Text: Liu Junhong
Edited by Wang Yisu
Following WeRide, another L4 driverless company is also preparing to go public.
On October 18, Pony.ai, a driverless service provider, filed a prospectus with the United States SEC (Securities and Exchange Commission). Following Baidu, Waymo, and Tesla, according to the prospectus's plan for the use of funds, Pony.ai has become another player that is ready to expand and operate Robotaxi in October this year.
According to the prospectus, in 2022, 2023 and the first half of 2024, Pony.ai's revenue will be US$68.39 million, US$71.9 million and US$24.72 million respectively, with a cumulative revenue of more than US$165 million (about 1.2 billion yuan), and the overall revenue volume is not large.
It's worth noting that Pony.ai's revenue in the first half of 2023 is only about one-seventh of the full-year, which is very different from the roughly one-half ratio of most companies. In the first half of 2024, the year-on-year growth rate of revenue is close to 100%, although the growth rate is fast, it is difficult to predict the revenue for the whole year.
During the same period, Pony.ai's net losses were US$148 million, US$125 million and US$51.78 million, respectively, and the overall loss showed a narrowing trend. In the first half of this year, the net loss fell sharply by 25.6% year-on-year, and the company carried out relatively strict cost control.
Like many L4 driverless companies, Pony.ai is still in the early stages of commercialization. At present, the autonomous driving fleet, which is the main source of revenue, is still too small, with less than 500 Robotaxi (250+) and unmanned trucks (190+) combined. The application of the passenger and freight industry is still in the state of phased pilot.
At a time when L4 autonomous driving companies are turning to L2 and seeking faster profits, Pony.ai is one of the few companies that focuses on L4 business.
At present, although the L4 business is still in the early stages, Pony.ai is already working hard to get the company through the long years before the outbreak of Robotaxi through businesses such as unmanned trucks and technology licensing, and support the company's long-term R&D investment.
The good news is that judging from the operations of Pony.ai, Baidu, and Waymo, the Robotaxi industry has basically moved from technical verification to the stage of seeking scale effects.
"When Pony started in 2017, I told the team that it might take another eight years for the industry to become profitable, which happens to be 2025." Pony.ai CTO Lou Tiancheng said in an interview.
Under the sudden outbreak of Robotaxi in 2024, Pony.ai has finally waited for its own "spring".
Commercialization model switching, from technology to operation
Pony.ai has quietly transformed from an unmanned technology service provider to an operator.
Behind this is a fundamental shift in Pony.ai's business model.
In terms of overall revenue, although Pony.ai has generally shown a growth trend in the past two years, the company's revenue quality has continued to decline. According to the prospectus, from 2022 to the first half of 2024, Pony.ai's gross profit margin has declined sharply, from 46.9% in 2022 to 10.5% in the first half of 2024.
The decline in gross profit margin is that Pony.ai's revenue model has become "more and more bitter". Previously, Pony.ai mainly relied on "selling software services", but now, its revenue mainly comes from the operation of driverless fleets dominated by trucks.
Pony.ai's main revenue sources are Robotaxi (unmanned taxi), Robotruck (unmanned truck) and technology licensing. In the past two and a half years, Pony.ai's revenue structure has changed significantly.
Among them, the proportion of revenue from the technology licensing business (generally speaking, with a higher gross profit margin) decreased from 54.2% in 2022 to 22.3% in the first half of 2024, while the Robotruck business rose from 32.7% to 73.0%, becoming the absolute focus of revenue. As for the Robotaxi part that is currently the most anticipated in the industry, it has not yet been launched.
The changes in revenue and gross profit margin are related to the characteristics of these three businesses.
In terms of truck business, due to Pony.ai's official entry into Robotruck in 2021 and cooperation with Sinotrans in 2022, Qinghu Logistics was established in the form of a controlling shareholder. The establishment of the joint venture has allowed the deployment expenses and revenue of 160+ of the 190+ Robotrucks to be recorded in Pony.ai's accounts, resulting in a sharp increase in the proportion of business revenue and costs.
Secondly, in terms of technology licensing revenue, since Pony.ai only gave a two-year data comparison, it can still be seen that the revenue of this business has been growing. The decline in gross profit margin in 2023 is mainly due to the company's increased demand for related materials to cooperate with the project. The superposition of the above two has led to a large change in Pony.ai's gross profit margin and revenue proportion.
Finally, there is Robotaxi, mainly because the industry is too early. Pony.ai's Robotaxi fleet only has a total of 250+ vehicles, which is really difficult to generate revenue at scale.
However, the good news is that based on the data for the first half of 2024, assuming that Pony.ai's Robotaxi fleet has 250 vehicles, the corresponding daily revenue per bicycle is about $26. Compared with Baidu's Radish Run, Pony.ai's Robotaxi is basically "tied" with the average daily order volume of 15 orders, but the average customer unit price is more than twice that of Radish Express (including subsidies, the average customer unit price is 5 yuan).
In other words, Pony.ai's Robotaxi commercial operation experiment was not only successful, but also better than the "out of the circle" radish run.
In summary, Pony.ai's current operation is relatively smooth, and there is no perception that everyone thought that L4 companies were operating particularly badly in previous years. The company's Robotruck business, which the company focused on in the last stage, achieved rapid growth, the revenue of the technology licensing business grew steadily, and the operation of Robotaxi also achieved relatively good commercialization results.
In addition, under the premise of continuous business expansion, Pony.ai also narrowed its expenses in advance.
In comparison between 2023 and the first half of 2024, Pony.ai's total operating expenses have been lowered from US$78 million to US$74 million, and the overall investment strategy is relatively conservative. Based on the cash reserves in the 2023 annual report, the company's capital volume can ensure safe operation for about 3 years. In other words, if 2024 "didn't wait" for the Robotaxi outlet, Pony.ai is also ready for a certain amount of long-term preparations.
According to the prospectus's plan for the use of funds, Pony.ai's main goal of the IPO is consistent with the actions shown by Baidu and Waymo in October this year, which is to expand the commercialization scale of driverless cars.
Under the tuyere of Robotaxi, is Pony.ai ready?
Robotaxi is ready to go, are ponies ready?
As the largest independently operated L4 driverless company in China, Pony.ai's every move has been attracting much attention.
The most important thing is technology, Pony.ai's autonomous driving is similar to Tesla's FSD, both of which are the end-to-end architecture of One Model. However, unlike FSD's training method, which relies more on real-world data, Pony.ai focuses more on building a world model through "simulation".
This is because, based on the development experience directly from L4, Pony.ai has long recognized that data is a "double-edged sword" for end-to-end intelligent driving models.
It is true that at a time when end-to-end has become the mainstream intelligent driving development model, most intelligent driving manufacturers, including Baidu and Tesla, attach great importance to the value of data. After all, in the end-to-end (including "segmented" and One Model) development models, vendors can not only save the cost of "ad hoc development" through data training. It can also rely on the model's imitation of the "old driver" to drive to achieve a better riding experience.
However, in Lou Tiancheng's view, this kind of training method will make it difficult for intelligent driving to surpass the human level, "The essence of end-to-end or large language models is only to fit existing data, and do not give some intelligent logic." So the capabilities of the model will be limited by the representation of the data." In other words, human data can no longer guide AI.
How to break through the upper limit of end-to-end intelligent driving capabilities driven by real driving data?
Pony.ai has built a set of world models that can accurately replicate and dynamically respond to real-world conditions. By allowing AI to "play freely" in the training process, autonomous driving can find a better driving strategy than "human driving", so as to achieve the goal of safer autonomous driving.
Such a technical concept allows Pony.ai to realize the operation of Robotaxi with less manpower.
According to the prospectus, Pony.ai currently has a pure R&D team of only 601 people. The overall team size and composition are very close to the Waymo model of 5 years ago (950 people in total, 350 in software engineering). As a comparison, on the side of car companies, He Xiaopeng, director of Xiaopeng Motors, said in May this year that Xiaopeng Motors' intelligent driving team will expand by 4,000 people this year.
Of course, this is also related to the number of vehicles supported and operated by the company, and the engineering team of the car company will be much larger than that of the L4 manufacturer, and it is still doing both traditional autonomous driving and end-to-end autonomous driving, and the number of teams will be relatively larger. With the end-to-end implementation, the intelligent driving team of car companies will also be more streamlined.
At the same time, the advantages of Pony.ai's autonomous driving technology also allow the company's AI algorithms to break through the limitations of vehicles and various platforms. Robotaxi and Robotruck (and possible mass production passenger cars in the future) can all implement the same autonomous driving system (virtual driver) at different levels of sensor hardware, end-to-end training, intelligent driving deployment, and data processing.
This means that in the next foreseeable "horse race" race in the Robotaxi industry, Pony.ai can continue to reap software service fees (virtual drivers) with higher gross profit margins by virtue of its general autonomous driving capabilities for different models. In addition, under the leverage of the software revenue model, Pony.ai will get rid of its current identity as an unmanned fleet operator and achieve higher profits and valuations.
As a result, Pony.ai's most important thing to do now is to rapidly expand the scale of fleet operations. As soon as possible, we will complete the verification of the capabilities and profitability of our own Robotaxi and Robotruck on a larger scale.
Perhaps, GAC's "emergency investment" before Pony.ai submitted IPO documents, and this year's unveiling of the Bozhi 4X Robotaxi model in cooperation with Toyota at the Beijing Auto Show means that car companies have already "geared up" to enter the Robotaxi industry.
I still remember that in 2022, many companies in the field of autonomous driving fell under the "capital winter" due to the delay in autonomous driving technology. In just two years, Robotaxi is far from opening up on the streets, and only funds and policies are left to bring large-scale "east wind".