A few days ago, a bizarre commercial case of a “former driver extorting the CEO of a company” brought the “Four AI (Artificial Intelligence) Dragons” back into the focus of public opinion. The protagonist of this case, Kuang Sees Technology as one of the “Four AI Dragons”. The core of the case was that the driver wanted to sell the company's “recordings of sensitive information” to competitors.

This detail has become a vivid footnote to the current intense competition among Chinese AI companies.

Business competition, IPO competition. 2021, which is yet to end, is destined to become a listing year for China's AI industry. The “Four AI Dragons” IPOs have achieved substantial breakthroughs: on September 9, the Kuangshi Science and Technology Innovation Board IPO was approved; on August 27, Shangtang Technology submitted a prospectus to the Hong Kong Stock Exchange; on July 20, Yuncong Technology's initial listing on the Science and Technology Innovation Board was approved; Etu Technology terminated its IPO on the Science and Technology Innovation Board in July, and there is news recently that it will go public in Hong Kong. In addition, companies in segments such as healthcare and chips on the AI circuit have also accelerated their listing.

The listing period is also a day to catch the exam. Over the past few years, with strong policy support and long-term capital pursuit, AI in China has gradually become an “influencer” industry, and everyone talks about it for themselves. But as far as AI companies are concerned, what is their actual operating situation? What are the R & D capabilities and technical reserves? What are the commercialization paths of each company? What will happen to the industry pattern? How much capital investment is needed in the future? How to solve the problem of insufficient commercialization in the application field? How to compete with the industry's traditional leader “+AI”? How to deal with the crushing of the Internet tech giants' “all in AI” strategies?

As the “Four AI Dragons” concentrated on the IPO exam, many problems were revealed for the first time by disassembling prospectuses and comparing financial data and business models. As of press release, the reporter tried to contact the “Four AI Dragons” for an interview, but was unable to do so due to the market silence period.

Although they are collectively known as the “Four Little Dragons of AI,” they are actually “one big and three small.”

As the boss of the “Four AI Dragons”, Shangtang Technology's revenue is significantly higher than that of the other three companies. According to the prospectus, Shangtang Technology's revenue from 2018 to 2020 and the first half of 2021 was 1,853 billion yuan, 3,027 billion yuan, 3.446 billion yuan and 1,652 billion yuan respectively. Judging from the average annual revenue scale of the past three years, Shangtang Technology's revenue is equivalent to more than 2 times that of Kuangshi Technology, 4 times that of Yuncong Technology, and 6 times that of Yitu Technology.

Behind the scale of revenue is a difference in business models.

In the prospectus, Shangtang Technology positioned itself as an artificial intelligence software platform-based company.

Currently, Shangtang Technology mainly has four major platforms: SenseFoundry-Enterprise (Shangtang's Ark Enterprise Open Platform) for smart commerce, SenseFoundry (Shangtang Ark City Open Platform) for smart cities, SenseMe, SenseMars and SenseCare platforms for smart lives, and SenseAuto (Shangtang Seying Smart Car Platform) for smart cars — covering almost all business, life, and travel scenarios.

It's easy to see that behind the goal of “empowering all industries”, Shangtang Technology's core position for itself is a platform, which can also be called an “AI factory.” The company once claimed that in order to support the continuous operation of the entire “factory”, it invested about 5 billion yuan to build a supercomputing center and open source core algorithms. Prior to the publication of the prospectus, Shangtang Technology launched its new artificial intelligence infrastructure — the Sensecore Shangtang AI Device.

The construction of a new artificial intelligence infrastructure platform shows the commercial ambition of Shangtang Technology. In essence, it may also reflect the company's predictions about the future competitive pattern of the AI industry. “Project outsourcing does not last long, and the platform can only make a long profit.”

The reporter interviewed many investors in the AI field and learned that over the years, Chinese AI companies have basically been based on the ToB model, which is essentially a customized “outsourcing” business. Seemingly tall AI companies are doing the “hardest work” of doing algorithms and SDK projects for hardware companies, and “their implementation is narrow, and demand is unstable.”

As a result, Xu Li, CEO of Shangtang Technology, once publicly stated that its strategy is “1+1+X”, that is, one core base is Shangtang's Sensecore AI device; one core technology is Shangtang Technology's own AI technology such as face and human analysis; X is used in an unlimited number of industries.

Check out the prospectus. As of June 30, 2021, Shangtang Technology has accumulated more than 22,000 artificial intelligence models that enable different applications. That's an astonishing number.

In addition to Shangtang Technology, Kuangshi Technology and others are also working hard to build their own AI platforms. In 2020, Prospect Technology released the AI productivity platform Brain++, with the intention of shortening the R&D cycle of AI algorithms and improving the efficiency of AI and industry integration; Yuncong Technology released its industry-level artificial intelligence product and capability platform “Qingzhou” platform in July 2020.

However, unlike Shangtang Technology, the actual construction progress and commercialization capabilities of these platforms were not fully disclosed in the prospectus.

So, how are they making money now?

According to the prospectus, Prospectus Technology's business mainly includes three categories: consumer IoT solutions, urban IoT solutions, and supply chain IoT solutions; Etu Technology is positioned as a “computing power provider”, with artificial intelligence chip technology and algorithm technology as the core; Yuncong Technology targets the four fields of smart finance, smart governance, smart mobility, and smart business. Human-computer collaborative operating systems and artificial intelligence solutions are its main revenue components.

There are so many complicated proper terms that often make laypeople unaware of this. But aside from these, unlike Shangtang Technology's core business model, Kuangshi Technology, Etu Technology, and Yuncong Technology are all deeply involved in different vertical markets.

For example, Prospect Technology's core strategy can be simplified to “hard core AIoT” — focusing on applying AI capabilities to IoT scenarios.

Currently, the Internet of Things places more emphasis on direct interconnection between applications and devices, and still lacks intelligent sensing, analysis, and collaboration capabilities. Based on this, Prospect Technology built an integrated product system of “algorithm+software+hardware” and launched the AIoT operating system in the IoT era.

Why did Prospect Technology choose the Internet of Things? In its prospectus statement, it focuses on industries where the pain points of the industry are clear and algorithms can generate great value. Among them, it is particularly different from other AI companies. Prospect Technology not only uses AI technology, but also hardware, and is more vertically integrated.

“The technical threshold and path of each company when it started were close, and commercialization scenarios also began in the fields of consumption, security, finance, etc., but each company's statement in the prospectus was different.” A VC investor who has been following the AI industry for a long time told reporters, “As of today, in the context of increasing homogenization of algorithm capabilities, whether as a platform or as a vertical, each company is essentially trying to establish its own core technical barriers and find opportunities for 'large-scale' applications in different industries.”

The different direction of the “Four AI Dragons” business model has brought four financial statements with huge commonalities and differences.

Overall, the commonalities are: continued growth in operating income, increasing loss amounts, extremely expensive R&D investment, and large customer dependency and cash flow that are far from improving; the difference is the wide gap in absolute indicators such as revenue scale, R&D investment, and proportion of R&D personnel, increasingly differentiated revenue growth rates, and volatile gross profit margins.

First, let's look at revenue data. From 2018 to 2020, Shangtang Technology's revenue was 1,853 billion yuan, 3,027 billion yuan, and 3.446 billion yuan respectively; Kuangshi Technology's revenue was 854 million yuan, 1.26 billion yuan, 1,391 million yuan, and Yuncong Technology's revenue was 484 million yuan, 807 million yuan, and 755 million yuan respectively. Etu Technology's revenue in 2018 and 2019 was 304 million yuan and 717 million yuan respectively.

In absolute terms, Shangtang Technology is far ahead; in terms of growth, Shangtang Technology continues to rise, and Kuangshi Technology and Yuncong Technology are relatively stable. In the prospectus, Shangtang Technology stated that the company's 2020 revenue scale already ranked first in Asia in the same industry.

In the first half of 2021, Kuangshi Technology's revenue was 670 million yuan, an increase of 91.27% over 350 million yuan in the same period in 2020; Shangtang Technology's revenue was 1.65 billion yuan, an increase of 91.86%. The rapid increase in revenue from Shangtang Technology and Kuangshi Technology proves that their tracks still have long-term value.

The second is investment in research and development. According to the prospectus, all four companies have invested most of their revenue in research and development.

The total R&D expenditure of Shangtang Technology over the past three and a half years reached 6.991 billion yuan. Specifically, from 2018 to 2020 and the first half of 2021, R&D investment accounted for 45.9%, 63.3%, 71.21%, and 107.3% of revenue, respectively. From 2018 to 2020, Prospect Technology's R&D expenses accounted for 70.94%, 82.15%, and 64.44% of revenue for the same period, respectively. The R&D expenses of the other two companies also account for more than 50% of revenue.

In terms of absolute value, Shangtang Technology is also the largest R&D investment, and its share of investment is still increasing rapidly, confirming the company's “all in technology” strategic position.

Another direct measure of R&D investment is R&D talent.

According to the prospectus, Shangtang Technology has the largest number of employees, over 5,000, with the highest proportion of R&D talents, close to 70%; Kuangshi Technology, Yuncong Technology, and Yitu Technology also account for more than 50% of R&D talents. In the first half of 2021, Shangtang Technology had 3,593 R&D personnel, the absolute number exceeding the other three companies combined.

Looking at the gross profit margin, which reflects the quality of profit, there is a huge difference between the four companies. In 2020, Shangtang Technology's gross margin was 70.6%, Kuangshi Technology was 33.11%, and Yuncong Technology was 43.46%. Etu Technology's gross margin in 2019 was 63.89%.

Why is this happening?

Looking at the prospectus, Shangtang Technology mainly uses software services and maintains a high gross profit margin all year round; Prospect Technology, which integrates software and hardware as a solution, has had very obvious gross margin fluctuations in recent years, showing a downward trend. Changes in business model trends may be the core cause of differences in gross margin.

In terms of customer structure, it is currently difficult for all companies to escape the dependency of major customers. Generally speaking, big customers mean big revenue, but they also mean product customization, and large-scale replication advantages cannot be formed.

Changes in major customers can also lead to operational instability. Compared with 2018, 4 of Prospect Technology's top five customers in 2019 were “replaced”, while its top five customers changed 3 more in 2020.

On September 9, at the site of the 66th review meeting of the Science and Technology Innovation Board Listing Committee in 2021, the Science and Technology Innovation Board Listing Committee requested Prospect Technology to further explain the company's core technical competitiveness and future development prospects in response to the situation where Prospect Technology's main customers were unstable, unconcentrated, and non-industry leaders.

Finally, there is the profit situation. The four companies are generally still in a period of loss.

According to the prospectus, from 2018 to 2020 and the first half of 2021, Shangtang Technology's net losses were 3.433 billion yuan, 4.968 billion yuan, 12.158 billion yuan and 3.713 billion yuan respectively. Prospect Technology lost 12.77 billion yuan in the three years from 2018 to 2020, and Etu Technology lost 6.1 billion yuan in two and a half years. Yuncong Technology lost 690 million yuan in 2020.

However, it should be mentioned that most of these astonishing loss figures include “fair value losses” (mainly losses due to employee options). For example, after deducting net losses of Shangtang Technology, adjusted net losses were $220 million, $1.04 billion, $880 million, and $730 million, respectively.

How do you view the high investment in R&D and continued losses? The reporter was unable to interview these four companies. However, the reporter asked veterans in the AI investment industry. The general consensus is that AI companies are not afraid of losing money; they are afraid that they will not lose money or invest in the short term.

“Investing in AI means investing in scientists. Differences in technology and experience among algorithm engineers determine the productivity of algorithm models. In the future, AI technology will have obvious winner-take-all characteristics, and the best algorithm models will be used more widely. On this track, as long as there are enough scientists, the commercial value will be realized sooner or later.” The AI investor mentioned above told reporters.

Is that really the case? Different people have different answers.

The reality is that although it has been developing for many years, judging from the revenue structure, the TOB business for governments and institutions is still the core mainstay of the “Four AI Dragons” that support revenue.

At the same time, when it comes to commercializing market segments, although each has made great efforts to expand, the reality is that the AI capabilities of the traditional giants of the original territory are rapidly improving, and their competitive advantage is not that obvious.

For example, in the smart city market that currently accounts for the largest share of revenue, AI companies are facing strong competitive pressure brought about by the original hardware manufacturers Hikvision and Dahua Co., Ltd.

In the security market, companies such as Hikvision obtained similar technical capabilities through self-construction or procurement, and steadily occupied market share. Behind this is the fact that some of the technologies that AI companies were proud of in the past are being generalized and popularized. Especially in the face of traditional software and hardware vendors with obvious channel and supply chain advantages, it is becoming more and more difficult for AI companies' algorithm longboards to form differentiated advantages. To this end, some of the “Four AI Dragons” have begun to select supporting hardware as an overall solution integrating software and hardware to cope with sinking competition.

At the same time, it also drove them to find a new battleground: after selling their previous core medical business, Etu Technology targeted a new scenario of autonomous driving; Cloud positioned itself as a human-robot collaboration solution provider from technology. In the process of continuing to expand in the financial field, it added research investment in robotics and IoT technology to try to build more standardized AI products; and Prospect Technology entered AIoT, looking at the logistics business as a future growth point.

However, it is not easy to expand into new markets and new tracks. In its prospectus, Shangtang Technology clearly stated, “We are not familiar with the new vertical... The vertices we decide to expand may have one or more existing market leaders. Such companies may be more competitive than us because they have experience in doing business in this market, deeper industry insight, and higher customer brand awareness.”

Simply put, on the vertical track, the technology window period for pure AI algorithm companies is being shortened, and it is necessary to complete the transformation of technology to commercial use within the window period as much as possible.

Therefore, unlike the three AI companies such as Kuangshi Technology, Shangtang Technology chose to follow a different path — AI platformization. Is this Kangzhuang Avenue?

In fact, as early as when Shangtang Technology was founded, Internet giants such as BAT were already making AI platform-based layouts, and these comprehensive AI platforms all introduced free policies. Their logic is: by attracting more enterprise users to settle in, building an ecosystem and accumulating data, and forming a closed profit loop of “AI solicits customers, cloud computing earns money” with existing cloud computing and other businesses.

At present, domestic AI open platforms have formed a relatively clear pattern: Internet technology giants such as Baidu, Ali, Tencent, and Huawei focus on comprehensive AI open platforms, accounting for more than 80% of the domestic market share; while Shang Tang, Kuang Si, and iFLY have created AI open platforms in segments such as computer vision and speech recognition, competing for the remaining 20% of the market share.

For some time to come, the “Four AI Dragons” will face the double pressure of business competition and financial distress regardless of whether their development strategy is vertical downturn or platformization. The intention to rush to the market at this point is self-evident.

Using survival as the standard, the “Four AI Dragons” test run has only just begun.