The problem of autonomous driving actually has only one issue.

Article | Financial Street Lao Li

“The enthusiasm of the autonomous driving industry has decreased recently. With the poor capital situation, even if there is money, it will not be invested in autonomous driving.” This is the opinion raised by a partner after investigating a well-known domestic autonomous driving company in the third quarter of this year. This viewpoint is not an exception, but represents the attitude of investors.

The capital market dislikes seeing situations such as IPO breaking, market value plummeting, bankruptcy, layoffs and internal struggles occurring frequently in the autonomous driving industry. Recently, Mobileye was listed at a discount, and then Argo AI announced its bankruptcy. Domestic star companies such as also experienced layoffs. According to Lao Li’s investigation, the cash flow of many domestic autonomous driving startups can only last for about six months.

In the article “Autonomous Driving, Can’t Help A-Dou?” link, Lao Li shared the development trend of autonomous driving from a macro perspective. Today, Lao Li will share with you more concretely from the perspectives of business and technology. Why must autonomous driving technology come before commercialization? What are the current issues with autonomous driving for bicycles? How much longer can vehicle-road collaboration fly?

Without technology, how to talk about commercialization?

So far, the only problem encountered by the autonomous driving industry is that the immature technology makes it difficult to mass-produce products, resulting in the inability to commercialize. Once the cash flow of a non-commercialized company is broken, the result is bankruptcy.

Generally speaking, entrepreneurial companies that rely on capital accumulation for scaling, such as those in commercial model innovation industries, can eventually stand out in the competition. Therefore, most internet companies are essentially driven by capital, and their moat is scale. As a result, there are usually no more than three giants in these fields.

However, the autonomous driving industry is different. Whether it is L3 or below level of assisted driving, or L4 or above level of high-end autonomous driving, the industry is a situation of hundreds of companies competing with each other. At least 10 companies can be found in each subdivision in the industry. For entrepreneurial companies that focus on technological innovation, there will be no revenue without mass production, and bankruptcy is only a matter of time.

The autonomous driving unicorn company Argo AI joint-supported by Volkswagen and Ford almost had zero revenue in its first three years. This year’s launch of Robotaxi was also unsuccessful, leading to the company’s bankruptcy following the cessation of shareholder funding.

Last year, Lao Li conducted due diligence on multiple self-driving companies, including and Momenta. According to incomplete statistics, the average revenue of the top ten autonomous driving companies’ main business in the industry is less than 300 million yuan. Even the smallest autonomous driving company’s annual expenditure far exceeds this average number. The main cost is personnel and equipment. The salary of an autonomous driving algorithm engineer easily reaches millions, with dozens of employees in each company, and the cost reaches tens of millions. The cost of a high-end autonomous driving system easily reaches hundreds of thousands of yuan, not to mention R&D costs.

Last year, Argo AI went to China to raise funds. When asked by the CEO of a domestic new energy leading OEM for the production schedule, the foreign answer was straightforward, “I want to know too…” The reason why kept switching tracks was also due to the inability to mass-produce. There were even differences in business positioning among the founders.

Currently, there are only two options for everyone: “switch to low-speed and commercial vehicle scenarios to develop high-end autonomous driving or continue to adhere to high-end autonomous driving for passenger cars.”

The recently popular mushroom-connected cars focus on building autonomous driving city public service fleets, including autonomous driving sanitation vehicles, patrol cars, connecting vehicles, logistics vehicles, buses, taxis, etc. However, from the capital perspective, these are not fully commercialized on the C-end because most of the buyers are governments and state-owned enterprises.

Compared with autonomous driving integrators, companies that manufacture hardware have a better situation. Although Mobileye went public with a discount, it always had a stable cash flow and carried the development direction of many autonomous driving chip companies, so the development situation of domestic chip companies such as Horizon Robotics and Black Sesame Technologies was far better than that of autonomous driving companies.

If a car is an autonomous and safe taxi, and the other car is driven by a human, which one would you choose? Most friends would choose the former. Many friends will find that the problem with autonomous driving is essentially not a commercial model problem but a technical problem. When it comes to this, we must talk about the continuous debate in the industry over the technical direction.## Autonomous driving for bicycles: slower, but more practical

There are two main technological routes when it comes to autonomous driving: one focuses on individual bicycles, represented by Tesla; the other focuses on vehicle-to-vehicle cooperation. The latter is the current approach in China.

Many Chinese car manufacturers and autonomous driving companies consider the individual bicycle model as the primary development direction. They believe that in terms of collaboration, both in low-speed and high-speed scenarios, it is easier to accomplish. Thus, companies such as Xiao Ma Zhixing, Momenta, and ZhiXingZhe are pursuing this route.

“We think that with the current level of cooperation, the individual bicycle route is faster than the vehicle-to-vehicle route. Although the cost of a single vehicle is high, most technical problems and supplier issues can be controlled by enterprises themselves. However, in the vehicle-to-vehicle scenario, multiple factors outside the vehicle must be considered,” said an expert at the Chang’an Automotive Research Institute during research conducted by Old Li.

“There is also a progressive route issue here. At present, suppliers are unable to provide complete L4 autonomous driving solutions. Also, regulations are not yet in place. Therefore, taking a progressive step-by-step approach, from L2 to L3, will be more secure from the perspective of technological implementation, commercialization of vehicle technology, and products,” suggested an expert from Volkswagen during a closed-door meeting.

Ignoring legal issues and focusing only on technology and cost, autonomous driving for bicycles also faces challenges in achieving commercialization.

A specialist at Ampofo is a loyal fan of Tesla and one of the leading figures in autonomous driving in China. He has worked for Xiaopeng Motors and Tesla, and believes “although Tesla’s pure vision may have its limits, its advantage is low cost. Low cost means high production. Relying on laser radar alone is expensive in China. Lasers often cost tens of thousands of yuan and cannot be mass-produced.”

Old Li believes that this point goes straight to the heart of the issue. If autonomous driving systems for individual bicycles cost more than ¥50,000 (USD $7,883), large-scale production is impossible. Currently, a single Velodyne lidar costs at least ¥100,000 (USD $15,765). According to Old Li, M’s solution used by Ji Ke costs around ¥30,000 (USD $4,729), while Huawei’s solution provided to Chang’an costs around ¥20,000 (USD $3,152). Essentially, these solutions are all L2-L3 advanced driving assistance systems and are far from achieving L4 autonomous driving.

For, cutting directly into L4 at the beginning was essentially a move to differentiate themselves from Huawei and Mobileye. In the L2-L3 field, where the industry’s top players gather, there are typical foreign automotive suppliers, A – Autoliv (Veoneer), B – Bosch, C – Continental, D – Delphi (Ampo), as well as technology companies Mobileye and Huawei. The competition in this industry is based on technological foundation and scale capabilities, and companies like have no advantages in this field.

Ultimately, for companies like WeRide,, Argo AI, their advantages are definitely in L4 and above, taking a differentiated route. However, the problem is that it is difficult to implement and many “experts” also pointed out that it may prove that autonomous driving of a single vehicle is a dead end.

From Mr. Li’s point of view, the more mature level of single-vehicle autonomous driving at present is L2-L3. In scenarios above L3, both perception and decision-making layers face problems of technological immaturity and high cost. Many friends say that time can solve these problems, but the problem is that five years ago, everyone said it could be solved, three years ago, they also said it could be solved, and now investors no longer believe it.

V2X, Let the Bullet Fly a Little Longer

In the early stage of the development of autonomous driving for single vehicles, some people in the industry proposed to rely on V2X to carry out high-level autonomous driving. Compared with single-vehicle intelligent autonomous driving, V2X has the core advantage of global perception, which can avoid safety problems caused by sensor failures, obstruction, and severe weather that single-vehicle intelligence may encounter.

What is global perception? To put it simply, it is to conduct autonomous driving from a “God’s eye view”. Vehicles and road infrastructure are equipped with sensors and “God’s hand” is used to drive the vehicle to achieve autonomous driving. Companies such as Alibaba, Baidu, Mushroom Car, and HoloMatic are representative enterprises of V2X.In roadside scenarios, vehicle-road collaboration can accurately analyze and optimize the positioning and path of vehicles on roads, improve traffic efficiency, support traffic accident tracing and analysis, and play a positive role in improving the efficiency and precision of traffic management. Therefore, the transportation and communication industry has been promoting the development of vehicle-road collaboration.

As Changan Automobile experts mentioned, vehicle-road collaboration means more collaborative entities. From the perspective of demonstration areas, China has formed a relatively complete system of roles and responsibilities among “government, industry, and enterprises.” The embryonic form of industrial development has emerged. It should be said that this is a comprehensive mobilization of “cross-industry collaboration and multi-level cooperation” industrial battle. Currently, numerous intelligent connected vehicle testing demonstration zones have been built in China, and the vast majority are vehicle-road collaboration demonstrations. However, the problem is that it is difficult to complete such a large-scale infrastructure construction on a 9.6 million square kilometer land. Even if only national highways are built, it requires a huge investment.

Although the opinions and guidance of various ministries and commissions have been issued to the industrial development planning and layout of various provinces and cities, and various cities have also formulated and promulgated road test management methods. Though more than 10 automotive companies have released or planned to release C-V2X production models, and more than 1,000 test licenses have been issued, and more than 3,000 kilometers of roads have been open to testing. In reality, vehicle-road collaboration faces more problems than single vehicle automated driving.

First, collaboration issues. Vehicle-road collaboration automated driving involves cross-fusion in the fields of automobiles, information and communication, electronics and electrical engineering, artificial intelligence, and control theory. Local governments can hardly establish coordination and communication mechanisms for responsible entities, and it is difficult for enterprises to allocate investment and benefits. Simply put, who will pay for such a large infrastructure investment?

Secondly, vehicle-road collaboration automated driving has strong big data characteristics. Huge basic traffic operation data face great challenges in storage, processing, and security protection. Currently, in local test areas, many vehicles must have strong “server” support. These hidden costs are not paid by users, and users will not pay for security. It can be said that, in terms of the current technology level, vehicle-road collaboration is even less mature than single vehicle automated driving.

Moreover, the current cooperative driving and autonomous driving functions are still in the early commercialization stage, and there is no standardized evaluation and testing system for these functions. In terms of hardware aspect, the functionality of vehicles and road-side facilities needs improvement. Due to the strong combination of vehicle collaboration and infrastructure, there is great potential for development.

What is the ultimate endgame for the autonomous driving industry? No one can say for sure. How should autonomous driving companies commercialize their products? It’s also hard to say. The industry has been debating over autonomous driving issues for years, but decisions are largely based on individual preferences.

Li believes that the ideas in the capital market may be simpler and more practical –

We do not exclude any technology. When L3 and below technologies can be mass-produced, we will pursue companies in these areas. When L4 and above technologies achieve breakthroughs, we will pursue companies in these areas. If both have potential, we will invest in both. If neither has potential, we will leave.

This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email