As the export volume of new energy vehicles (NEVs) from mainland China increases, the pace of overseas expansion by related automakers is also accelerating. In this context, the Hong Kong Special Administrative Region (SAR) is becoming a key area for these automakers to establish a foothold, serving as the so-called "going global" starting point.

Looking at actual sales, Hong Kong, with a population of over 7 million, primarily relies on buses or the Mass Transit Railway (MTR) for transportation. The acceptance of NEVs is yet to become widespread, and there is a limit to sales volume. Therefore, developing the local market is not the core demand of the companies involved.

Advertisement

The trend of crossing the river has been formed, with mainland automakers and Hong Kong moving towards each other in what can be described as a "two-way rush." The latter's enthusiasm stems from an inherent desire for economic diversification. To promote innovative technology, subsidies and assistance in securing cornerstone investments that could reach hundreds of millions of Hong Kong dollars, coupled with the prospect of "landing first and going public later," form the basic appeal to automakers.

The attempt to accelerate the "welding" of technological research and development advantages with the industrial chain is yet to be seen in terms of efficiency. Relevant data indicates that for many years, the proportion of local R&D investment has been lower than that of many core cities within the Guangdong-Hong Kong-Macao Greater Bay Area. After decades of "deindustrialization," Hong Kong does not have many advantages or accumulations in related industrial chains. The technological innovation that relies on it is also more challenging to carry out.

Under these circumstances, mainland NEV manufacturers (including original equipment manufacturers and supporting enterprises) are flocking to Hong Kong to establish research and development centers, data centers, and even international headquarters. The rationale is to leverage its R&D and talent advantages, with approaches including the establishment of financial management centers and becoming leading enterprises to bring upstream and downstream partners to Hong Kong. The specific operations and micro-industry logic followed are not widely disclosed.

On August 9, at an industry salon held in the Hong Kong SAR Science Park, several individuals from the upstream and downstream of the new energy vehicle industry chain discussed the logic of this wave of going global from multiple perspectives of mainland China and Hong Kong. This may help to eliminate some uncertainties related to the issue, such as what are the future breakthroughs and challenges for NEVs? Why is it necessary for companies to go global? What are the attractions of Hong Kong? And what other work can be done?

Demand Side: Why Cars Are No Longer Just Cars?

In the global automotive industry development map, China's new energy vehicles have emerged as a strong contender. Taking 2023 data as an example: China's automobile production and sales broke through 30 million vehicles for the first time, maintaining the world's number one market for 15 consecutive years. Among them, the production and sales of new energy vehicles have been ranked first globally for nine consecutive years.

After more than 20 years of exploration and development, as of April this year, the retail share of new energy vehicles in mainland China has historically surpassed that of fuel vehicles for the first time. With continuous competition and cooperation among new forces and multiple automakers in technology, supply chain, and market aspects, the industry as a whole is constantly innovating and accumulating technological advantages.Upon close examination of the so-called technological advantages, a key area is the convergence of new energy vehicles with intelligent technologies, transforming into smart cars. This signifies that they are becoming carriers of the Internet of Things (IoT), interconnecting the automotive internet, energy networks, and intelligent transportation networks.

The fundamental idea behind this interconnection is that the integration of intelligent technology with new energy vehicles becomes more profound, offering users a more immersive experience. According to Dai Dali, CTO of NIO (Chief Technology Officer): "The industry has reached a critical period for the emergence of new species and new industries." As cars transition from traditional electromechanical support and integration to new energy, the acceleration of intelligence and internet capabilities leads to significant changes in the connotation and extension of both the automotive and the industry.

Expanding from the traditional manufacturing industry within the field of mechanical and electronic manufacturing, a plethora of new concepts related to comfort, intelligence, and safety emerges continuously. However, at the core, they mainly involve two directions:

First, the enhancement of traditional technologies and models. For instance, how to achieve longer endurance for new energy sources (batteries), minimizing range anxiety while meeting user demands for comfort, safety, and environmental friendliness. The main breakthroughs in this field are, on one hand, how to accelerate the application of new technologies such as solid-state batteries, characterized by higher efficiency, safety, and energy density; on the other hand, it is about continuously improving technological levels in areas such as high-voltage fast charging and power systems.

Second, the so-called "intelligence" technology, which is believed to have the potential to change the nature of cars and promote the restructure of the automotive industry. Due to the possibility of "disruptive innovation," this has become a focal point of contention among manufacturers. It is also in this area that "car manufacturing," traditionally a part of the manufacturing industry, begins to increasingly acquire technological attributes.

Breaking down the logic of the integration of intelligence and automobiles, it can be divided into different domains, such as intelligent interaction and autonomous driving. However, for users, it can be unified into the perceived quality of intelligent experiences. To enhance the level of perception, manufacturers are collaborating with partners to upgrade automotive hardware and software from a more comfortable, safe, and intelligent perspective.

The upgrading process often involves the collaboration and progress of both hardware and software. Taking the line-controlled vehicle chassis as an example, to achieve greater safety and comfort, a combination of hardware and software is required. Algorithms provide more experiences and functions, while hardware applies integrated assembly and upgraded manufacturing techniques to ensure a better foundation for support.

Other areas that involve the combination of hardware and software and are more perceptible to users are intelligent interaction and autonomous driving. To achieve human-machine interaction and integration of people and vehicles, the common approach is also through higher computing power and better algorithms to integrate ecosystems.

To achieve ecosystem integration and technological interconnection, one of the core focuses of various efforts in the automotive industry is the deployment of AI large models. For instance, to achieve autonomous driving, a current hot topic in the industry is the so-called "end-to-end solution."

Prior to this, the technological evolution of autonomous driving has been relatively fast. The mainstream solution in 2023 was still light high-precision map urban driving, but by 2024, it evolved into the so-called "end-to-end," which integrates perception, planning, and control into a single module.This actually borrows a concept from the field of deep learning, "end-to-end," which in English is "End-to-End (E2E)." It refers to an AI model that can output the final result as long as the raw data is inputted. In the field of autonomous driving, it means that only one model is needed to convert the perceptual information collected by sensors into specific operational commands, allowing the car to drive automatically.

So far, there is no exact answer in the domestic industry about how Tesla's end-to-end autonomous driving solution is implemented (there might be, but it has not been disclosed). However, this does not hinder it from becoming a new breakthrough for various companies trying to build competitive advantages.

According to Wang Xiaogang, co-founder of SenseTime in Hong Kong and President of the Jueying Intelligent Automobile Business Group, "When the end-to-end autonomous driving solution was first proposed in the industry in 2022, most people did not believe it. But this year, Tesla's FSD (Full Self-Driving) entered China, triggering anxiety and demand for end-to-end solutions."

Before this, the mainstream solution in the autonomous driving industry was actually based on rules. To solve practical problems, a large number of algorithm engineers are needed to formulate relevant rules. However, with the acceleration of large AI models into the practical field, it is claimed that the intervention of the model can achieve functions that used to require very complex rules, which can improve R&D efficiency.

Understanding the technological upgrade on this basis also includes many different aspects. For example, rule-based autonomous driving requires a lot of modules to implement functions, but only some modules such as the perception module are based on neural networks, and other modules need to formulate rules. After being modeled, it unifies all modules, and by inputting video, it can output driving planning trajectories, achieving basic functions such as pedestrian and vehicle avoidance.

Being able to rely solely on pure vision (without relying on LiDAR or high-precision maps), and inputting images and videos to output trajectories, is the foundation of the true end-to-end that Professor Wang Xiaogang refers to. In his view, only in this way can a higher ceiling be built, and the understanding of complex scenarios is more comprehensive.

This is actually only half of his vision for the future of autonomous driving, "brain plus cerebellum." The reason for this judgment can be understood as: end-to-end autonomous driving is similar to the human cerebellum, which can directly output actions after receiving various signals. On this basis, multimodal large models will play a more diverse role, systematically analyzing, judging, and making decisions for various complex scenarios.

"It can recognize taxis, ambulances, know who to give way to first, and make better decisions. For situations that have not been defined in advance, such as the road to the right being closed, and people's postures and gestures, it can make automatic judgments," said Wang Xiaogang.

Although the implementation path is different, to meet similar related functions, it has put very high requirements on the infrastructure of automotive companies. Taking Tesla as an example, to support the iteration of its end-to-end autonomous driving, the basis for the transition from rule-driven to data-driven is a large amount of high-quality data, as well as support from infrastructure with ultra-large computing power.

At the first quarter financial report meeting in 2024, Tesla has revealed to the outside world: it has expanded the training AI cluster to 35,000 H100 GPUs. According to the plan, by the end of 2024, Tesla will also invest an additional $1.5 billion in the supercomputing cluster, with the goal of increasing the total computing power of its supercomputing center to 100 exaFLOPS (a computing power indicator, equivalent to 100,000 petaFLOPS).Supply Side: What Can Hong Kong Do?

As an industry benchmark, Tesla's process of improving computational power, also known as the "great strength creates miracles" or "brute force computing" model, stems from the fact that higher computational power leads to corresponding improvements in model iteration efficiency, iteration methods, and problem-solving efficiency. Simply put, high computational power brings a larger capacity, allowing for more attempts and trial-and-error.

However, the reason why the intelligent development of new energy vehicle (NEV) companies is still in its infancy is that mainland car manufacturers believe there are other ways to break through.

In the short term, the current goal of the relevant car companies is to solve the mass production and landing problem of urban NOA (Navigation on Autopilot) and to improve the level of autonomous driving. In the medium to long term, the improvement of computational power can be achieved through self-building or cooperation. In fact, in terms of intelligent computing centers, Geely, Changan, and the "Weilai, Xiaopeng, and NIO" are all preparing, either by self-building or by cooperating with third parties. For intelligent driving suppliers, having an intelligent computing center has become a standard for "entering the industry."

Taking the aforementioned SenseTime's Jueying intelligent driving as an example, according to the information revealed in its latest financial report: the number of GPUs in its intelligent computing center has reached 45,000, with an overall computational power scale of 12,000 petaFLOPS. Compared to the beginning of 2023, this number has doubled.

Looking at the absolute value of computational power, domestic companies still have a certain or even significant gap compared to their foreign counterparts. And due to factors such as chips, it is difficult to catch up in the short term. However, in addition to the aforementioned "great strength creates miracles" model, companies are also willing to believe that for new energy vehicles to achieve intelligent features such as autonomous driving, computational power is only one aspect, and data and algorithms complement each other.

This is also the key area where intelligent driving suppliers are deeply cooperating with vehicle manufacturers. Admitting that Tesla and others have extremely strong infrastructure, one source of product differentiation is the cleaning, combination, and high-quality screening of data. In the view of suppliers, the deeper the cooperation (infrastructure, platform to data), and the more it breaks away from the standardized product output model, the more likely it is to produce breakthroughs, and the more likely it is to keep up with or even lead the pace of automotive technology iteration.

The constraints and breakthrough models currently faced by the development of new energy vehicles are quite representative. Taking the example of a mainland research institution reported by "Caijing" magazine, which set up a research and development center in Hong Kong, it also encountered the problem of computational power constraints when developing a multimodal large model for the medical industry. In addition to actively cooperating with domestic computational power suppliers, it is willing to carry out research and development work in Hong Kong, valuing Hong Kong's unique advantages in basic research, international exchange of personnel and information, and the connection of industry rules and legal systems between mainland and the world.

When a field is slightly weak, consciously strengthening other aspects may bring certain additional quantity or even breakthroughs. To achieve this, a series of related talents are naturally an indispensable competitive advantage. In fact, according to Huo Jian, General Manager of Alibaba Cloud Intelligence Group's automotive energy industry solution, there are three key elements for car companies to develop autonomous driving research and development: the third is the cloud computing partner that provides computational power support, the second is capital, and the first is the talent echelon.

"New energy vehicles belong to a typical capital-intensive, technology-intensive, and talent-intensive industry. Hong Kong (its industry and talent soil) meets the relevant requirements." Wang Ruoyu, co-founder and vice president of TianTong WeiShi, described, "Hong Kong gathers talents from all over the world, and local universities can also cultivate and output talents, which is crucial for the high-tech industry."In terms of talent cultivation and technology utilization, industry insiders have previously expressed to Caijing a view that, on the surface, Hong Kong has not had any significant breakthrough products in recent years, and talents often have to move north to start businesses due to lack of funding and policy support. However, Hong Kong's R&D is actually more international and relatively more foundational. This implies that if R&D resources can be well integrated with industrial chain scenarios, the potential is enormous.

Specifically, in the field of new energy vehicles, the areas that currently need to be focused on breaking through include complex environment perception, intelligent decision-making control, human-machine interaction, and human-vehicle co-driving. In the related technologies such as new electronic and electrical architecture, multi-sensor technology integration perception, intelligent computing platforms, new types of smart terminals, and smart car topological maps, Hong Kong is also considered to have a certain basic scientific research leadership advantage.

This may also be the common logic behind a series of new energy vehicle companies from the mainland building global innovation platform strategic fulcrums in Hong Kong. For companies, by establishing R&D centers in Hong Kong as a "bridgehead," they can not only recruit and cultivate global talents by leveraging Hong Kong's international advantages but also cultivate talents through cooperation with local universities and research institutes, thereby promoting technological and industrial progress. As the founder and chairman of NIO, William Li, said to Caijing, "Why come to Hong Kong? We want to rely on the motherland, be based in Hong Kong, and face the world. We must make full use of Hong Kong's unique advantages and its open and inclusive cultural atmosphere."

For Hong Kong, as the starting point for mainland enterprises' global business over the past decades, in addition to traditional strengths such as financial and legal norms, to seize this new opportunity, it has also begun to rebuild and complement the chain through a clear industrial leadership approach.

When sorting out the basic ideas of this new industrial blueprint, Zhang Manli, the Deputy Director of the Innovation and Technology Bureau of the Hong Kong Special Administrative Region Government, said, "Hong Kong must not only consolidate its upstream basic scientific research but also make up for past shortcomings, strengthen the transformation of achievements in the middle stream, and the industrialization and commercialization in the downstream. It is necessary to build a new type of economic entity by constructing a technology industry system in Hong Kong."