For SenseTime, whose business model has not yet been connected, AI technology may only be chasing hot spots and enhancing brand awareness when it goes to the Olympic Games, and the implementation and commercialization process of its application scenarios may not bring substantial changes.

@新熵 Original

Author丨Yan Cao Editor丨Jiuli

The 2024 Paris Olympics with the blessing of AI black technology is romantic and fashionable. Major model manufacturers are also taking advantage of the hot spots to show their muscles and roll up crazy inside and outside the arena.

In the direction of involution, on the one hand, there are large-scale model products represented by chatGPT, Wenxin Yiyan, and Byte Doubao, which provide C-end users with richer services before, during and after the game, such as hot sports agents, AI search, AI interpretation, etc. On the other hand, large model manufacturers represented by Alibaba, with the help of past technology and ecology, are oriented to the B-end market, and export AI large models and other related technologies in terms of game broadcasting, athlete training and digital communication (digital human anchors), etc., and have made their presence felt.

As one of the four AI tigers, SenseTime did not miss this event, but brought a new model to help the event. According to media reports, SenseTime's AI smart basketball products will participate in the journey of the Chinese national basketball team throughout the event, providing sports data analysis and competitive strategy optimization support. However, compared with the AI technology strength and brand voice of leading technology companies, the AI model of SenseTime, which is slightly less influential, is what is the quality? This global event may be a litmus test.

01

SenseTime participated in the "involution" of the Olympic Games, but it was difficult to hide the shortcomings of commercialization

According to media reports, SenseTime's new large-scale model and application mainly provide technical support for Chinese athletes from training and preparation, event data analysis, to smart broadcasting, and improving the viewing experience.

For example, SenseTime's AI smart basketball product will participate in the entire journey of the Chinese national basketball team, which can realize real-time motion capture and analysis, present multi-dimensional sports performance data, and provide scientific and effective decision-making support for athletes' training and competition.

SenseTime's smart event broadcasting technology will be used for the broadcast of table tennis and archery, providing audiences with an immersive viewing experience. Based on the InnoMotion event broadcast solution jointly developed by SenseTime and SMT, SenseTime's 3D non-perceptual motion capture technology can be used to achieve multi-person, large-scale, and multi-angle multi-scene coverage without wearable devices, and obtain spatial motion and attitude information in real time.

The unveiling of SenseTime's AI model at the Olympic Games has also brought the spotlight of the media to this AI giant again. However, while some netizens cheered for domestic AI manufacturers to go abroad, some netizens believed that this was just SenseTime's habitual "hot spots".

Because for a long time, problems such as the lack of AI application scenarios, insufficient competitiveness of computing power, immature technology, and unconnected business models are still in front of SenseTime. The actual value of SenseTime's trip to the Paris Olympics is also very doubtful.

First of all, the commercialization difficulties of the AI industry have not been completely solved because of the advent of the era of large models. On the one hand, although general-purpose and platform-based AI such as Baidu and iFLYTEK have gradually started their commercialization journey with the help of their original products and ecosystems, their application scenarios are still very limited, so the revenue scale and user data brought by them are not ideal.

On the other hand, the general-purpose and platform-based AI pattern has been initially determined, so that the vertical AI large model represented by SenseTime has been squeezed in terms of living space and even lacks a sense of existence. The most obvious change is that SenseTime, which also tried to follow the platform model in the past, was forced to return to the road of customized development due to the pressure of commercialization.

In this regard, the market believes that it is essentially the same as a traditional development outsourcing company. Because it is up to the customer to customize the demand scenario, it means that they are all very personalized and segmented long-tail needs, rather than general needs. This also leads to high development costs, difficult delivery, and long account cycles, which affect cash flow.

At the same time, because the demand is too long-tailed, the products developed according to customer needs will not be reused by other customers and scenarios, affecting the standardization and scale of its products, thereby limiting its future development space and imagination in the secondary capital market.

Secondly, the biggest cost of deploying AI technology is investment in technology research and development and computing power reserves. As it turned out, SenseTime paid a high cost as a result. According to the 2023 annual report released by SenseTime, the company achieved revenue of 3.406 billion yuan last year, a year-on-year decrease of 10.57%, and an annual loss of 6.495 billion yuan, an increase in the amount of loss compared with the previous year.

According to the statistics of previous financial reports, from 2018 to 2023, SenseTime Technology achieved losses of 3.433 billion yuan, 4.968 billion yuan, 12.16 billion yuan, 17.18 billion yuan, 6.093 billion yuan and 6.495 billion yuan respectively, with a total loss of 50.324 billion yuan in six years.

Under the boom of AI large models, SenseTime's revenue in generative AI has also risen, with its performance contribution increasing from 395 million yuan in 2022 to 1.184 billion yuan in 2023, an increase of 199.9%. Although the performance contribution of this segment has increased significantly, it is a drop in the bucket compared to the scale of its losses in recent years. This also means that it may be difficult for AI models to become a magic weapon for turning losses into profits. It is not a lifesaver for SenseTime to get out of the commercialization dilemma.

According to analysis, SenseTime is currently monetizing through customer demand customization. However, it is not willing to give up on its idea of platform-based AI. However, from the perspective of AI technology, products, and industry ecology, SenseTime may not have the conditions to build an AI large model platform.

Because the application of large AI models requires not only computing power, but also rich application scenarios, corresponding products and even industry ecosystems. Baidu, for example, in the past two years, Baidu has reconstructed all the products in its mobile ecosystem with the help of AI models, which has brought rich returns. Taking Baidu's first-quarter financial report as an example, with the support of the AI model, although its revenue was the same as the same period last year; However, non-IAS net profit increased by 22% year-on-year.

On the other hand, SenseTime has not changed its loss-making status because of the AI model, so even though the proportion of its revenue contribution is increasing significantly, its business model is difficult to say that it is going through.

02

Hoarding computing power is a huge investment, betting on large models or a double-edged sword

As we all know, the "100 model war" in recent years has further intensified the involution of the industry. As the second echelon of the industry, SenseTime is facing a severe test. Especially in terms of computing power reserves, major model manufacturers can do their best. SenseTime Technology, which is already lacking in strength, naturally does not dare to be left behind.

As a result, SenseTime has invested heavily in computing power, and its gross profit margin has been damaged, which has further affected its large-scale profitability. According to media reports, in 2023, when the momentum of AI large models is in full swing, some experts said that SenseTime has rich experience in parallel training of kilocalories in the past five years, and its single-task training can mobilize 4,000 A100 GPU cards at the same time, with an equivalent computing power of 10,000 A100, which has reached the threshold of training GPT3 or even GPT4.

However, this does not mean that the technical level of SenseTime is very good. On the contrary, because AI giants are running wild in "hoarding computing power", and SenseTime can only become a follower of the top players.

For example, Baidu has the largest single intelligent computing center in Asia, Yangquan Intelligent Computing Center, with a computing power scale of 4EFLOPS (40 billion floating point calculations per second), a construction area of about 120,000 square meters, and a Shenyang Intelligent Computing Center. iFLYTEK has joined hands with Huawei to build an independent and controllable large-scale model computing power base based on the Ascend ecosystem, the Feixing-1 platform. And Internet giants such as Ali and Tencent are all investing heavily in computing power without exception.

In addition to the huge cost of computing power reserves, SenseTime's AI technology and application are not focused enough and it is difficult to achieve large-scale growth, which is also its fatal injury.

At a time when its commercialization capabilities are insufficient and technology applications have not yet been transformed into commercial value, SenseTime keeps up with the pace of large manufacturers with relatively mature business models and invests heavily in computing power, which will inevitably bring huge trial and error costs and even irretrievable sunk costs. This dilemma of SenseTime can be seen through its continuous pursuit of hot spots and even the transformation of the track of AI technology application.

It is understood that since the establishment of SenseTime, its core products include AI large devices, Ririxin large model series, and smart car brand SenseAuto, and its core technology is face recognition, image recognition, text recognition, medical image recognition, video and other AI vision technologiesHowever, in terms of product commercialization and technology, there has always been a sense of going around and chasing hot spots.

For example, as early as 2021, SenseTime went public with its forward-looking layout on the concept of "metaverse", which once became the focus of the market. At the time of the IPO, SenseTime's market capitalization was as high as HK$140 billion, with a maximum market value of HK$300 billion. However, it turns out that after the "outlet" of the metaverse that has been popular by the media, the concept of the metaverse of SenseTime Technology is only a flash in the pan.

In addition, in addition to not missing the Olympic Games, SenseTime also showed its "AR smart bus" equipped with its L4 autonomous driving solution at the 2023 Hangzhou Asian Games, showing off its technical muscles and attracting a lot of attention.

It is understood that the founder team of SenseTime Technology that came out of a laboratory of the Chinese University of Hong Kong naturally carries technical genes, so it has also attracted the blessing of many star capitals, plus SenseTime Technology is also very good at telling stories with the help of the wind, so it has also made some capital see its grand vision.

According to incomplete statistics, before SenseTime went public, the company received a total of 12 rounds of financing, with a cumulative financing amount of up to 5.22 billion US dollars.

It's just that SenseTime, which blindly uses the tuyere to tell stories, has hit walls everywhere in the actual implementation of its AI technology. Although from the perspective of its business layout, SenseTime Technology is evenly exposed to rain and dew in smart cities, new energy vehicles, medical health, international sports events and other industries, but from the perspective of its revenue structure, smart business and smart city are the two main sources of income for SenseTime.

Moreover, these tracks have also been attacked by giants such as Alibaba, Tencent, ByteDance, and Baidu. Even the security track can only survive in the shadow of industry giants such as Hikvision.

In fact, compared with AI giants such as Baidu, Alibaba, and iFLYTEK, they have quickly realized the commercialization of generative AI by relying on their original business fundamentals. However, when SenseTime Technology was going around and couldn't make a profit, some capital began to lose confidence and patience in it.

For example, according to media reports, in 2023, SoftBank, as an important shareholder of SenseTime, will continue to reduce its holdings, and even until a few days before the release of its financial report, it will continue to reduce its holdings; As for Alibaba, in April 2023, Ali reduced its holdings of 40 million shares of SenseTime, with a cash amount of more than HK $280 million. On June 21, Alibaba sold 50 million SenseTime shares at an average price of HK$2.2670 per share due to its own business adjustments.

It can be seen that SenseTime's bet on AI large models may be its last opportunity to find a commercial landing scenario. It's just that SenseTime, which has invested a huge amount of money, can rely on AI large models to turn things around against the wind in the current state that the "revenue generation scale" of large models is very limited?

03

Large-scale model manufacturers bet on the sports track, or opportunities and challenges coexist

In fact, SenseTime is not the only one betting on the sports track through AI large models. For example, in June 2024, Shanghai Sport University released the first sports AI model: the "Shangti Sports Model". According to reports, the product is mainly used in sports literature, action recognition and technical and tactical analysis, allowing people to see the possibility of AI application in the sports industry.

But given that the industry is just getting started, there may be both opportunities and challenges. Specifically, before the advent of the first sports AI model, many AI technologies have been applied in the world's top competitions, and have played a positive role in the level of athletes, the efficiency and accuracy of referees' decision-making.

For instance, in January 2024, the NBA invested in six start-ups, namely avatar production studio Dimension Studio, motion analytics company Reboot Motion, data tracking company Tonsser, and AI analytics and scheduling platforms Fastbreak.ai, Plantiga, and TextQL. Among the 6 companies, 5 focus on AI technology and 1 develops virtual technology.

At the same time, in sports technology companies such as Keep, some somatosensory sports + home fitness application scenarios have also made AI and large models and other technologies useful, and some companies are also actively developing and promoting artificial intelligence digital sports projects.

According to the "2022-2028 Sports Intelligence Development Trend Forecast" produced by PricewaterhouseCoopers, China's sports intelligence market will develop rapidly in the next few years, and the overall market share will reach the level of 100 billion.

Although the advent of the AI era has injected impetus into the vigorous development of sports. However, it cannot be ignored that AI technology is also a double-edged sword, which may jeopardize the fairness and justice of sports competitions, and even shake the identity and status of athletes. It can even be used by criminals to falsify the results of competitions, thus deviating the sports industry from its essence.

For example, during the Qatar World Cup, SAOT, a semi-automatic offside recognition technology derived from the VAR system, caused controversy, such as "leg hair offside", "hair level goal" and other ridiculous complaints. Moreover, the addition of technology will also challenge the principle of fairness adhered to in sports competitions, and may even turn sports competitions into "off-field competitions" to compete for scientific and technological strength, gradually blurring the identity of athletes.

In addition, the emergence of AI-mediated reports and various fake videos is also controversial for the application of AI in the sports track. For example, on February 19, Messi clarified on Weibo that "Miami International did not appear on the Hong Kong China tour", but then the video was questioned, and it may have been used to reshape it with the help of AI, which caused a stir.

It can be seen that the progress of AI technology has a positive impact on the development of the industry, but also hides many risks. For SenseTime, which has not yet had a business model, this rush to the Olympic Games may still only be chasing hot spots and enhancing brand awareness, and it may not bring substantial changes to the landing and commercialization process of its application scenarios.

epilogue

As one of the four AI tigers, SenseTime originally had excellent AI technology and a core technical team. However, in the early years of the AI industry, it was difficult to find a variety of application scenarios, and business models could only be found in popular scenarios such as face recognition.

As the industry technology becomes more and more mature, although the advent of the era of AI large models seems to bring opportunities to SenseTime, because the application of general AI large models is becoming more and more homogeneous, and the vertical large models lack sufficient application scenarios, technology continues to look for business opportunities in various vertical tracks.

It's a pity that in the process of going around, SenseTime has spent huge computing power reserves and technology research and development funds, but has not found a solution that can be mass-produced and standardized, and can only make profits through the long-tail customization needs of customers. It is precisely because of this that its brand value is very limited, and its stock price has been reduced to penny stocks for a time.

After the tide passed, it was only known who was swimming naked. When SenseTime sounds like a "high-level" AI large model application, it can only follow the hot spots, and the AI application of this Olympic Games may not have much practical value for SenseTime. It is foreseeable that when the heat subsides, the commercialization of SenseTime's AI will inevitably be another chicken feather.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com