On the evening of Oct.
8, the latest version of the Horizon prospectus was posted on HKEx’s website, indicating that the company had passed the listing hearing and IPO entered the final countdown.
Horizon first submitted the table in March this year, was listed overseas by the Securities Regulatory Commission in August, submitted to the Hong Kong Stock Exchange again in October, and updated the data for the first half of 2024.
Through the hearing, followed by the formal ringing of the bell and other processes, IPO is only a foot in the door.
In March, Black Sesame and Horizon submitted their IPO applications almost at the same time.
The former is the first autopilot chip company in Hong Kong and the second company listed on the HKEx under Chapter 18C, but the new stock was broken on its first day of listing.
The latter hit the capital market for the first time and also experienced some twists and turns.
From the tuyere point of view, in addition to the technological revolution of artificial intelligence and chips, the horizon has also encountered the opportunity of autopilot.
It is against this background that hard technology companies such as Black Sesame, Horizon and Xinchi Technology, whether they compete in the wrong place or meet each other, have finally found their own way in an inner circle, although rugged is the norm.
At the key node of IPO, Black Sesame and Horizon both face the same problem, with persistent losses and relatively high customer concentration.
On the one hand, as the darling of the investment community, the company’s valuation has been rising over the past few years, but loss is an unavoidable problem.
After all, the tuyere period of the industrial track is limited, in the early stage of development, economies of scale have not yet been released, only higher R & D investment and deeper customer service, it is possible to get tickets and sit on the poker table.
On the other hand, large customers with a high degree of concentration are often two sides of the same coin.
In the short and medium term, it will have an impact on the company’s operating performance and core financial data.
in the long slope and heavy snow track, gradually getting rid of the problem of revenue dependence on head customers is a compulsory course for all start-ups.
01, the moat behind R & D, get the prospectus, the industry likes to use the horizon and black sesame to do horizontal core performance comparison, but the two companies in the business focus and entry point, there are many differences.
However, in terms of R & D investment, Horizon has really “invested a lot of money” in recent years.
Just from the 2023 data, black sesame accumulated revenue of 2.
677 billion yuan, R & D investment of 1.
363 billion yuan, accounting for about half of the revenue.
This year, Horizon revenue was 1.
552 billion yuan, and R & D investment was 2.
36 billion yuan, accounting for 150% of that year’s revenue, much higher than 50% of black sesame.
Where did the huge R & D expenses go? According to the prospectus, the employee with the highest R & D cost on the flat line is the employee input, which reached 837 million yuan in the first half of 2024, and as of June 30 this year, the company’s R & D team had a total of 1696 people, accounting for 73% of the total staff, an increase of 218 people over the end of last year.
Thus it can be seen that horizon’s R & D expenditure is mainly paid for the welfare of R & D personnel, accounting for 60% of the total R & D.
But specific investment in which business R & D sectors, the prospectus does not specify.
Industry insiders told Auto Commune / C Dimension that Horizon has been committed to strengthening the research and development of software algorithms in the past two years, further promoting the combination of software and hardware, and software algorithm development, especially production-level algorithms.
is the focus of investment.
In addition, streaming, software tools and so on are also more expensive.
However, according to the prospectus, the horizon’s losses were narrowing from 2021 to 2023, with adjusted net losses of 1.
103 billion yuan, 1.
891 billion yuan and 1.
635 billion yuan respectively, compared with 804 million yuan in the first half of this year, 19.
3% lower than the same period last year.
The background of the horizon has never changed. ” Dr. Yu Kai, the founder of the company, has stressed this point many times in public.
In his view, Horizon is essentially an algorithm company, a software company in the cloak of chips, an atypical chip company.
These are all regarded as moats on the horizon.
From the point of view of the industry, the horizon’s high R & D investment in recent years is protecting this special moat.
The R & D investment on the horizon will be 1.
144 billion yuan in 2021, rising to 1.
88 billion yuan in 2022, 2.
366 billion yuan in 2023 and 1.
42 billion yuan in the first half of 2024.
Yu Kai once said that technology has never been a moat.
He said in an interview, “Technology has no moat, it is the wave of technology of every generation, you have to get past it, once you are overturned, there will be no you.
What is the core of a technology company? Through a wave of technology, you always go on, but what you precipitate is not technology, but brands, customer relations, and so on.
These are the real moats.
” Horizon management knows this well.
At the Haikou World New Energy Automobile Conference last year, Chen Liming, president of Horizon, said in an interview that the focus of the industry is on the hot domain convergence and BEV Transformer, but the horizon has a long-term vision, and the chip design process should not only consider Transformer, but should focus on this as a basis to think about the opportunities and technological breakthroughs brought by Transformer.
BEV+Transformer is one of the trends, but the future will definitely be replaced by new trends that must be taken into account in the design of the product.
” A year later, at the scene of this year’s Haikou World New Energy Automobile Conference, Chen Liming shared new thoughts in the field of perceptual algorithms.
BEV+Transformer is the current mainstream algorithm, but at the same time BEV+Transformer also has another problem, that is, computing resources.
BEV itself is a dense algorithm, which distributes computing resources evenly to the surrounding environment, which will cause a huge waste of computing resources when there is no need for computing resources, and it is difficult to balance the computing power.
” Horizon is aware of some challenges existing in BEV+Transformer, and proposes a perceptual architecture of Sparse4D long-term sequence pure sparse fusion, which makes full use of computing resources through the sparse sensing framework, hoping to solve the problem of insufficient computing resources.
At the same time, through the feature fusion of long time series, the object is tracked better, and the end-to-end algorithm from image input to trajectory output is realized as a whole.
It is reported that Horizon opened up related algorithms in January, and the company believes that these trends will succeed BEV and Transformer as the next generation of perceptual architecture and bring new breakthroughs to the whole perception.
First of all, the perceptual neural network, then enter the neural network of regulation and control, and then go down to the modular end-to-end, and finally the development direction should be the end-to-end of the whole stack.
” , 02, the new inflection point of the integration of software and hardware, at presentHorizon’s main business consists of three parts: product solutions, licensing and services, and non-vehicle solutions.
In terms of revenue share, the volume of licensing and service business is the largest.
It is reported that Horizon allows customers to purchase algorithm modules independently, as well as other components of the algorithm and technology stack.
Because the marginal cost of software sales is low, the gross profit margin of this business can be as high as 89%, which is much higher than 61% of the product solution.
In addition, the revenue share of licensing and services has also continued to increase in the past few years, from about 40 per cent in 2021 to about 50 per cent in 2022, and the revenue share of licensing and services has reached 62 per cent in 2023, which is the core business of the horizon.
In the first half of this year, Horizon Licensing and Licensing Services accounted for 73.
9% of revenue, or about 690 million yuan.
According to the prospectus, revenue from horizon licensing and licensing services continues to grow, mainly due to strong licensing demand for algorithms, development tools and software for ADAS and AD solutions and related services Last year, Horizon licensed Cool Ruicheng, a software joint venture between Horizon and Volkswagen Group, to use related algorithms and software, and earned a total of 351 million yuan from Kerry Cheng in the first half of this year.
Horizon has stressed over the past few years that it is a “software and hardware solution provider”, playing the role of a chip provider while providing software algorithms.
However, before the smart driving chip journey 5 (referred to as J5), the company’s software and hardware solution still has a lot of room for improvement.
Starting with J6, Horizon is trying to create a “model room” based on software, algorithms and chips.
The hardware is J6 series chips, and the software is Horizon SuperDrive (hereinafter referred to as HSD), the urban intelligent driving solution.
HSD works with J6 to create a high-level intelligent driving solution with the cooperation of software and hardware.
J6 is an important inflection point for the horizon, which is an important attempt for the company’s chip + software to achieve higher performance, and it is also a key leap to challenge the final competition of smart driving and to achieve the two dimensions of scale up (performance improvement, performance optimization) and scale out (scene generalization).
Among them, scale up mainly has two routes: “rule-based” and “end-to-end”.
The latter is very popular this year, which greatly improves the performance limit of self-driving.
In fact, even if it is the industry’s leading company, it is difficult to do one of scale up or scale out to the extreme at this stage, and it is even more challenging to take care of it at the same time.
Mullison, chief architect of Horizon algorithm platform, once mentioned in an interview that the end-to-end advantage is to raise the performance upper limit of some scenarios, but relatively speaking, it is more difficult to ensure the lower performance limit of the system.
In fact, Tesla has not completely solved this problem.
Of course, Tesla has done a very good job in the industry at present.
Frankly speaking, everyone’s urban NOA is not good enough, so there is still a lot of room for the industry to bring about the progress of urban experience solutions.
The end-to-end accumulation of the horizon mainly lies in iterative algorithms, engineering infrastructure and the ability to combine software and hardware.
” Chen Liming, president of Horizon, also made it clear at this year’s China Automobile Forum that end-to-end will not be achieved overnight, but will gradually iterate towards a complete end-to-end.
In the absence of sufficient data, it is still necessary to gradually implement end-to-end, module-by-module replacement, and complete end-to-end security at the same time, in this way of relatively solid engineering infrastructure and rapid iteration, step by step to improve the system performance upper limit, but also ensure its lower limit.
” According to the previous prospectus of Black Sesame, the company’s funds have mainly flowed to sales and R & D in recent years, of which R & D expenditure accounts for the highest proportion of costs, which were 250 million yuan, 590 million yuan and 760 million yuan respectively in the past three years.
According to Black Sesame, the autopilot Soc track has three main barriers: technology, R & D cycle & input, and customers.
In the customer field, black sesame has a relatively high degree of customer concentration.
The company, whose customers are mainly automotive OEM and Tier 1 suppliers, has accounted for 47 per cent, 40 per cent and 43 per cent of its revenue from its largest customers in the past three years, respectively.
The single chapter itself mentioned that when the company is from 0 to 1, the most difficult thing is the technology and how to serve the core project.
But from 1 to 100, we are bound to face more customers, and the biggest challenge at this time is how to allocate resources.
Revenue depends on head customers, which is not only a practical problem for black sesame, but also a “curse” for horizon business to break the game.
According to the prospectus, the top five clients on the horizon account for 60%, 53% and 68% of revenue, respectively.
According to the official data of Horizon, the Cheng family has shipped more than 6 million models, more than 270models have been produced on a fixed point, and more than 110models have been put on the market.
In addition, Horizon’s software and hardware solution has been adopted by 27 OEM (42 OEM brands) and equipped with 290models.
Thus it can be seen that although the number of customers on the horizon is expanding, it is still highly dependent on large customers.
At this stage, horizon’s revenue is mainly due to the contribution of Volkswagen, which accounts for nearly 40% of horizon’s total revenue.
After the success of IPO, with the expansion of market size and business territory, Horizon needs more customers to reduce the risk of the previous concentration of major customers.
However, judging from this year’s latest prospectus, the top five clients on the horizon have also changed.
In 2023, besides the top five customers of the company, such as ideal, Fretek, SAIC, Yingheng Technology and Core Cheng, Siwei Tuxin has replaced ideal as the second largest customer in the first half of this year, and BYD is also on the company’s top five customer list.
From a strategic point of view, the cooperation between Horizon, Fretek and Siwei Tu Xin is mainly focused on low-level intelligent driving, and high-level programs will continue to work hand in hand with the whole vehicle factory, which is also the main direction of future efforts.
It is worth mentioning that, in addition to the high customer concentration, Horizon, like other technology companies, faces the dilemma of burning money.
However, at the end of 2023, the company had 11.
36 billion yuan left in cash and cash equivalents, which is better than the 1.
2 billion yuan of Black Sesame Intelligence.
According to the Horizon prospectus, the company’s losses for fiscal 2024 are likely to continue to expand, mainly due to high investment in research and development.
Horizon disclosed in its prospectus that the loss is expected to increase significantly this year, which is mainly due to continued investment in research and development, in addition to factors such as changes in valuation and the core of Volkswagen’s joint venture plan.
The R & D investment is only one side.Face., For companies like Horizon, the cycle time is long, mass production speed is high, and adapting to customers is also a challenge.
Previously, Black Sesame revealed that taking the company’s Huashan series of autonomous driving chips as an example, it takes 3 to 4 years from product definition, streaming, sealing and testing, vehicle regulation certification, algorithm tool chain, functional safety certification to end customer verification, customer experience, and mass production.
Now, from car companies to Tier1 and Tier2, it has become a mature chain.
Everyone not only improves efficiency in running-in, but also improves engineering capabilities.
There is still room for cost reduction in every link of the chain.
How to control costs is a new topic for Horizons and a new challenge for Tier1 and vehicle manufacturers.
Return to First Electric Network Home>.
