(Report Producer/Author: Anxin Securities, Zhao Yang, Xia Yingtao)
1. Full-stack self-development is not the only option, a variety of paths to achieve automotive intelligence
How to view the "full-stack self-research" boom of automobile intelligence? In recent years, from intelligent cockpit to intelligent driving, the process of car intelligence has been accelerating. In the wave of intelligence, a large number of car companies have regarded "full-stack self-research" as the foundation for deeply cultivating the era of intelligence. Among them, the new car-making forces represented by "Wei Xiaoli" took intelligence as the main selling point as early as the early stage of entrepreneurship, and took the lead in carrying out self-research layout. At present, many traditional car companies, including Geely, Great Wall, SAIC and so on, have also announced a full-stack self-developed roadmap. For example, in the "2025 Strategy", Geely Automobile proposed to build a full-stack self-development system covering electronic and electrical architecture, vehicle basic software, intelligent cockpit software and automatic driving software; Great Wall Motors proposed to build a full-stack self-developed technology research and development system at the press conference; and Nezha Automobile launched a full-stack self-developed intelligent and safe car platform - Shanhai Platform.
What factors ultimately affect the form of cooperation between suppliers and OEMs? Car companies have been involved in the upstream supply chain on the one hand to promote the rapid development of the industry, but at the same time, it has also caused the market to think about the reshaping of the competitive landscape of the intelligent automobile industry chain. Among the first questions is whether the space of Tier 1 and Tier 2 will be squeezed in the future, and 2) how their roles will change. We believe that in order to answer the above questions, we first need to clarify which areas the automakers will be involved in. Therefore, in the first chapter of this report, we will first explore the following topics: 1) for intelligent driving, how large-scale R&D investment is required to complete the software and hardware full stacks; 2) what conditions need to be met by OEMs to achieve the full stack of software and hardware in the long run; 3) which factors will affect the boundaries of the field in which automakers are involved in self-research.
1.1. From the oem and Tier 2 respectively, look at the investment required for full-stack self-development
We looked at OEMs and smart driving Tier 2 to determine the required investment in the full stack of software and hardware. First of all, we selected Tesla, "Wei Xiaoli" and other new car-making forces, as well as some traditional OEMs, to analyze their average annual R & D expenditure and R & D investment, so as to initially determine the required investment of software and hardware full stack. Then we picked Momenta and Horizon, two smart driving Tier 2 unicorns. Considering that such enterprises mainly focus on technology and product research and development in the early stage of establishment, and will not be put into production and manufacturing on a large scale, we assume that their average annual financing amount is approximately the investment required for research and development, so as to further analyze and verify the required investment for full-stack self-development.
1.1.1. Car companies' perspective: Differences in R&D investment bring about differences in the path to achieving intelligence
Tesla, new car manufacturing forces and traditional car companies three types of automakers R & D investment is quite different. By comparing the new car-making forces such as Tesla and "Wei Xiaoli" and the average R&D investment of some traditional automakers in the past three years, from the absolute value of R&D investment, we find that the average annual investment in R&D of Tesla is more than 10 billion yuan; the R&D investment of new car-making forces such as "Wei Xiaoli" is in the range of 1.5-3.4 billion yuan; the R&D investment of traditional automakers is more differentiated, the average investment of SAIC Motor Group is 14.1 billion yuan, BYD is 6.36 billion yuan, and the average value of typical enterprises we select is 50 yuan per year About 100 million yuan.
Tesla is a typical representative of the full stack of software and hardware self-research, with an average annual R&D investment of more than 10 billion. Tesla's R&D investment from 2019 to 2021 was 9.369/97.29/169 million yuan, respectively, an average of about 12 billion yuan per year. Among them, in terms of intelligence, Tesla's research and development investment is mainly oriented to 3 directions, that is, the construction of software, hardware and data closed-loop. 1) Algorithm software: Tesla's Autopilot has realized L2+ level intelligent driving functions including automatic parking, urban street assisted driving, etc., and has completed large-scale commercialization, opening up a business model for software monetization, and its automatic driving optional package price is currently 12,000 US dollars. 2) Chip hardware: Tesla launched its self-developed FSD chip in 2019, with a computing power of 72TOPS, as the hardware basis of Autopilot 3.0, installed on models such as Model S\Model X\Model 3. 3) Build a data closed loop: In addition to the more familiar Autopilot and FSD chips on the market, Tesla has also invested a lot of resources in the construction of data closed loop. In the "AI Day" of 2021, Tesla released the supercomputer Dojo loaded with the self-developed AI training chip D1 for big data processing and analysis, training the entire autonomous driving system including Autopilot, and further accelerating the iterative upgrading of autonomous driving capabilities.
As a representative of the domestic software full stack, Xiaopeng Automobile has an average annual R&D investment of about 2.2 billion. In 2018, Xiaopeng launched the automatic driving system XPILOT, whose core functions and underlying algorithms are independently developed, while hardware parts such as domain controllers, sensors, etc. adopt the solutions of third-party vendors. From the first model G3 listed in 2018 to the P5 mass production at the end of 2021, Xiaopeng's automatic driving system has been upgraded from XPILOT 2.5 to XPILOT 4.0, in the previous upgrades, the company's intelligent driving functions have successively covered high-speed NGP, memory parking and other L2+ level functions, and according to Xiaopeng 2021 performance exchange meeting, the city NGP function is scheduled to be officially opened at the end of the second quarter of 2022, so as to achieve continuous coverage of the entire intelligent driving scene. Looking back at Xiaopeng's R&D investment, from 2019 to 2021H1, it was 2.07/17.26/1.399 billion yuan, respectively, with an average annual R&D of about 2.2 billion yuan.
The absolute value of R&D investment of traditional car companies is relatively high, and it is mainly invested in electrification and new model research and development, and there is still a lot of room for improvement in future intelligent investment. In the past three years, the average R&D investment of 6 traditional car companies selected, including SAIC, Geely, BYD, Great Wall, Changan and GAC, is about 5 billion. Although the absolute value of R&D investment in traditional car companies is relatively large, the R&D investment is currently relatively scattered, mainly reflected in the following two aspects: 1) The research and development of new models, The Great Wall launched 7 new models in 2020; BYD has also launched a number of new models in recent years, and has continuously upgraded the upgrading of existing models. 2) The development of technical projects, battery safety applications, engine production, and collision laboratory capability upgrade construction projects are the main capital investments of Changan in 2020; Geely focuses on the research and development of automobile engines, transmissions and automobile modeling correlations.
Among the above three types of car companies, Xiaopeng and Tesla have realized the full-stack self-development of automatic driving software and software and hardware respectively, and traditional car companies are currently focusing on model development design and electrification transformation, and the research and development of automatic driving started relatively late. Combined with the scale and investment direction of its R&D investment, we preliminarily predict that the R&D investment of at least 2 billion yuan per year is required to achieve the full stack of software, and the R&D investment scale of the full stack of software and hardware is about 10 billion yuan per year.
1.1.2. Tier 2 Perspective: Momenta and horizon, typical of both soft and hard ends
From the financing amount of the smart driving Tier 2, further analyze and verify the required investment of the full stack self-development. Above, from the perspective of R&D investment of car companies, we measured the R&D investment of software full stack and software and hardware full stack, in order to ensure the effectiveness and accuracy of the calculation, we selected Momenta and Horizon, two intelligent driving Tier 2 suppliers. Considering that the company mainly focused on technology and product research and development in the early stage of establishment, and has not yet entered the stage of large-scale production and manufacturing, we will make its average annual financing amount almost the investment required for research and development, so as to further analyze and verify the required investment for full-stack self-development from the perspective of financing amount.
Software and algorithm solution representative: Momenta can provide mass production solutions for autonomous driving software, and since its establishment, the amount of financing has reached nearly 10 billion. Founded in September 2016, Momenta is a company that provides software algorithms for perception, positioning, decision-making, and different levels of autonomous driving software solutions, which we see as representative of software and algorithm solutions. According to the information disclosed on the company's official website, Momenta has raised 9.75 billion yuan in financing since its establishment, with an average annual apportionment of 1.95 billion yuan. In terms of product lines, the company provides mass production autonomous driving (Mpilot) and fully unmanned driving (MSD) solutions based on a unified mass production sensor solution, and the current cooperation between SAIC Zhiji L7, Lotus, Great Wall Salon Zhixing and Momenta is based on the Mpilot business line; on MSD, Momenta and SAIC have joined forces to build the Robotaxi fleet.
Hardware solution representative: Horizon's chips have obtained more than 40 front-loading mass production projects, and the cumulative financing since its establishment has exceeded 14 billion yuan. Founded in July 2015, Horizon has launched a series of AI chips for intelligent driving and AIoT since its inception, and has achieved vehicle-grade chip preload, which we consider as a representative of hardware solutions. According to the information disclosed on the company's official website, Horizon has accumulated 14.3 billion yuan in financing since its establishment, with an average annual apportionment of 2.383 billion yuan. At present, Horizon has formed three major product matrices in automatic driving, such as the vehicle intelligent computing chip journey series, the vehicle intelligent computing platform Matrix series, and the development platform Tiangong Kaiwu. As of January 7, 2022, the two chips of Journey 2 and Journey 3 have shipped a total of 1 million pieces, and have obtained the fixed point of 40+ front-loading mass production projects of SAIC, BYD, Ideal and other car companies.
The R&D investment of software full stack\software and hardware full stack may be 20\5 billion yuan per year. Through the above combing of the R&D expenses and investment directions of different car companies and intelligent driving Tier 2, we found that Xiaopeng's average annual R&D investment and Momenta's average annual financing amount were basically the same, both of which were 2 billion yuan, further confirming that the R&D investment required for the software full stack was about 2 billion. As for the software and hardware full stack, considering that the software and hardware solution represents the unknown amount of round financing on the horizon part, and superimposed on its self-hematopoietic capacity, based on the average R&D investment of 2.383 billion yuan, we infer that the development cost of intelligent driving hardware, including chips, may be close to 3 billion yuan per year, while the R&D investment of the software and hardware full stack needs to reach 5 billion yuan per year. Tesla's average annual R&D investment of about 10 billion yuan includes its investment in electrification such as three-electric systems, and is not limited to the research and development of soft and hard stacks for autonomous driving.
1.2. Under the steady state of the industry, the R&D expense ratio returns to the stable value, and the volume of revenue becomes the key to whether R&D investment can be commercially sustained
The revenue scale of OEMs determines whether the full-stack self-development can be sustained. After clarifying the scale of R&D investment required for full-stack self-development, the next question we discuss is what kind of oem can achieve full-stack self-research of software and hardware.
Based on the R&D expense ratio of each car company, we further explore the requirements of full-stack self-research on the volume of operating income of car companies. First of all, we have counted the R& D expense rate of various car companies, and the R & D expense rate of traditional car companies is generally low, averaging 2.58%, while the R & D expense rate of the three new car-making forces "Wei Xiaoli" is above 10%. Tesla's R&D expense ratio in 2020 is 4.73%, between traditional car companies and new car-making forces.
Then we need to choose the appropriate R&D expense ratio to calculate the revenue scale required for full-stack self-development. We believe that the R& D expense ratio of traditional car manufacturers and Tesla can better represent the long-term steady state level of the automobile market, and the current revenue of new car-making forces is small, growing rapidly, and has not yet entered a steady state. From our retrospective of Tesla's R& D investment, the current development stage of the new car-making force is similar to Tesla in the 2015-2017 period, and its research and development expense rate is relatively high, and the main products have not yet reached a certain level. Referring to Tesla's R&D expense rate data, in the long run, when the operating income of the new car-making forces reaches a certain scale, its R&D expense rate is expected to drop significantly.
We believe that the R&D investment level of traditional car companies is determined under the long-term market game of the automobile industry and tends to be the reference value of the industry's steady state. Therefore, in the long run, when the technological revolution at a certain stage tends to be stable, the current round of technological evolution is represented by intelligence, and the automobile market develops to the steady-state stage again, the level of R& D investment of traditional car companies can be used as a reference value. In addition, considering the fluctuations in the level of R&D investment that may result from the business model changes brought about by Tesla (such as the increase in the overall gross profit margin brought about by the increase in the proportion of software subscription payments), we finally chose the R&D expense ratio of Tesla and traditional car companies and used it as the upper and lower limits of the R&D expense ratio range (2.58%-4.73%).
After calculating the R&D investment of the software and hardware full stack and the reasonable R&D expense rate range, we calculated that the revenue volume corresponding to the realization of the software stack should be more than 40 billion, and the corresponding revenue volume of the software and hardware full stack self-research institute should reach more than 100 billion. Based on Tesla's R&D expense rate, the revenue scale that needs to be realized to achieve software full stack, hardware full stack, and software and hardware full stack is 423/634/105.7 billion yuan, and the revenue scale that needs to be realized to achieve software full stack, hardware full stack, and software and hardware full stack is 776/1164/194.1 billion yuan, respectively. (Source: Future Think Tank)
2. Analogous to mobile phones, intelligence has not increased market concentration, and small and medium-sized car manufacturers are expected to exist for a long time
In the future, how many car companies can reach the revenue volume of full-stack self-research, we believe that the answer to this question mainly depends on the concentration and overall capacity of the intelligent automobile market. The total capacity of the global automotive market has not changed much, so market concentration is the key. Industry concentration increases, more car companies in the head will have enough income to support the full stack self-development, on the contrary, cooperation with third parties will become the main option for car companies. Therefore, the focus of this chapter is to analyze how the concentration of OEMs will change under the wave of intelligence.
Learn from the past, use smart phones as a reference. As products that have also experienced the wave of intelligence, the industrial changes faced by mobile phones and automobiles are highly similar. Among them, the mobile phone from the communication tool to the current smart phone, and now the car is also from the means of transportation to the development of intelligent terminals, we hope to discuss the development trend of the concentration of the mobile phone market by analogy with the automobile market.
Intelligence has not caused the concentration of the mobile phone market to increase, and the industry has shown a trend of brand dispersion in the early stage of intelligence. Looking back at the feature phone era, the mobile phone market has shown a highly concentrated trend, according to Garnter data, the world's top five brands in 2005 were Nokia, Motorola, Samsung, Sony Ericsson and LG, and their market concentration is close to 80%. After 2009, the wave of intellectualization struck, a number of emerging brands entered the market with the reshaping of the industrial chain, and smart phone brands such as Apple, Samsung, Xiaomi, Huawei, and HTC emerged, and the market concentration of the top five brands declined. After 2016, the smart phone market gradually changed from an incremental market to a stock market, and the head manufacturers further expanded their market share by virtue of their advantages in brand, supply chain and capital, and the market concentration gradually rose to more than 60%. However, on the other hand, the current market concentration in the era of smart machines has not yet recovered to the level of the functional machine period, mainly because the industrial chain of smart machines is more complex than that of functional machines, product demand is more fragmented, and there are more subdivided brands and market segments.
The automotive market CR5 is slightly over 30%, and the concentration is significantly lower than that of the mobile phone market. Different from the highly concentrated mobile phone market, according to GAD database statistics, the concentration of the world's top five automobile brand markets has exceeded 30% in recent years, while the market concentration of the top ten automobile brands is about 50%, and the concentration is significantly lower than that of the mobile phone market. We believe that the main reason for the high degree of discretization of the automobile market can be attributed to the following three points, namely, the high complexity of the industry, the long industrial chain, and the long product iteration cycle.
1) The number of auto parts is about 30,000, which is 100 times that of smartphones. Comparing the number of smartphones and auto parts, the current number of parts of smart phones includes chips, photoacoustic components, displays, structural parts, batteries, PCBs, etc., with the number of around 200. At present, the parts of a fuel vehicle are about 100 times that of smart phones, about 30,000 or so, covering a variety of electronic and mechanical components such as motors, batteries, electronic controls, circuit systems, engines, chassis, and dashboards
2) The auto parts supply chain has a pyramid structure, and the scale of suppliers is very large. Combined with the above, due to the small number of mobile phone parts, so there are fewer parts suppliers, most mobile phone manufacturers will directly dock with some core parts suppliers, taking Apple mobile phones as an example, the number of core suppliers is only 200. The automobile industry, with the vehicle manufacturing industry as the core, extends upward to the parts manufacturing industry and related basic industries. Upstream enterprises are responsible for the modularization, systematic development and design and manufacture of parts, forming a pyramid complex structure of "parts -components-components-system-system assemblies", so the scale of auto parts suppliers is larger than that of mobile phone parts suppliers, such as the number of suppliers of Volkswagen is as high as 40,000, several times the number of suppliers of Apple mobile phones.
3) Subject to the complexity of the industrial chain, the research and development cycle of automobiles is usually 4-5 years. In addition, automobile research and development is a very complex project, from the step point of view, to the vehicle development V mode process as an example, the development of a vehicle needs to go through from the vehicle demand \ power system requirements \ hardware and software requirements analysis, hardware implementation and software modeling, software and hardware unit testing, subsystem \ power system integration testing, vehicle calibration and verification of the whole set of processes; from the perspective of time, a car from research and development to the market generally takes 4-5 years, the average replacement cycle of German cars before is 7 years. Compared with smart phones, Apple's mobile phone iteration speed is basically one generation a year, and The Frequency of Android iteration is higher, almost half a year. Looking to the future, we believe that as many Internet companies join the car-making team, the R&D iteration speed of the entire automotive industry will accelerate, but it is expected that the overall R&D cycle will be about 2-3 years.
Automobiles are the most optional consumption with the highest single value, and a single market segment is enough to support the long-term survival of small and medium-sized car manufacturers. In addition to the relative complexity of the upstream supply chain, we believe that the numerous downstream segments are also one of the reasons for the low concentration of the automotive industry. As the optional consumption with the highest single value, the sales volume of a single market segment in the automotive industry is enough to support the long-term survival of a small and medium-sized car factory. According to the data of the owner's home, there are many small and medium-sized car manufacturers and subdivided brands whose annual sales level is only around 10,000 to 10,000 vehicles. At the same time, we have also seen that Great Wall Motors has launched two sub-brands of Euler and Mech Dragon in the past two years, focusing on the female market and the geek market respectively. Therefore, we judge that in the future smart car market, in addition to the head brand, most of the remaining market share may continue to be divided by small and medium-sized car companies or sub-brands.
With many new players entering the game, the concentration of the smart car market is expected to decline. With the opening of the era of automotive intelligence, most emerging roles have entered the game, and we have seen the shadow of the early stage of the development of smart phones. At the macro level, automotive market concentration has declined since the beginning of 2018, with CR5 falling from 34% to 32% and CR10 from 52% to 49%. At the micro level, many new players have entered the automotive intelligence, including sub-brands of traditional car manufacturers, Internet companies, ICT giants, and new car-making forces. Therefore, we expect that in the future, the concentration of the automobile market is expected to replicate the trend of the concentration of the mobile phone market ten years ago, and there will be a certain degree of decline.
Therefore, we believe that among the many reasons affecting the concentration of the automobile market, factors such as the large number of parts, the complexity of the industrial chain, the high unit price of products, and the large number of downstream market segments have not changed due to intelligence. Therefore, looking forward to the future, we do not think that in the short term, the concentration of the automobile market will increase significantly, on the contrary, as mentioned above, the concentration of the industry is expected to decline in the early stage of intelligence, in other words, in the future for a long time, there will be a large number of small and medium-sized car factories in the automobile market.
In summary, based on the above discussion on the ability margin and industry concentration of car companies, we have the following conclusions: Software full stack or software and hardware full stack requires long-term investment of car companies and a certain amount of revenue as a support, and in the future for a long period of time, there will still be a large number of small and medium-sized car companies in the industry, whether it is revenue or research and development can not meet this requirement.
Based on the above judgment, we expect that the future path to intelligence of car companies can be divided into the following three categories: 1) software and hardware full stack: this type of car companies aims to achieve the optimal solution of experience and cost through a high degree of coupling of algorithms and chips, but there are also the highest requirements for the research and development and revenue volume of car companies, and we believe that car companies that can achieve this path will be more similar to technology companies. 2) Software full stack: Such car companies will take intelligence as the selling point, and create a competitive advantage of differentiation through the continuous iteration ability of automatic driving algorithms. 3) Embrace third parties: Such car companies will use external software and hardware suppliers to achieve intelligence, which will maintain intelligence at an average level at the same time, with advantages in some directions, in the market segment has a high market share. (Source: Future Think Tank)
3. Climbing the ladder, the proportions of the three roles determine the desay Savih long-term space
3.1. The three roles of the future Tier 1: foundry, hardware solution provider, and automotive intelligence enabler
In the future, car companies will cooperate with Tier 1 with different capability levels to complete intelligent puzzles. Above, through the analysis of the capacity boundary of the depot and the concentration of the automobile market, we summarized the three major paths of the car company to intelligence, namely the full-stack self-research of software and hardware, the self-development of the software full-stack and the embrace of third parties. We believe that the realization of these three development paths requires the use of partners and third-party suppliers of different capability levels, which also correspond to the three possible roles of the future Tier 1: foundry, hardware supplier and automotive intelligent enabler.
3.1.1. Foundries: Large-scale manufacturing and cost control capabilities are key
The core capabilities of the foundry are large-scale manufacturing and cost control, and they enjoy a relatively limited profit margin. For automakers with full-stack software and hardware, they need an OEM role, similar to Luxun, Goertek and Apple. For the assembly business, precision manufacturing capabilities and cost advantages are key. For example, in terms of production and manufacturing, intelligent manufacturing technology can ensure the excellent level of yield and delivery of the factory; in terms of cost control, self-made parts and scale advantages have always been the two major weapons for cost reduction and efficiency increase. In terms of business model, based on the gross profit margin level disclosed in the annual reports of Luxun Precision and Goertek, such suppliers usually enjoy a gross profit margin of about 15-20%, and need to increase manufacturing efficiency, increase the proportion of self-research, or expand the volume of revenue to achieve growth.
3.1.2. Hardware Solution Providers: Understanding hardware and vehicle specifications does not happen overnight
The cooperation between Desay and Xiaopeng is an excellent example of a hardware solution provider empowering a software full-stack car factory. For automakers with full-stack software, they need a partner to provide hardware solutions. For example, the Xiaopeng P7 is equipped with Desay SV's intelligent driving domain controller IPU03 based on the mass production of Nvidia Xavier chips, and the tripartite cooperation model provides Nvidia with chips, reference designs and development boards; Desay SV is responsible for the design and production of hardware on the one hand, and also carries out software development of BSP, drivers and part of the middleware; Xiaopeng has developed a series of algorithms such as perception and decision-making and upper-level application software. We believe that the cooperation between Desay and Xiaopeng is an excellent example of a hardware solution provider empowering a software full-stack car factory.
We believe that unlike the foundry, hardware solution providers should have the following four capabilities in addition to basic manufacturing capabilities:
1) In-depth understanding of chips and board-level design capabilities. Hardware solution providers not only need to have an in-depth understanding of the architecture, computing power, and adaptation of different versions of chips, but also need to work with car companies to design board-level design, such as the size of the development board, the number of interfaces, electromagnetic interference, heat dissipation, etc. We believe that this ability does not work overnight, requires a lot of trial and error and time accumulation, and to a certain extent creates the scarcity of such roles.
2) Understanding of vehicle specification level standards and functional safety. The automotive industry has an independent system for safety and stability, which is mainly reflected in the explicit and implicit aspects, the explicit aspect lies in the mastery of various safety rules, technical specifications such as ISO, AEC and other standards, and the implicit aspect requires Tier 1 to cooperate with the car manufacturer for a long time, so as to gain the trust of the car manufacturer. We think both of these capabilities come from long-term interactions with automakers.
3) The development ability of the underlying software. In the process of creating hardware solutions, it will involve a series of low-level software development such as BSP, hardware drivers, middleware, etc. Gao Dapeng, general manager of Desay SV, once mentioned in an interview with Gaz Motors, "In 2018, the number of Desay SV software engineers has accounted for 67% of the entire R & D team, and in 2019 this data has reached 70%, and a large amount of work and costs are invested in software for the products delivered by the company every year." We believe that software capabilities are key to differentiating the foundry from the hardware solution provider, and will become increasingly important in the future.
4) Localization service and rapid response ability. Generally speaking, domain controllers need to cooperate with car companies to complete A\B\C multi-round tests from making prototypes to finally getting on the car, and constantly discover and solve problems such as adaptability and compatibility during the testing process, while optimizing design, maintaining stability and efficiency, so there are also higher requirements for local services and timely response capabilities of hardware suppliers.
From the perspective of business model, due to the need for more capabilities, hardware solution providers will also enjoy higher gross interest rates than the foundry, usually around 20-25% (refer to Desay SV's current gross margin level).
3.1.3. Automotive intelligent enabler: business model upgrade, software and hardware architecture and program iteration capabilities become the winners and losers
For automakers that fully embrace third parties, they need a role in providing intelligent overall solutions, that is, automotive intelligence enablers. Such roles are similar to the strong Tier 1 in the traditional automotive industry, such as Bosch's direct supply of standard assemblies and modules to OEMs. We believe that software and hardware total solution providers will face two main challenges:
1) For the accumulation of intelligent driving software and hardware architecture. The overall solution supplier must first determine the appropriate hardware matching according to the needs of the depot, such as the number, type, position, etc. of the sensor, as well as the calculation unit scheme, such as computing power, cost, etc.; and then based on the hardware matching, with the appropriate software and algorithms, and finally achieve intelligence. Therefore, we believe that the accumulation of software and hardware architecture for various types of intelligent driving is the top priority of the intelligent enabler of automobiles.
2) Through the data closed loop, the program iteration ability is formed. At present, the actual data is collected through the mass-produced models, the data is returned to the analysis and labeling, and then the new algorithm is trained and iterated, and the continuous iteration ability of the algorithm is finally formed. Therefore, we believe that this ability also determines the advantages and disadvantages of the intelligent enabler of the car.
In terms of business model, considering the addition of software algorithms and the stronger position as a turnkey total solution provider, we expect the gross profit margin of such total solution providers to reach around 35-40% in reference to the gross profit margin of Bosch's subsidiary in India (more than 80% of the automotive business).
3.2. Hardware solution providers: There is no need to worry too much about the competitive landscape, and pay attention to the technical feedback brought by high-quality customers
At this stage, Desai is mainly to provide hardware solutions, and the domain controllers are constantly iterating out new and mass-producing. At present, Desay SV mainly provides hardware solutions, and its domain controller product matrix includes the main cost-effective IPU01\IPU02 and the main high-performance IPU03\IPU04. Based on the IPU02 developed by Texas Instruments (TI) Jacinto TDA4 chip, the computing power range is 4-32TOPS, which supports multiple scenarios of high-low speed automatic driving and realizes L2+ level automatic driving. The IPU03 with NVIDIA Xaiver hash rate 30TOPS has been mass-produced on the Scale of Xiaopeng P7, and Xiaopeng P5 will continue to use IPU03. IPU04 is a new generation of intelligent driving domain controller following IPU03, equipped with NVIDIA Orin X chip, hash rate coverage 254-2000TOPS, has been rolled off the production line at the end of September 2021, and received orders from many OEMs including Ideal.
The first-mover advantage is significant, and there is no need to worry too much about the competitive landscape. We believe that the company's advantages in hardware solutions are mainly now: 1) in-depth cooperation with NVIDIA, occupying a first-mover advantage. Desay SV is the only domestic Tier 1 with NVIDIA, and in the Xavier era, the company took the lead in running through the NVIDIA tool chain through IPU03, occupying a first-mover advantage in NVIDIA's ecosystem. 2) Deep understanding of vehicle specification level standards and functional safety levels. As a veteran Tier 1, Desay SV has accumulated a deep and comprehensive understanding of vehicle regulations and functional safety through long-term cooperation with car companies, and the current IPU03 has reached the ISO26262 functional safety ASIL D level. 3) The ability to quickly land enables car companies to shorten the development cycle. IPU03 is currently the only large-scale mass production of domain controllers, while IPU04 has also been off the production line at the end of September 2021, and is expected to be mass production in 2022, through the rapid landing of domain controller products, to help car companies in the early stage of intelligence, quickly launch products, fully occupy the user's mind.
Pay attention to the dual boost effect of the above business model on the company's revenue side and technical side. We believe that the follow-up Desay SV for the main intelligent car companies will continue to provide hardware solutions business model. For such car companies, high-level automatic driving capabilities are the core competitiveness, so mastering automatic driving-related algorithms for software self-research is an inevitable choice. We recommend paying attention to the dual boost effect of the above business model on the company's revenue side and the technical side: 1) Revenue boost effect, high-level autonomous driving car companies have large sales, and the unit price of superimposed domain controllers is high, and cooperating with them has a greater role in promoting the company's revenue side. 2) Technology feedback effect, we believe that because in the early stage of intelligence, the boundaries of various roles in the industry are blurred, so the cooperation between all parties in the project will be more in-depth and the interaction will be more frequent. Therefore, cooperation with high-level autonomous vehicle companies can feed back Desay SV's capabilities in autonomous driving, so that the accumulation of chips, algorithms, operating systems, and autonomous driving software and hardware architectures has reached a higher level.
3.3. Hardware foundry: will not be the main role of Desay SV in the future
Hardware foundry will not be the main business model of Desay SV in the future. At present, the degradation of the business model is one of the several concerns of the market for Desai. Specifically, the market believes that in the face of software and hardware full-stack self-developed enterprises, The role of Desay SV will be regressed from a hardware solution provider to a foundry party. However, we believe that such concerns are not necessary, mainly because based on the above argument, only a small number of car manufacturers that have the ability to achieve full-stack self-research of software and hardware in the future are only a small number, and most car manufacturers cannot reach a certain amount of revenue volume to support full-stack self-research. In the long run, we do not rule out that Desai will expand the scope of cooperation with customers in the case of rich production capacity, but hardware OEM will not become the company's main business model.
3.4. Automotive Intelligent Enabler: Business model upgrade under the proposition of "Chuang ling zhi xing"
In the above article, we have explained that in the future, there will be a large number of third-party car manufacturers in the intelligent automotive industry from the two dimensions of the overall concentration and development trend of the automotive industry and the R&D investment and volume required to achieve full-stack self-development. Based on this, we believe that there are a large number of car manufacturers who have demand for the overall solution of software and hardware, and the intelligent driving Tier 1 is expected to evolve into a "turnkey" business model in the future and become an enabler of automotive intelligence.
Different from Tier 2's core competitiveness in technical advantages, we believe that the core competitiveness of automotive intelligent enablers lies in: 1) channel construction on the depot side; 2) the ability to understand intelligent software and hardware architecture; 3) the diversity of solutions to meet different needs. In terms of channels, Desay SV, as a veteran Tier 1 that has been deeply involved in the automotive market for many years, has a customer range covering European and American car companies, Japanese car companies, and domestic independent brand car companies, and has established deep cooperative relations with many brands. Therefore, we believe that the core of the discussion is in the latter two, namely the ability to understand the software and hardware framework for intelligent driving, and the diversity of solutions.
Forward-looking R&D + large-scale mass production, dual-track parallel development of software and hardware architecture accumulation. We believe that Desay SV's accumulation of intelligent driving software and hardware architecture mainly comes from two aspects: 1) R&D centers have been set up in many places for intelligent driving technology research and development. The company has set up R&D branches in Singapore, Europe, Nanjing, Chengdu, Shanghai, Shenzhen and other places to develop new products in the field of internet of vehicles and intelligent driving, in addition, the company set up a new research team in Singapore in 2019, responsible for the development of L4 and L5 level autonomous driving and related cutting-edge technologies such as network security. In the same year, the company obtained a licence (M1) for the first phase of road testing of driverless vehicles, which allowed the Desay SV test vehicle to drive on public roads in special areas of Singapore. 2) Strengthen the accumulation of intelligent driving software and hardware architecture through long-term interactive cooperation and research and development with car companies. At present, the entire intelligent automobile industry is in the early stage of development, car companies and suppliers are not the traditional vertical, assembly line cooperation model, the division of labor between the two sides of the boundary is very blurred, there is no standardized products on the market, so through the in-depth cooperation with some of the main intelligent car companies, Desay SV Whether it is the demand for intelligent driving of car companies, or the software and hardware architecture, can be precipitated.
Cooperation is the mainstay, supplemented by self-research, or the optimal solution at the current algorithm level. In addition to the hardware and architecture level, the overall solution of intelligent driving also requires algorithms, including perception, decision-making, control and other algorithms, of which the algorithms of perception and decision-making are particularly important. At the algorithm level, we believe that the optimal solution at present is cooperation-based and self-researched. First of all, the current L2 level ADAS algorithm includes adaptive cruise, lane keeping, automatic brake assist, automatic parking, etc. has been relatively mature, there are more algorithms on the market Suppliers can provide relevant capabilities, so from the pursuit of efficiency, commercial division of labor logic, the purchase of mature algorithms is a more appropriate choice. Secondly, high-level intelligent driving algorithms, such as high-speed pilotage, valet parking, etc. have higher requirements for perception algorithms, high-precision maps and positioning, and their research and development requires a lot of investment and continuous iteration, and the company is still in the early stage of exploration of the above algorithm development, so it is also a choice to cooperate with the excellent Tier 2 in the industry.
Invest in intelligent driving Tier 2, make up for the shortcomings of software algorithms, and build an intelligent driving ecosystem. According to the information disclosed in Thesay SV's annual report, the company invested MOMENTA 63.59 million yuan in 2019 to complement autonomous driving technology. According to nullmax official website and Desay SV annual report, in 2019 the company invested 34.36 million yuan in Nullmax Newwood Technology, and the cooperation between the two sides was fully carried out in the aspects of automatic driving hardware, software, testing and data, nullmax is mainly responsible for multiple modules such as target monitoring and identification, and provides support in sensor selection and underlying software to jointly create an autonomous driving front-loading solution for car manufacturers. Based on maxieYE's official website and Desay SV's annual report, in 2021, Desay SV led the investment, and MAXIEYE completed a 300 million series of financing, and reached a strategic cooperation to jointly work on the research and development of high-end autonomous driving with large computing power, and cooperated to build L1 to L4 full-stack autonomous driving program development and operation service capabilities. At the same time, the "Jiukui Plan" was released to deploy the core technology of autonomous driving of commercial vehicles and accelerate the commercialization of autonomous driving. We believe that investing in the above-mentioned intelligent driving Tier 2 can, to a certain extent, make up for the company's current shortcomings in software algorithms, and have a positive effect on its evolution to "automotive intelligent enabler".
Short-term attention to the overall solution of software and hardware on the revenue side of the improvement effect. At present, the market's focus on Desay SV in the field of intelligent driving is still on its domain controller, but we believe that as the company's business model evolves to an enabler of automotive intelligence, the company's sales of various types of sensor hardware, such as cameras, millimeter wave radar, ultrasonic radar, etc., will also become an important revenue growth pole. According to the statistics of automotive electronics and software, the current realization of L2 level automatic driving requires 5-8 millimeter-wave radar, 7-10 cameras, 12 ultrasonic radars, such as the Xiaopeng P7 equipped with IPU03 The whole vehicle is equipped with 31 sensors, including 5 millimeter wave radar, 12 ultrasonic radars, 4 surround view cameras, 10 high-perception cameras and high-precision positioning systems. We believe that theoretically choosing a product from the same vendor is beneficial for hardware adaptation, sensor fusion, and subsequent services.
Long-term optimism about gross margin optimization brought about by business model upgrades. In the previous article, we mentioned that due to the addition of software and the change of the status of the industrial chain, the automotive intelligent enabler will enjoy a higher level of profit, and the gross profit margin is expected to reach about 35-40%. In our view, as far as Desai is concerned, this goal is not untouchable, and the following two aspects can be observed:
1) Whether the proportion of the company's self-developed software and algorithms has increased. Although we judged above that the company's exploration of perception, decision-making and other algorithms is still in its infancy, including application software, car control algorithms, middleware, OTA and other fields can become the company's first development object. We expect that as the proportion of self-developed software algorithms rises, the company's profit level will show an upward trend.
2) Whether there are technical services for carrying out similar consultations. We refer to Bosch's role in the traditional automotive industry chain, in the early stage of the development of each new technology, it will first carry out joint research and development with the head car factory, and then after the various schemes are formed and verified, they will be introduced into other car manufacturers. At this time, if the relevant car manufacturer requires a customized solution, Bosch will charge the relevant technical development fee. In our view, the cost of such technology development is more similar to that of consulting services, which is a high value-added business model. Referring to the gross profit margin level of Bosch's Indian subsidiary, we expect that in the future, as the enabler of automotive intelligence, Desay SV's gross interest rate is expected to reach more than 35%.
4. Profit forecast and investment analysis
4.1. Basic Assumptions and Operating Income Projections
First, intelligent cockpit business
1) The company has been deeply engaged in the field of IVI for many years, the downstream customers are of high quality, and the follow-up market share is expected to increase steadily;
2) Liquid crystal instruments, multi-screen systems, etc. currently have relatively low penetration rates in low-end models, and the follow-up penetration rate is expected to increase rapidly;
3) Considering that the company's manufacturing capacity and customer resources belong to the first echelon in China, and the advantages of superimposed localization services, it is expected that the market share in the fields of LCD instruments and multi-screen systems will increase rapidly;
4) The cockpit domain controller market is still in the early stage of development, and the future development space is large;
5) According to the information disclosed by the company, it and Qualcomm have reached a cooperation on the fourth generation of intelligent cockpit chips, and it is expected that the cockpit domain controller will become a new growth point for smart cockpits;
6) Suppose that in the company's smart cockpit business, the proportion of overseas customer revenue remains stable;
7) We expect the gross margin of the cockpit business to increase steadily.
Second, intelligent driving business
1) According to the information disclosed by the company, IPU02 was developed simultaneously with the TDA 4 chip, assuming mass production in 2022;
2) According to the information disclosed by the company, IPU04 has obtained a number of project designations, including ideal cars, and is currently in the process of research and development, assuming mass production in 2022;
3) IPU03 has begun mass production at present, and is expected to grow with the sales growth of Xiaopeng P7 and P5 in the future;
4) With the company's transformation to an intelligent enabler, it is expected that the sales of sensors such as cameras, millimeter wave radar, and ultrasonic radar will grow synchronously with the volume of domain controllers;
5) With the increase in the proportion of self-developed software algorithms and the upgrading of industrial roles, we expect the company's profit level in intelligent driving to show an upward trend;
3. Intelligent network connection service
1) It is expected that the intelligent network connection service will maintain medium and high-speed growth in the future;
2) Refer to the gross profit margin level of the company's intelligent network connection business in 2021, assuming that the subsequent gross profit margin remains stable.
4. Other business
1) It is assumed that the growth rate of other businesses is basically the same as the overall growth trend of the company
4.2. Investment Analysis
The company is the core target of automotive intelligence, the card slot advantage is obvious, and the follow-up growth logic is clear. It is optimistic about its expansion from intelligent cockpit to intelligent driving, and from automotive electronics suppliers to automotive intelligent enablers.
(This article is for informational purposes only and does not represent any of our investment advice.) For usage information, see the original report. ）
Featured report source: [Future Think Tank].