Shulou(Shulou.com)06/03 Report--

Source: the power of machines

This article is about 6280 words. It is recommended to read it for 12 minutes.

This article introduces that on the evening of May 22nd, the Shanghai Stock Exchange disclosed the announcement of the 33rd Review Conference that the Cambrian Science and Technology Innovation Board of China Science and Technology will be launched on June 2. At this point, the first batch of Chinese AI companies officially opened the floodgates, pouring out a quasi-listed queue composed of 18 unicorns.

After 2015, the third wave of machine learning was detonated by the emergence of "AlphaGo", which happened to hit the youth of a generation.

In the hot land of creating wealth, such as image technology and the security market, China's first batch of artificial intelligence companies have emerged one after another, leaping every five years and once every seven years, accurately stepping on the nodes of the times.

Their high-frequency exposure, their bold exploration, the initial construction of a perfect commercial closed loop, with the joy of maturity, with the troubles of youth, they stand out from unicorns and enter the new queue of the secondary market, and the first batch of AI's "bar mitzvah" symphonies play slowly.

"the primary market can help you run fast, but it is the secondary market that really allows you to take off. "that's what listing is all about.

But the premise of taking off is to grow solid arms, and the experience of ripening in the capital greenhouse in the early years has branded AI with deep To VC inertia.

The stubborn contradiction between cost and scale is like a chattering opposing debater, constantly questioning the argument that "artificial intelligence is a good business", making self-proof bound to be difficult and long.

On the evening of May 22nd, the Shanghai Stock Exchange disclosed the announcement of the 33rd Review Conference that the Cambrian period of China Science and Technology will be launched on June 2, Science and Technology Innovation Board. At this point, the first batch of Chinese AI companies officially opened the floodgates, pouring out a "quasi-listed" queue composed of 18 unicorns.

"if you do not find a break-even point, it all depends on blood transfusion, ignoring that technology will certainly not choose to go public now. "

Song Chunyu, a partner in Lenovo venture capital, told the heart of the machine. As a well-known unicorn catcher, Lenovo Venture Capital has successfully "captured" two "quasi-listed" companies in the AI sector-Xia Technology and Cambrian. According to the heart of the machine, after several twists and turns in the Hong Kong Stock Exchange, neglect has now shifted its focus to the domestic internal medicine board.

"the stability of the company's infrastructure and the level of investment have tended to be balanced, and now we can move on to the next stage," Song Chunyu explained. For example, 3000 GPU is enough to meet the business needs in the next three years, without unlimited investment.

Since the company first submitted its application for listing to Hong Kong stocks in August 2019, a number of AI companies have been "disclosed" about their listing progress, including not only those companies that are good at machine vision (CV) and voice recognition algorithms, but also companies such as AI chips and smart hardware.

According to the comprehensive arrangement of public materials, the companies to be listed include those that have clearly expressed their intention to IPO, but have not disclosed the timetable and landing section. Shangtang Technology has explicitly denied the listing plan, so it has not been included in it; Squirrel AI is an education brand of Yuxue Education Group.

"these companies share some common iconic characteristics, seize technological breakthroughs and initially achieve a commercial closed loop. Song Chunyu said, but there is still a big gap with giants such as Google and Amazon. The primary market has already "spent" a lot of money, and there are demands for withdrawal, and then social capital relay is needed to help good companies speed up their products and commercialize the closed loop.

Nuggets Kechuang board

Science and Technology Innovation Board Mingjin opened the market in July 2019. Over the past three quarters, each company has raised an average of 1.19 billion yuan.

Among the first batch of 18 AI companies to be listed, as many as 14 companies made it clear that impact Kechuang board. Due to technical and business sensitivity, vision companies choose significantly fewer than voice categories, which usually include overseas listings as options.

Compared with Science and Technology Innovation Board's five sets of listing standards, the market value threshold of Standard one is the lowest (1 billion yuan), and the core focus is profitability. in order to seek stability, the vast majority of enterprises choose the one with lower risk. However, AI's choice is slightly different.

In addition to stone technology "from Daliu" selection criteria one, like Hong soft Technology, Cambrian also selected criteria two, the core focus on R & D investment. If the news of technology seeking Science and Technology Innovation Board listing is true, it is also possible to choose standard two.

Science and Technology Innovation Board five sets of listing standards and conditions, from Ernst & Young's China Accounting Newsletter

AIoT companies like Stone Technology have no sense of disobedience even if they choose to go public on the gem. However, Standard 2 is the only standard that requires R & D investment by enterprises, an investor told Machine Heart.

According to the requirements, the cumulative R & D investment of IPO enterprises in the past three years shall not be less than 15% of the cumulative operating income in the past three years, which is mainly suitable for enterprises that have reached the commercial stage through continuous R & D investment in key areas.

"companies like Cambrian and Kuangshi have a very high R & D investment and leading technology, which is more in line with Science and Technology Innovation Board's original intention of aiming at" hard technology ". Song Chunyu thinks.

Although most companies are listed for the first time, there are also two companies that have moved from the new third board, Shadow Spectrum Technology and Xiao I Robot. As long as we can successfully land on the Kechuang board, enterprises, shareholders and investors all have the opportunity to achieve a "win-win" situation. "the most important thing for this kind of companies is to do a good job in threshold valuation management. "the above investor told the heart of the machine.

According to the public information and other comprehensive arrangement.

From the perspective of macro AI industry chain, 18 companies to be listed cover the basic layer, the technical layer and the application layer. Basic layer at least (2), Cambrian is a veritable "AI chip" the first, the vast majority concentrated in the technology layer and application layer.

The technical layer is the core of artificial intelligence. With the advantage of talents and the number of AI papers in the world, the layout of Chinese AI companies in voice and visual algorithms has been relatively perfect. Nine of the companies to be listed this time belong to the technical level.

The application layer is also the most active area in China's AI market. Seven companies are concentrated in this sector, applying more mature voice, computer vision and other technologies to a number of segmentation scenarios. Data show that among the financing events in the artificial intelligence industry segment from 2014 to 2019, the largest number of applications were invested in products and solutions. Thus it can be seen that capital also favors companies close to commercial realization.

Capital investment products and solutions are the most widely used, from the "Investment Research Institute Chong period Capital 2019 China artificial Intelligence Industry Investment and financing White Paper"

Focusing on the technology layer, we found that CV companies (including application layer companies directly related to CV) accounted for the majority, a total of 10. There are few voice AI companies, only 5. And of the top five companies with the highest valuation, with the exception of Ming Lue Technology, all are CV, or directly related to CV technology.

The reason for the formation of such a distribution is not only related to the maturity of CV technology, but more importantly, China provides a unique application scene. There is a "real need" here, Hu Yangzhong, president of Haikangwei, once told the media.

Historical data show that Ping an City, especially the "Xueliang Project" has continuously promoted the growth of the domestic security market. CV companies have not only dug up the first bucket of gold here, but the security business is also one of their main sources of revenue. For example, the revenue of the Internet of things business in cities that ignore technology accounts for 73%. According to Yao Zhiqiang, co-founder of Yuncong Technology, the company's biggest source of revenue is also the security business.

The scale growth trend of the domestic security market is very obvious. The Source Prospect Research Institute "the Development Plan of China Security Industry"

By contrast, the way for Voice AI to go public appears to be "arduous". Although the "age" of these companies is generally higher than that of the CV category, they are significantly lower in terms of revenue and valuation.

Voice AI was founded even earlier. Almost all the companies founded before 2012 are Voice AI, and Xiao I Robot can even be traced back to 2001. So far, these companies have been established for as little as eight years to nearly two decades. However, in 2018, Yunzhisheng Camp received 100 million yuan, and this year, cloud technology has exceeded 600 million. Yunzhisheng is valued at about 8 billion, and it is also the most highly valued voice AI unicorn, not only behind the four Little Dragons of CV, but also further away from Yuncong's valuation of 25 billion.

Voice AI spans the availability phase, basically after the outbreak of deep learning in 2016, before the voice technology was still in the stage of development, during which the start-up company will have a difficult life, a head of AI startup told Machine Heart.

Moreover, voice interaction involves a very complex technology chain, including core technologies such as acoustic processing, speech recognition, semantic understanding and speech synthesis, which is much more complex than computer vision, which greatly affects the commercial realization of these companies. That's why iFLYTEK is still being questioned after it went public in 2008.

Turning its attention from the technology layer to the application layer, squirrel AI and Todong Technology, which are to be listed on the market, come from the fields of education and finance, respectively. Big entertainment gives birth to shadow spectrum technology and extreme chain technology. In addition to a good information base (education may be slightly inferior), what these industries have in common is that the market demand is really huge. For example, the first source of income for banks is lending, and the core demand for lending is risk control.

Interestingly, these four companies are currently the only ones willing to be "honest" about their revenue. "the cash flow of the education industry is very abundant, so the hematopoietic ability of squirrel AI is also very strong, almost able to achieve cash flow balance. "Squirrel AI CEO Li Haoyang said in an interview.

The four application layer AI companies, Tong Shield Technology, Squirrel AI, Shadow Spectrum Technology and Polar chain Technology, are also the only companies that are willing to be "honest" about their revenue (2016, 2017, 2018).

Sky-high price valuation VS 50 times market sales rate

Most of the most expensive AI companies in the world also appear in the proposed listing lineup. You must choose and ignore the highest valuation of science and technology, Cambrian, Yuncong technology and other companies are also "expensive". However, many people don't understand why companies with low revenues are worth billions of dollars.

According to the public information and other comprehensive collation, in order to facilitate sorting, unified conversion to RMB valuation.

According to the prospectus, it was not until the second half of 2018 that the net profit of science and technology became positive, with a profit of more than 30 million yuan in the first half of 2019. The Cambrian has been losing money, with a loss of 1.18 billion in three years.

It must be admitted that the true state of commercialization must not be as good as PR advertised. "go out and ask CEO Li Zhifei said frankly in an interview with the domestic media.

At the same time, however, these AI companies are pursuing a steep sales curve as much as possible. For example, revenue has been rising rapidly in the last three years, achieving business income of 67.8 million yuan, 313 million yuan and 1.427 billion yuan in 2016, 2017 and 2018, respectively. In the first half of 2019, 949 million yuan of income has been generated, with a compound growth rate of more than 358%. According to Yao Zhiqiang, Yuncong's annual compound growth rate of science and technology is about 300%.

Cambrian revenue has grown by more than 50 times in three years, which is an excellent performance in the AI chip industry.

Most companies are reluctant to release revenue data and sort them out according to public information.

Since PS is more suitable for fast-growing technology companies because of dilution of profits, and is often used to measure whether valuations are too high, compare the PS of these companies.

For example, if Skytech is valued at 28 billion yuan and revenue is 1.4 billion in 2018, then the PS is about 20. Compared with Hong soft Technology (the market-to-sales ratio of the year is about 50), it is basically in line with the PS valuation method.

Let's take a look at the Cambrian. Wind data show that Science and Technology Innovation Board semiconductor industry average market-to-sales ratio (PS) 21.42. Based on a valuation of $22 billion, the Cambrian market-to-sales ratio (PS) is about 50, significantly higher than the average, with revenues of 440 million in 2019. However, compared with Lan Qi's market-to-sales ratio of 51 (2019), it is also basically in line with the PS valuation method.

We found that compared with Science and Technology Innovation Board's "elder" worth, the valuation of AI is actually not "too outrageous". "at present, Science and Technology Innovation Board's valuation is relatively high, and this is a necessary stage," a researcher from a securities research institute told Machine Heart. "this is also hoping to attract hard-tech enterprises to list in Science and Technology Innovation Board with relatively good valuation and liquidity. After all, (Science and Technology Innovation Board) shoulders the important task of leading economic transformation. "

However, many people still have reservations about this. It is difficult to give a reasonable price-to-sales ratio (PS), and Internet companies also burn money, as long as the scale can grow rapidly and consider profits after burning out a certain scale, one investor told Machine Heart. However, the price-to-sales ratio of more than 10 times is already a little high, not to mention dozens of times, at least in the short term.

More importantly, some "can't even see S (sales)," a practitioner in the chip industry told us. For example, companies such as RF chips with a certain sales scale generally have a market-to-sales ratio (PS) of 6-8 times. No matter what chips they do, AI chips also need proof of sales revenue.

Is AI a good business?

The difficulty of performance growth is also reflected in the struggle on the business model of these proposed listed AI companies. At least from the perspective of gross profit margin, technologically advanced is not necessarily a good business.

A survey of Silicon Valley's well-known venture capital A16Z found that global AI companies have one surprising thing in common: gross margins are too low, usually 50-60%, far lower than the 60-80% gross margin of SaaS business. IFLYTEK, for example, has a gross profit margin of about 46 per cent; ignoring the security business with the largest technology revenue, it was 59 per cent in the first half of 2019.

AI itself is a cost center. Computing power and labor account for "about 40%" of the cost. Wang Yong (pseudonym), CEO of a domestic CV company with a valuation of 1 billion yuan, told us that tens of millions of algorithms come first, and cloud service fees account for most of the expenditure. In addition, in order to ensure the high accuracy of the algorithm model, the workflow does not leave manual.

At present, in-depth learning is mainly focused on supervised learning, and data labeling is also an important work. According to the survey data of Alibaba Group, among the nearly 500000 artificial intelligence trainers in China, the vast majority are engaged in simple tasks such as data labeling.

For businesses that require particularly high precision, it is more at ease to do it yourself, Wang Yong explained. According to the prospectus, there are also 405 data-tagged employees, accounting for 17.2% of the company's total population, second only to the R & D team.

The cost of data sources is significantly higher than the cost of cloud services.

Source: Nakahashi Research: research on Market and Technology Trends in 2020-artificial Intelligence

The flip side of high-cost investment is the difficulty of business scale.

Most AI companies cut into segments (areas) are customized, and new data may be needed when deploying each new customer. When you write an algorithm to sell to An and then sell it to B, you need to retrain the model or write the interface according to the scenario of Company B. In particular, the marginal cases (corner case) of the two companies are often different, and the treatment of these marginal cases constitutes one of the most costly parts of the company.

"is it the essence of business that you are so tired that you can't get any money? "the head of a domestic machine vision start-up company once issued such a sigh.

Generally speaking, the production of delivery not only determines the delivery cycle, but also determines the gross profit margin of the enterprise. AI Vision's current business model is the traditional project-based model. Huang Yan, an investor in Shangtang and managing partner of CDH Innovation and growth Fund, once told Caixin that the project system means that side B (enterprise level) can default on its accounts and may kick out the company and find someone else to replace it. "how far can this model go? How big is it? "

AI essentially creates a new business type. in the view of some a16z investors, AI applications are like ordinary software and can be sold many times, but they need a large number of professionals to provide services each time, so that it is impossible to copy and scale traditional software with zero margin.

However, it is not impossible to effectively improve the situation. "the standardization of low-level tools can also reduce costs from another dimension," Yao Zhiqiang told the author. Many AI companies, including Yuncong Technology, are committed to the standardization of low-level tools.

Ignoring the basic productivity platform Brain++ of science and technology

For example, the basic productivity platform Brain++, which focuses on the painstaking efforts of science and technology, is like the assembly line invented by Ford, providing one-stop AI engineering solutions for R & D personnel. In addition to automating basic work such as data acquisition, cleaning and preprocessing, AutoML technology can also automate algorithm production.

The marked increase in the company's gross profit margin is also proof of the effectiveness of this effort. The data show that the share of data source costs of disregard technology has fallen sharply for three consecutive years from 2016 to 2018, and accordingly, the gross profit margin of disregard technology has risen to more than 60% in 2018.

At first, a large number of them are customized, but when the industry has three or five customers, there will be more and more products, or more and more reusable parts, Minghui Wu, CEO of Mingluo Technology, once explained to the media.

In spite of this, TO B business is very difficult to expand rapidly, especially the customer mentality is the most difficult to accelerate, Wang Yong said, "without a certain critical point, it is very difficult to push." Moreover, technical logic is not equal to business logic. For example, the direction of deep learning favors end-to-end learning, but black boxes are not good for business scenarios that require decisions.

Early private capital can cover up these inefficiencies in the short term. However, can long-term product or market optimization completely solve the problem? Some investors say this is not clear.

Get rid of "emptiness" to reality and gradually return to rationality

It's time to squeeze out the bubble.

A senior investor who has seen much of the primary and secondary market upside down sighs that the primary market values the withdrawal of IPO, which means both fame and wealth, but the secondary market may not pick up the red-hot potato. Science and Technology Innovation Board's choice of listed companies is still more cautious, the money-burning model, the valuation is too high, may not be able to pass.

"the primary market can make you run fast, but what really makes you take off is the secondary market. A middle manager of a listed company told Machine Heart.

The competition between enterprises is not a direct contest between enterprises, but a contest of their respective economic alliances (upstream and downstream). After entering the capital market, resources related to the capital market (such as brand building, upstream and downstream resources) will follow, thus promoting the faster development of enterprises.

Tencent and Google also achieved rapid growth after going public, Song Chunyu added.

Song Chunyu does not agree with those who pay too much attention to the company's financial indicators. "using a set of financial indicators to measure a tech startup company that has been established for a few years is very unreasonable, will stifle innovation, and is not in line with Science and Technology Innovation Board's original intention. "

60% to 70% of the US economic growth should be driven by the new economy, while the development of the US new economy is largely due to the developed venture capital industry and the Nasdaq market, as pointed out by Paul Krugman, the winner of the Nobel Prize in economics.

In order to support scientific and technological innovation enterprises, China was faced with a big basic deficiency, that is, there is no Chinese "Nasnak". Science and Technology Innovation Board's demand for a higher market capitalization is tantamount to replacing hard indicators with market games, indirectly attracting social capital and giving China the initiative in advanced technologies, such as semiconductors and artificial intelligence.

"if social capital does not come in, the money will still go to stocks or real estate of mature companies in traditional industries. "Song Chunyu said.

However, it is unknown how these companies that used to rely on financing will grow in the next few years. In 2019, fluent English had a net loss of nearly 600 million.

When it comes to profitability, "We don't have much to worry about," Yao said. The high investment situation of most companies in the early stage has improved recently. with the expansion of revenue, the relative cost will be reduced rapidly. "even without Science and Technology Innovation Board, the listing time would only be postponed for another two or three years. "

Although iFLYTEK, which is 20 years old, has been working hard on the path of commercialization, in Yao Zhiqiang's view, this does not represent the future of this batch of AI companies.

When iFLYTEK was founded, it had not yet reached the technical dividend period. However, the landing of AI needs to be combined with a lot of points, so the industry of iFLYTEK is a little heavier than the current AI company. Yao Zhiqiang explained that, for example, some acquisitions will be made in order to do education, and after being deeply bound to the industry, it is also more vulnerable to the development cycle of the industry.

In addition to resources and leverage, the company's listing also means that the competition with competitors has expanded from AI-based in the past to all-round commercial competition of products, brands and channels. At present, Science and Technology Innovation Board pricing has included the expectation of rapid growth of the enterprise, once the growth falls short of the expectation, the stock price will fall.

Any company can go public, but time will tell.

-end-

Https://www.toutiao.com/i6831899425103675916/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.