Zhongguo Lianhe Zaogu Holdings Limited released a research report stating that end-to-end large models are the main way to achieve advanced intelligent driving functions. The end-to-end neural network can fully simplify the computation steps, reduce the need for manual feature engineering, and identify the correlations in the data to fully improve computing efficiency. With the increase of effective behavioral trajectory data, end-to-end intelligent driving large models are expected to become a solution for advanced intelligent driving.

Guolian Securities' main points are as follows:

How to evaluate the ability of intelligent driving end-to-end large models?

With the development of large models, end-to-end intelligent driving large models have an emerging effect. Guolian Securities believes that the main parameters for evaluating the model are the scale of vehicle trajectory data, training data capability, and software development ability. The scale of vehicle trajectory data includes accumulating sales and accumulated mileage of centralized domain architecture and models with larger vehicle computing power. Training model capability mainly includes the computing power of the intelligent calculation center, cloud training capability, and data storage capability. The computing power of training is the key to computation speed. The cloud architecture optimizes computing power layout, and the data storage capability determines the trainable vehicle trajectory data scale. The software development ability is difficult to evaluate due to various code not being open sourced. The R&D cost is used to replace the software development ability from two dimensions, namely the scale of vehicle trajectory data and the training data capability.

How is the progress of intelligent driving end-to-end large models of current car companies?

Analysis of the progress of end-to-end large models of intelligent driving of various car companies at the current point in time from two dimensions: the whole vehicle and the training end. The architecture and vehicle sales volume are the main concerns for the vehicle end. E/E architecture: Tesla and new forces are leading, and self-owned brands are accelerating to follow up. After the domain controller-style architecture, the sales volume and cumulative mileage of the models are positively correlated, and Li Auto is in the lead, and Huawei and Xpeng are rapidly catching up, and self-owned brands are expected to accelerate in the future. The training end mainly focuses on the deployment of computing power and the level of cloudization. The deployment of computing power: Tesla is in the lead, Huawei's construction is accelerating, and domestic car companies and Internet companies are establishing partnership to accelerate the deployment of computing power. Cloud computing capabilities: Huawei Cloud, Aliyun, Tencent Cloud, and Baidu Intelligent Cloud are accelerating the integration of their market shares.

Current situation: Tesla is leading, and domestic brands are accelerating.

Guolian Securities stated that from the current situation, Tesla is expected to continue to lead with its advantage in computing power and data scale, and Huawei and Li Auto have higher R&D expenses and larger data scale, which enables them to catch up quickly. Nio and Xpeng may be affected by data volume and other factors that reduce iteration efficiency. Among other self-owned brands, due to the small number of centralized E/E architecture models, the number of data accumulation models cannot be achieved in the short term and computing power is in the construction stage.

Investment advice: focus on leading end-to-end whole vehicle manufacturers and intelligent/domain architecture providers.

Whole vehicle end: Huawei-related car companies are expected to fully benefit from Huawei's leading computing power to achieve a function reversal. Self-developed Li Auto and Xpeng are expected to take the lead in realizing the leading position of large models with the end-to-end landing rhythm. Parts end: focus on the direction of E/E architecture upgrade and intelligent driving vehicle computing power. Key recommended high-speed connector supplier Electric Connector Technology (300679.SZ); wire-controlled chassis domain supplier Bethel Automotive Safety Systems (603596.SH), Ningbo Tuopu Group (601689.SH), Shanghai Baolong Automotive Corporation (603197.SH), Anhui Zhongding Sealing Parts (000887.SZ); domain controller core suppliers Jingwei Hirun (688326.SH); end-to-end intelligent driving domain controller core suppliers Huizhou Desay SV Automotive (002920.SZ), Keboda Technology (603786.SH), and Foryou Corporation (002906.SZ).

Risk warning: Intelligent end of vehicle progress is lower than expected; computing power construction is lower than expected.