relatively closed ecology. We believe that on the contrary, mastering the ability of self research in the whole stack requires more in-depth collaboration and cooperation of partners."

　　After talking so much, I turn around and ask myself, everyone is claiming that the entire stack is self developed. What is the significance for the entire industry? In this era of redefining the boundaries of the industrial chain, it may be that people have a deeper understanding that "full stack self research is a capability, not a business model.".

　　Calos Bueno, who proposed the concept of "full stack", also pointed out that no one can be familiar with all aspects, but as a full stack, one can clearly see how each stack works up and down.