Written by | Wen Yehao
Edit | Wang Pan
As a leading enterprise in the AI industry, SenseTime is currently concerned by the market, and it is not too much to say that it is a "barometer" of the AI track. This makes its every move, whether it is financing or listing, to be talked about by the outside world.
The stock price fluctuations after the lifting of the ban on SenseTime's restricted shares have once again triggered questions and concerns about the AI track. Overnight, the Internet context has nailed the label of "charm" to the once-popular AI track.
In this context, SenseTime has made frequent moves, not only announcing the extension of the share lock-up period by the management on the day of the lifting of the ban, but also recently buying back 6.7 million shares for HK$14.07 million. It is worth noting that this is the first repurchase since SenseTime went public.
From the perspective of SenseTime, there are two motivations for the above actions, one is to eliminate market noise and alleviate the pessimism of the outside world on AI; The second is to show confidence in the value of the stock, business prospects, strategy and long-term development from the company level.
On the one hand, it is the "charm" of the Internet context, and on the other hand, it is the "self-proof" at the company level. In the current noisy market, who on both sides is more sober?
AI has no ghost stories
In fact, before the fluctuation of SenseTime's stock price, the market's doubts about the AI track have never stopped.
Before the large-scale IPO of AI players, the "bearers" mainly fired at the financial reports of star companies, accusing them of slowly getting rid of the quagmire of losses; With the listing of SenseTime and Cloud, the more intuitive stock price performance has instantly become the focus of doubt, whether it is Geling Deep Pupil or Cambrian, or Stock Exchange, which has fluctuated stock prices, has been pushed to the cusp of public opinion.
In the context of the market downturn and the general decline of Chinese stocks, why the public will be particularly strict with AI players may be related to the high expectations they carry.
Once upon a time, people's discussion of AI was no less than the current metaverse and Web 3.0, and at that time, the "AI substitution theory" that made the public anxious was endless. Today, in the face of the not-so-beautiful performance statements and stock price performance of AI companies, viewers have regarded AI as a "pig" that has fallen into the wind.
Is this really the case? Putting aside the interference caused by shallow information, trying to understand the deep logic behind it will find that the AI track actually does not have so many "ghost stories" to speak of.
Taking THE AI leader SenseTime as an example, first of all, it needs to be made clear that SenseTime has not yet achieved profitability, and the stock price fluctuations after the lifting of the ban are objective facts, since the "result" is beyond doubt, it is naturally the "cause" that has caused the public to misunderstand.
From the perspective of stock price, the fluctuation of SenseTime's stock price is actually based on the changes in the shares in circulation before and after the ban period, before the lifting of the ban, the free circulation of shares accounted for less than 2% of the total share capital, and the total amount of shares lifted involved 25.287 billion shares, accounting for 75.91% of its total share capital. Obviously, this is not the collapse caused by the bursting of the "bubble", but the result of the reduction of volume and liquidity, which is expected.
Secondly, in response to the clichéd problem of losses, the market also has a misunderstanding of SenseTime. Looking at the amount of losses alone, SenseTime lost 4.96 billion yuan, 12.16 billion yuan and 17.14 billion yuan respectively from 2019 to 2021, but it is necessary to take into account the abnormal disturbances caused by fair value changes, and the so-called large losses are largely due to the operation of financial norms, rather than the actual business losses of SenseTime. Excluding fair value changes, SenseTime's adjusted net loss for 2021 was only $1.42 billion.
Once we understand the above two layers of logic, we will find that the industry cloud hanging over SenseTime is actually not terrible. In fact, in the face of doubts about the context of the Internet, investment institutions have come to the opposite conclusion.
After the lifting of the ban on restricted shares, in addition to SenseTime's own announcement of repurchases, HSBC, Everbright and other institutions gave SenseTime an overweight rating, and early investors such as Zhongping Capital and CDH Investment also spoke out in support, indicating their recognition of SenseTime's long-term development prospects. Zhongping Capital has pointed out that SenseTime's stock price fluctuations provide a rare entry opportunity for long-term investors in the field of artificial intelligence.
At the moment when negative emotions are full of emotions, this may be interpreted as an institutional group for SenseTime, but if you analyze the transformation of SenseTime's business, the prediction and rating of investment institutions are not "empty", but there are traces to follow - washing away the "flashy" SenseTime is on the road to rebuilding AI confidence.
Why Sindh Corpus proves itself
In the context of SenseTime's repurchase, institutions giving overweight ratings, and investors' optimistic statements, assessing whether SenseTime is "self-evident" successful, it seems necessary to answer the two ancient problems of the AI industry - financial performance and commercialization process.
In terms of performance, in 2021, SenseTime's revenue was 4.7 billion yuan, up 36.39% year-on-year; Gross profit reached RMB3.278 billion, up 34.77% year-on-year, continuing to maintain its growth trend.
However, the growth of financial data is only the result, and the business-level de-virtualization is the main reason, which is also the core logic of rebuilding AI confidence.
Looking back at the past decade of the domestic AI track, the commercialization problem is undoubtedly a mountain on the path of industrial development, which is also a common problem that existed in the previous AI companies such as SenseTime that were born out of the laboratory: technology ran in front of commercialization, but it is difficult to adapt to the industry, and commercial exports are slightly single.
Taking the four head companies of the AI track as an example, the four players all started from the CV (computer vision) track, and it is inevitable that some people will joke that AI manufacturers are still "camera" business. Ridicule is ridicule, if you really want to delve deeper, there are many CV direction subdivision scenes, and the two scenes of face recognition and medical image analysis alone are enough to break the homogenization filter, and it can only be summarized by the word "camera".
Behind the ridicule, perhaps out of the outside world still retains the stereotype that the AI industry is "addicted to scientific research and ignores the landing", after all, in the public perception, AI is still a "castle in the air", and its life scene is still separated. As everyone knows, the current AI track has become more diverse, quietly distributed in various well-known fields.
For example, the small red book that mainly "marks my life", compared with the UGC content community such as Zhihu and Tieba, the small red book that focuses more on the mixed output of graphics and text has significant advantages in photo presentation and beautification, which also makes it a "grass planting" platform loved by users.
On the surface, Xiaohongshu and AI can not see any relationship, in fact, Xiaohongshu users will use key point tracking, advanced beauty, sticker effects and other technologies whenever they edit content through their rich graphic creation tools.
For example, in the industrial field, from the intelligent management of industrial infrastructure such as railways and power grids, to the product defect detection of assembly lines, and the production safety protection of production line workers, the application of SenseTime's AI industrial engine can be seen everywhere.
In the field of agriculture, which is related to the issue of food and clothing, the value of AI is gradually being released. Taking wheat production monitoring scenarios as an example, the remote sensing imaging technology commonly used in the past can easily confuse wheat with other crops, and the monitoring efficiency is also very low. SenseTime combines its own AI capabilities with remote sensing to not only accurately identify wheat distribution, but also accurately monitor wheat growth changes, thereby improving the efficiency of monitoring and governance.
The same logic is also reflected in the two protagonists of SenseTime's new business line, smart cars and meta-universes. Both are the hottest tracks at the moment, which inevitably makes people wonder whether there is a "rubbing concept", but with the landing of the scenes at all ends of TheRmotime, the suspicion of the outside world is becoming less and less.
In fact, although SenseTime Intelligent Vehicle Business Group was only established in the first quarter of this year, SenseAuto SenseTime's smart car platform has been polished for 5 years, and in 2021 alone, it has established cooperative relations with more than 30 car companies.
Among Honda China's two pure electric debut models, there is no shortage of SenseTime. It is reported that the "DMC driver status perception system" equipped with it can sense the driver's fatigue driving, driving distraction and other behaviors, and through the human-vehicle interaction system timely early warning, effectively avoid the dangerous situation on the way. The system is based on the AI perception algorithm capability of SenseAuto SenseTime's Smart Car Platform.
If it is said that the promotion of the intelligent car business is SenseTime's grasp of the current outlet, then betting on a longer-term and more tortuous metacosm is SenseTime's growth logic based on long-term thinking.
However, compared with the smart cars running on the road, the metaverse is still in the embryonic stage, as SenseTime CEO Xu Li said: "In the metaverse, it is more difficult for us to find the so-called answer." ”
In the face of great uncertainty, SenseTime's choice of play is a multi-line layout. In terms of cultural creativity, SenseTime has reshaped Art Collections such as Xu Beihong's famous work "Song Ren Pi Long Whistling Words" into digital cultural and creative products through AI+AR technology, and put them on the shelves of its Digital Cat Cultural and Creative Platform; In the technical field, SenseTime has successively entered the fields of virtual people, offline AR/MR, and broken the boundary between virtual and real through the capabilities of the SenseMARS Mars mixed reality platform.
In many meta-universe scenes, The SenseTime AR minibus, which blends with smart cars, is playing a more important role. Although AR is not a new thing, but limited by technology and cost, the previous AR applications are often biased towards the experimental nature, not only in the content is too rough, the product side also lacks practical value.
SenseTime's AR minibus is not the case, which uses the window as the screen to combine a variety of pictures such as aerial photography outside the car and the video screen in the car into a visual spectacle of virtual reality, and passengers only need to be in the car to have a panoramic view of various urban cultural landscapes, realizing the balance between the immersion and functionality of AR technology.
Whether it is the DMC driver state perception system or the AR minibus, it is not difficult to glimpse the transformation of SenseTime's AI philosophy - upgrading one paper after another into one application scenario after another.
Based on this, SenseTime's march towards smart cars and meta-universes may not be aimlessly chasing the wind as the outside world speculates, but the industry answer given after fully thinking about "what AI can do". This also means that the current SenseTime technology has stepped out of the "laboratory" and evolved into a mature AI company.
Young AI still needs patience
Although SenseTime has "self-proved" its success through diversified landing cases, the AI industry will still face a long road.
To take a simple example, although AR minibuses can autonomously connect passengers, their automatic driving is limited to a fixed route within the demarcated range, which does not meet the ultimate expectations of autonomous driving in the Internet context. In other words, although the value of AI is certain, there is still a gap between the AI industry and the ultimate goal of the general public at this stage.
However, unlike the "business inspiration" that burst out overnight, the evolution of the technology field requires money and time from beginning to end.
According to gartner's emerging technology maturity curve theory, emerging technologies need to go through five stages from birth to landing, such as budding period, expected expansion period, bubble bursting trough period, steady recovery period and production maturity period, and this process is particularly long. This is also why 40 years in time and space, the bionic rachel in Blade Runner can only survive in science fiction movies.
In addition to the slow technological process, the AI industry is also facing a "commercial bottleneck", the bottleneck mentioned here is not that manufacturers can not find commercial exports, but the industry is based on the contradiction between cost and benefit, and is temporarily unwilling to pay for "uneconomical" solutions. After all, for small and medium-sized enterprise owners, the most concerned thing is not how advanced the technology is, but whether they can only spend a small amount of money and do big things.
Obviously, the current AI application is difficult to meet the expectations of bosses, otherwise there will be no "AI disenchantment". The logic behind this is that AI vendors are difficult to meet the different commercialization needs of various industries and companies.
Taking the CV scenario as an example, due to the lack of ability of AI manufacturers to mass-produce highly available AI models, massive data annotation cleaning still needs to be done manually, which is like a "labor-intensive industry", resulting in high solution costs and difficulty in entering the wider market. This can be seen from the financial reports of various AI manufacturers - the ToG business has always occupied half of the country, and the long-tail scene is difficult to get on the table.
But to put it another way, once a player has a "killer technology" with availability and economy, they can break through the current shackles. The purpose of the high investment of a number of AI companies is this, after all, who can take the lead in walking out of the "black barrier area", who can eat crabs.
Looking around the entire AI track, the first players to eat crabs must be the most well-laid out manufacturers at the moment, and SenseTime is undoubtedly one of them. The reason is that SenseTime has found the right path for AI landing - platformization.
As early as two years ago, SenseTime held high the banner of "platformization" and looked for an outlet to get rid of the manpower-intensive status quo in the industry. Today, SenseTime has built a SenseCore AI device covering the computing power layer, algorithm layer and platform layer, and implemented the strategy into practical application areas.
It is reported that SenseTime AI large device has a complete, large-scale algorithm production capacity, based on this, the cost of AI solutions has decreased significantly, which in turn enables enterprises in all walks of life to deploy AI under the condition of limited budget to achieve large-scale landing of AI.
Among the many cases, the performance of SenseTime AI in the "StarCraft 2" game is worth mentioning.
Due to the special game mechanics, the confrontation intensity of "StarCraft 2" has always been a "ceiling" in many games, and the ai technology in the past can only barely reach the level of ordinary players through cheating - the "alpha dog" who defeated humans in the Go project, and even the "simple human machine" built into the game when first exploring "StarCraft 2".
Through the AI device, SenseTime's AI model DI-star completed about 100 million rounds of battles in only 5 weeks, and finally showed its micro-manipulation and decision-making ability that was not inferior to that of first-line professional players in the game with the top professional players of human beings. Such decision-making intelligence can play a miraculous role in a large number of decision-making AI application scenarios in the field of intelligent transportation in the future.
The data shows that based on ai large devices, the average number of models produced by each R&D personnel of SenseTime has increased sharply from 0.44 in 2019 to 7.96. It can be seen that SenseTime's AI large device has achieved a double improvement in the production efficiency and usability of AI models.
The persistence of long-termism, the transformation of AI commercialization from virtual to real, and the unremitting investment in AI infrastructure to finally achieve the industrial mass production of AI models, etc., all show that the current SenseTime is no longer the stunned youth of the technology "naked swimming" in the Internet context. On the contrary, it does not hesitate to use profits to pay for technological progress, and it is gradually entering the "harvest season". The industry and the market, while looking at players on the AI road in multiple dimensions, should also give them more time.