Ali clears Shangtang, the large model is not an AI "talisman"

Source: "Shenmou Finance" (ID: chutou0325), author: Zhang Wei

Image source: Generated by Unbounded AI

The "Gold Swallowing Beast" who has no expected profit plan has lost his support, and capital no longer trusts SenseTime.

After five years of investment and several reductions in holdings, Ali finally cleared the AI giant SenseTime.

On July 21, SenseTime-W, a Hong Kong-listed company, announced that it had received notice from Taobao Holding Limited, a wholly-owned subsidiary of Alibaba, that it had sold all Class B shares of SenseTime in an orderly manner. This means that Ali chose to withdraw from Shangtang Technology, which had invested for many years.

Even though SenseTime stated after reducing its holdings, the company’s business, operations, and business cooperation arrangements with partners have not been affected. But the announcement couldn't save investors' lost confidence and falling stock prices.

** On the day when Ali’s shareholding reduction announcement was released, SenseTime closed at 1.72 Hong Kong dollars per share. The stock price has fallen by 22.52% this year, and the latest total market value is only 57.6 billion Hong Kong dollars. **

In fact, Shang Tang lost more than Ali.

In May 2022, Softbank Group, the second largest shareholder of SenseTime, began to reduce its holdings and cash out. In less than a year, Softbank Group has reduced its holdings of SenseTime three times, and its shareholding ratio has also dropped from 18.02% to 15.99%; , March, and April reduced its holdings of SenseTime stocks several times, and its shareholding ratio dropped from 17.97% to 13.95%.

In the first half of this year, SenseTime took advantage of the large-scale model and relied on the "SenseTime SenseNova" large-scale model. In April, the stock price rebounded to 3.7 Hong Kong dollars per share. Although it is still far from the peak of 9.7 Hong Kong dollars per share, it is still showing positive signs.

Why did SenseTime lose the trust of capital?

01 Nothing more than to make money

From a macro point of view, it is normal logic for capital to re-evaluate risks and withdraw after the end of the Fed’s interest rate hike cycle. But from a microscopic point of view, SenseTime, which has lost 43 billion in five years, also has its own reasons.

From the perspective of the general environment, it is normal for SenseTime to encounter capital reductions. This year’s AI concept stocks are not very easy. For example, leading concept stocks such as 360, Cambridge Technology, and Kunlun Wanwei have also experienced reductions in their holdings.

The reduction of AI concept stocks is closely related to the world economy. When the Fed’s interest rate hike cycle draws to a close, the valuation of technology companies will face revaluation.

In 2023, the interest rate hike cycle is coming to an end, and it hits the popularity of ChatGPT again, and native AI companies such as SenseTime, HKUST Xunfei, and Cambrian will soar. However, in just a few months, investors have differentiated. Due to the complexity and risk aversion of the market in the past two years, investment institutions have become more cautious. Industries with high investment and long-termism like AI have made investment institutions fearful. , so companies with weak profitability are more cautious.

** Weak profitability is a deep-rooted problem of Shangtang, and it is also the key to losing the trust of capital. **

Since 2019, the annual reports submitted by SenseTime have not been "passed" every year, and they are all in a state of loss. In 2019, data showed that in 2021, SenseTime Group achieved operating income of 4.7 billion yuan, a year-on-year increase of 36.4%, and an adjusted loss of 1.42 billion yuan.

**According to public data, since 2018, SenseTime has accumulated losses of 43.8 billion yuan. **

The money thrown out and the return that cannot be seen make capital worry about the future growth of SenseTime.

From 2018 to 2021, SenseTime’s research and development expenses were 849 million yuan, 1.916 billion yuan, 2.454 billion yuan and 3.614 billion yuan respectively, and the proportion of research and development expenses in the current total revenue was 45.9%, 63.8%, 71.3% and 76.89%. In 2022, SenseTime's R&D expenditure will further increase to 4.014 billion yuan, a year-on-year increase of 11.07%, and the R&D expense ratio will reach 105.38%.

Looking at the combination of R&D and revenue, SenseTime’s total revenue will decrease in 2022, but R&D expenses will still increase.

In 2022, the total operating income of SenseTime will be 3.809 billion, a year-on-year decrease of 18.97%, and the R&D expenditure will be 414 million, a year-on-year increase of 11.07%. SenseTime's R&D investment is getting higher and higher, but the returns are getting less and less. After five years of burning money, the revenue has not improved. How can capital be in a hurry.

** Losses are one thing, and the wobbly ** business is another.

The volatility of SenseTime is manifested in the fact that it follows wherever there is an outlet. In terms of the company's development strategy, SenseTime has made many attempts, from TOB to TOC, TOG, actively developing customers, from smart medical care to smart cities, and SenseTime has included all the current AI hot tracks.

However, except for the smart city, SenseTime’s newly developed business can be regarded as a taste, and it has little advantage in the industry track. The overly complicated business also makes SenseTime waste a lot of time. In the AI industry that is racing against time, SenseTime It can be said that the gain outweighs the gain.

Specifically, SenseTime has four major business segments: Smart City, Smart Life, Smart Car, and Smart Business. Previously, nearly 90% of SenseTime's revenue came from smart commerce and smart cities.

Compared with 2021, SenseTime's business structure in 2022 has undergone significant changes. The revenue of smart business and smart city, which SenseTime relied on before, has declined, while the share of smart life and smart cars has risen sharply.

SenseTime's commercial customers, single customer revenue, and business revenue have all declined, and business revenue has decreased by 25.2% year-on-year. In the smart city sector, although the number of service customers increased by 15.7% year-on-year, the revenue of this business fell by 48.8% year-on-year.

Against the background of the decline in the main business, although the two sectors of smart life and smart cars did not support the revenue of SenseTime as a whole, they maintained positive growth.

As the penetration rate of new energy vehicles continues to rise, intelligent configuration is becoming one of the key areas for differentiated competition among auto companies, and SenseTime's auto business has begun to grow. In 2022, SenseTime's smart car segment will achieve revenue of 300 million yuan, a year-on-year increase of 59%, and its proportion of the group's total revenue will increase from 4% last year to 8%.

The smart life business is the fastest-growing business of SenseTime.

Judging from the financial report, the increase in revenue of SenseTime Smart Life is mainly due to the new business of AIGC application. The smart life sector to which the AIGC application belongs achieved a year-on-year increase of 129.9% in revenue to 960 million yuan last year, accounting for 25.1% of the total revenue in the same period, an increase of 16.3% from 2021.

02 The AI "Four Little Dragons" who share the same disease

"If no one asks you under the ten-year window, you will become famous in one fell swoop and the world knows it." This common saying sums up the status quo of the AI industry.

AI not only has a high investment cost and a long cycle, but also follows the business principle of "fast fish eat slow fish". The research and development results may be wasted if one step is slow. Besides, the other "Three Little Dragons" were silent.

As for the plight of the Four Tigers, the reason is mainly related to the bottleneck of commercialization, or to put it more bluntly, Yuncong, Megvii, and Yitu have the same problem as SenseTime—insufficient liquidity.

Cloudwalk Technology’s total revenue in 2022 is 526 million yuan, a decrease of 51.06% over the same period; the net profit loss attributable to shareholders of listed companies is 869 million yuan, an increase of 237 million yuan from the previous year’s 632 million yuan.

Although Megvii and Yitu have not released recent data, they have lost a lot from past financial reports. The latest prospectus of Megvii Technology shows that as of the first half of 2021, the cumulative loss exceeded 16.5 billion yuan; the financial information of Yitu Technology’s previous prospectus was updated to the first half of 2020, and the cumulative loss during the three-and-a-half year reporting period Over 7.2 billion yuan.

Limited by the dilemma of high investment and difficult profitability, government support is often the best way for AI companies and the computer vision industry to run smoothly. Therefore, in the early days, the "four little dragons" of AI mostly focused on ToG-end business. For example, the smart city business with a large proportion of SenseTime, including City Ark SenseFoundry products, etc., as well as Yuncong's digital city-related business.

According to the statistics of Qianzhan Industry Research Institute, the demand for domestic computer vision services mainly comes from several industries such as security, government affairs, manufacturing, and retail, and the income of AI companies mainly comes from providing intelligent image monitoring, equipment operation and maintenance and testing for enterprises in the above industries Serve.

However, the ToG end market is limited, and the business cycle is long, not to mention the fierce competition in smart city, smart security and other businesses, and the market concentration is relatively high.

In 2022, SenseTime (market share 23.1%) tops the list for five consecutive years, followed by Hikvision (8.3%), Chuangxinqizhi (8.1%), Megvii Technology (8.0%), CloudWalk Technology ( 3.3%), Smart Eye Technology (2.2%). The top six players in China's computer vision AI market account for 53% of the overall market.

In the past two years, the growth in demand from the government and retail industries has been lower than expected, which has directly affected the smart governance and smart city business income of the AI Four Tigers. For example, the scale of the computer vision industry has also begun to shrink. According to the "China Artificial Intelligence Software Market Share in 2022" report released by IDC, the computer vision AI market scale is 12.30 billion yuan, a year-on-year decrease of 19.1%, becoming one of the market segments with the largest decline.

In sharp contrast to this, the to C business is obviously insufficient, and this directly limits the increment of business, making it difficult to maintain sustainable development, and the controllability of the business is weak, and it is easily affected by factors such as policies.

Like the AI chess-playing robot launched by SenseTime — “Yuanluobo”, a chess-playing robot. As SenseTime's first consumer-oriented product, its sales volume did not meet expectations. At the beginning of its release, Tmall pre-orders were only more than 200.

03 The large model is not an AI "talisman"

Judging from the trend of SenseTime this year, commercialization around large models is the main strategic plan.

The birth of ChatGPT 3 has triggered a wave of companies competing for large models. As a native AI company, SenseTime is one of the fastest-moving companies. On April 10, SenseTime released the basic large model "SenseNova" at the technical exchange meeting, relying on the large AI device SenseCore to realize the research and development system of "large model + large computing power".

The same is true for Megvii and Cloudwalk. Large-scale models are the key development strategy.

On February 10, Zhang Xiangyu, head of basic scientific research at the Megvii Research Institute, said that Megvii has opened up relevant underlying technologies and focused on four directions: the general image model, the video understanding model, the computational photography model, and the autonomous driving perception model. layout. On May 18, Yuncong released the "Calm" large-scale model product.

Yitu Technology has stayed away from this wave of "big model craze". Since the failure of the listing, Yitu Technology has been silent.

Yuncong, who has insufficient blood, chose to raise money to build a large model. Zhou Xi, chairman and general manager of Yunwalk Technology, believes that as long as the investment is in place, the computing power threshold is not insurmountable, "we must invest 100 to 200 million to do this."

Before the release of the "calm" model, Yuncong Technology threw out a targeted capital increase plan. According to the announcement, the funds raised for the research and development project of the "Industry Elf" large-scale model will not exceed 3.635 billion yuan.

Compared with other manufacturers, CloudWalk entered the market late. When various industry models have emerged in the market, CloudWalk's project has just begun preparations.

Recently, CloudWalk Technology announced the launch of the CloudWalk Pedestrian Basic Model, which uses more than 2 billion data. At present, this technology has been widely used in the security deployment and control of mines, construction sites, and special places, and monitors abnormal and irregular behaviors such as the wearing compliance and posture behavior of operators.

In addition, SenseTime has launched a large language model "SenseChat" that benchmarks against ChatGPT, an AI text creation platform "SenseMirage", and an AI digital human video generation platform "SenseAvatar". In June, SenseTime, together with the Shanghai AI Lab, the Chinese University of Hong Kong, Fudan University, etc., released a large language model with hundreds of billions of parameters, "Scholar Puyu" (InternLM).

For Shangtang, which is still "bleeding", it is more worth discussing a question before it: Can the large model alleviate the embarrassment of Shangtang's urgent need to make profits?

In fact, SenseTime prefers to use large models to empower its four business segments, rather than compete with the first echelon for large models.

This is also reflected in SenseTime's financial report. In the 2022 financial report, SenseTime mentioned that "in the future, we will use SenseCore's large device and large model capabilities to upgrade the four major sectors of smart life, smart cars, smart commerce, and smart cities. Through Make full use of the capabilities of the large model, continuously upgrade and strengthen the product matrix of the four major business lines, and improve product competitiveness.”

Under the background that the large-scale model bubble dissipates and returns to rationality, and SenseTime is in urgent need of capital, it is obviously unrealistic to compete with large-scale Internet companies on large-scale models. So can Shangtang be sure to make money?

Judging from the financial report, it has earned but not much. In the past two years, the smart life and smart cars that have focused on efforts have achieved growth.

In terms of prospects, there is still room for growth. For example, the number of customers served by SenseTime and customer revenue have increased. SenseTime's smart life business segment has served a total of 317 customers throughout the year, a year-on-year increase of 23%, and single-customer revenue increased by 86%.

Looking at the competition of large models, the position of the AI "four little dragons" is a bit embarrassing. According to public information, the first echelon of large-scale models is Tencent, Baidu, Ali, and Huawei. The core appeal of these four large-scale models is to reconstruct the bottom of cloud computing.

The current level of SenseTime is slightly inferior to that of Dachang, and it is in the second echelon.

China International Finance Securities believes that compared with other companies that provide large language models, SenseTime has advantages in technology and talent, computing power supply, and capital. However, compared with Internet companies, Huawei, and iFlytek, SenseTime There are slight deficiencies in the stable profit model and the ecological construction of landing applications.

In terms of technology investment, commercialization and other levels, even the first echelon still has a long way to go before the technology is implemented, not to mention SenseTime and Yuncong, which have weaker "hematopoietic" capabilities.

Simultaneously, the large model begins to defoam. According to data from the third-party website SimilarWeb, in June this year, the global traffic of ChatGPT’s Throne and mobile clients decreased by 9.7% month-on-month. In addition, Microsoft Bing, Character. The traffic of websites such as AI also declined to varying degrees in June.

Knowing that he entered the wilderness late, he sticks to the computer vision business. At this stage, Megvii's enterprise business product system can be divided into three series: identification products, perception analysis products, and innovative product solutions.

"The goal of Megvii is to innovate AI technology that affects the physical world." Megvii co-founder and CEO Yin Qi emphasized at the meeting.

The large-scale model dividend has subsided, and there is really not much time left for AI companies that have not yet made a profit. Ali’s liquidation is a signal-capital’s risk rating for AI companies has risen.