The smart car track is both imaginative and realistic.
The imagination lies in that the intelligentization of automobiles, especially autonomous driving, is one of the industries with the highest threshold for AI (artificial intelligence) to land. Apple CEO Cook once commented, "Autonomous driving is the cradle of all artificial intelligence (AI) projects, and it is also one of the most difficult AI projects we are working on."
However, the reality is that there are still many difficulties to be overcome before the real commercialization and large-scale launch of L4 autonomous driving. However, intelligent assisted driving for pre-installed mass production has already become a red ocean market.
However, judging from the many industries where AI is applied, car intelligence is still a track that cannot be ignored.
Megvii Technology, which was born on the AI track (once known as one of China's four AI tigers), is also returning to the smart car industry. A few years ago, Megvii entered the cockpit AI visual solution track (cooperated with NIO on the in-cabin AI visual interaction project), but has since disappeared.
Interestingly, in Megvii’s previously disclosed prospectus, apart from some awards in top autonomous driving competitions, there is almost no discussion about business related to car intelligence.
Megvii's road to listing is also quite bumpy.
From submitting the prospectus to the Hong Kong Stock Exchange in 2019, to moving to the Science and Technology Innovation Board in 2021, in March 2022 and March 2023, the issuance registration process was suspended twice. On July 15, Megvii restarted the IPO registration process again.
Previously, relevant regulatory agencies had raised doubts about Megvii's business prospects: including unstable customers, low industry concentration, core technology competitiveness and future development prospects.
Recently, Liu Wei, president of Megvii's smart driving business, revealed that the company has now realized a four-in-one model of "sensing, mapping, tracking, and forecasting", focusing on cost-effective L2+ mass production solutions. At the same time, it is also accelerating the deployment of hardware such as smart driving chips and sensors.
However, the company's in-cabin AI vision project was also reported to be in a state of pause and adjustment. For many AI companies, the extension of face recognition technology in the automotive industry was originally logical.
At that time, Chengdu Megvii Artificial Intelligence Research Institute led the research and development of in-vehicle AI vision products and solutions, including gesture recognition, expression recognition, and gaze tracking in multi-modal human-computer interaction systems; to meet the needs of smart and safe driving and operating vehicle supervision face recognition, driver status detection, front collision and lane departure warning, etc.
Judging from the actual terminal mass production data in recent years, the performance of the cockpit visual AI interactive track is acceptable. However, most companies are still unable to achieve profitability, and some companies even show signs of performance decline.
According to the monitoring data of Gaogong Intelligent Vehicle Research Institute, from January to June 2023 in the Chinese market (excluding imports and exports), 1,327,300 passenger cars equipped with IMS systems (visual perception interaction) as standard equipment, a year-on-year increase of 76.62%; among them, 893,000 DMS pre-installed standard equipment, a year-on-year increase of 45.37%.
However, the market competition has become fierce. Some software and hardware integration solution suppliers even offer a price for hardware fee + free software. Companies that originally focused on pure software solutions are all integrating software and hardware, but their profitability is not good.
For example, the gross profit margin of software is relatively high, and there are many synergies between the underlying technology and the application of AI in other industries. However, the "ceiling" effect that exists naturally in the automotive industry makes it difficult to scale up the revenue from a single software module.
Once involved in hardware, it means that the gross profit margin is rapidly compressed. But this is the only way out. "What we have to do now is to firmly invest in the technology research and development of large models, and at the same time do a good job in AI hardware carriers and build high barriers." Yin Qi, co-founder and CEO of Megvii Technology, said frankly.
Previously, it was also discussed in the Prospectus of Megvii Technology: Consumer IoT solutions are mainly delivered in the form of cloud services and software, so the gross profit margin is relatively high; while urban IoT and supply chain IoT provide a full-stack solution integrating software and hardware. Solution, hardware costs account for a relatively high proportion of the cost structure, so the gross profit margin is relatively low.
Since 2018, Megvii's overall business gross profit margin has continued to decline from the highest 62.23% to around 30%. In addition to hardware costs lowering the gross profit margin, fierce competition in the industry is also a major factor.
For example, take ArcSoft Technology, which has the largest amount of pre-installed DMS, as an example. The company wrote in this year's semi-annual report: being challenged by other competitors in the industry, it is facing the decline in gross profit margins caused by intensified market competition and the Risk of declining market share.
According to data, in the first half of this year, ArcSoft's smart driving and other IoT smart device vision solutions achieved operating income of 22.4466 million yuan (including software revenue of 16.72 million yuan), a year-on-year decrease of 15.01%. Among them, the in-cabin visual perception interaction is still the main source of revenue contribution.
In addition, Israel-listed Cipia Vision (customers include FAW Group, SAIC Motor, Ford, etc.) will have revenue of $5.3 million in 2022, an 11% increase over 2021. However, corporate margins fell to 21.9% from 39.6%. One of the reasons is that the gross profit margin has dropped sharply due to the software-hardware integrated delivery solution, and last year it lost 13 million US dollars.
The financial situation of Smart Eye, which has received orders from many car companies such as BMW, Audi, Polestar, and GM, is also not optimistic. The company will achieve global shipments of more than 1 million sets in 2022, and double its revenue year-on-year, but its losses will also increase significantly.
This time, Megvii’s autonomous driving business is positioned to create affordable high-end smart driving solutions for mainstream models. "Compared to peers, Megvii can further reduce costs by 20%-30%"
However, the cost reduction strategy given by Megvii is nothing more than a clear trend in the industry: strong vision, de-high-precision maps, and de-RTK. At the same time, step-by-step solutions are given for different price ranges.
The good news is that from January to June 2023 in the Chinese market (excluding imports and exports), the share of standard L2 (including L2+) equipped with Chinese local system solutions in the Chinese market (excluding imports and exports) has risen to 14.08%. In addition, in high-end tracks such as NOA, local Chinese forces have already gained the upper hand.
For example, the Hyper GT, the first mass-produced model of GAC Aian's high-end new energy brand Haobo, comes with a top-of-the-line solution from Heduo Technology.
According to the monitoring data of Gaogong Intelligent Automobile Research Institute, from January to June 2023, 3.2435 million new passenger cars equipped with L2 (including L2+) equipped with pre-installed standard equipment in the Chinese market (excluding imports and exports) will be delivered, a year-on-year increase of 37.65%; The standard configuration rate was 34.90%, an increase of about 8 percentage points year-on-year.
However, competition is also fierce.
Taking Nezha as an example, besides SenseTime, MINIEYE, Freetech and many other ADAS partners. At the same time, there is also a TA PILOT intelligent driving system with a full-stack self-developed system, and TA PILOT 5.0, which is under development, plans to realize the full-stack self-developed software and hardware of the system.
This means that in addition to providing solutions on perception algorithms and some hardware, it is difficult for external third-party collaboration suppliers to obtain greater benefits from them. A similar situation also appeared in GAC.
As we all know, GAC has invested in Heduo Technology (high-end smart driving) before, and has in-depth mass production cooperation with Zhihua Technology (surround view, parking), Horizon (chip, cockpit interaction), and smart driving technology (L2/L2+).
At the same time, the X Lab team of GAC Research Institute, established in 2022, is also deeply cultivating the self-developed full-stack self-developed algorithm for autonomous driving based on pure visual routes. At present, various technologies have been tested in the parking and driving fields before mass production.
In this regard, Liu Wei’s judgment is that OEMs’ cognition and thinking on full-stack self-development are changing, and the core appeal is shifting from full-stack self-development to full-stack controllability.
However, this is more of a "placebo" shot for investors from the perspective of suppliers. Beginning this year, some leading car companies have begun to license core technologies to the outside world. For example, the cooperation between Xiaopeng and Zhiji with Volkswagen and Audi, Tesla's public permission to authorize FSD, and so on.
At the same time, even if a car company only achieves full-stack control, its bargaining power over suppliers will also be greatly improved, and the latter's profit margins will also be greatly reduced.
At the hardware level, traditional foundries (originally hardware partners of many start-up companies) also hope to seek breakthroughs in the smart car track. For example, Luxshare Precision has established a subsidiary (Lisheng Intelligent) to lay out intelligent driving solutions.
In addition, L4 autonomous driving companies such as Pony.ai, WeRide, and Qingzhou Zhihang have entered the pre-installation market in turn to compete for high-end smart driving orders. From traditional Tier1, ADAS solution providers, autonomous driving companies to different cross-border players, smart The pre-installed track has been completely reduced to the "Red Sea".
In particular, entering 2023, the smart car track has been driven from pure technology and has truly entered a new round of market-driven cycle. Next, for suppliers and car companies, the fight is not just about the "stacking" of flagship high-end models, but cost-effectiveness.
Behind the three words "price ratio" is the price war.
In the view of Gaogong Intelligent Automobile Research Institute, under the background that leading car companies continue to strengthen their self-research capabilities, in the past, AI start-ups placed their hopes on a model based on software algorithms, and the opportunity to become bigger and stronger is gradually being weakened ; And it is not easy to enter the software and hardware integration track from software.
Previously, a research report from Jingwei Hengrun disclosed that the company judged that the software for L1/L2 function suites is free, and once the software is made, it will be free; currently L3 is not free, but once the software is made Later, it will also be free.
Now automakers are working on algorithms for autonomous driving, just to understand the konw-how in it, and turn the demands into product demands. The premium of software will tend to be zero, and suppliers must make hardware if they want to grow bigger.
This point can also be seen from the strategic transformation of ArcSoft Technology. The company also provided a single software algorithm solution in the early stage, and began to lay out the integrated software and hardware vehicle vision solution.
At present, both software and hardware manufacturers are developing towards the integrated software and hardware model; on the one hand, avoiding the risks brought by price wars and self-research by car companies, and making two-handed preparations; on the other hand, a single software or hardware model , it is difficult to form a synergistic effect on cost reduction and efficiency increase.
However, under such a background, coupled with the "low price" order grabbing, the risk of sustainable growth of enterprises is also highlighted. Especially for AI companies fighting on multiple fronts, there are many difficulties.