Product recommendation, traffic management, article generation, code programming, movie special effects production... Since the birth of ChatGPT, the AIGC wave has swept the world, and the upstream and downstream industry chains have also shined brilliantly.
The high prosperity of the market is intuitively reflected in the stock price. Regardless of whether AI companies are profitable or not, their stock prices are mostly on the rise. The stock prices of some games, media and other companies related to the concept of AI are also generally rising. However, from a phased point of view, the trend of individual stocks has begun to diverge recently, some continue to carnival, and some have a correction. Is this a precursor to the return of the AI big model to value investment? Or is it a brief lull before the market rises further?
The stock price of AI concept "parted ways", soaring and callback coexist
Artificial intelligence has become a popular target in the current investment field, and the outside world is generally optimistic about its future development. Therefore, entering 2023, the stock prices of many AI companies have begun to "surge".
In the U.S. stock market, when the market closed on May 30, C3.ai rose by more than 33.4%, and Bullfrog AI Holdings rose by about 18.5%. Nvidia has increased by more than 160% since the beginning of this year, and once broke through the trillion-dollar market value mark. In the A-share market, according to incomplete statistics, AI chip-related concept stocks rose by more than 22.3% at the highest point this year, and "AI+" concept-related sectors all rose by more than 50%. Therefore, many investment institutions believe that the AI sector will be this year. The "main force" of the stock market rise.
While stock prices have risen, good news is still coming out frequently. The first is policy promotion. The "14th Five-Year Plan" has listed AI as the "highest priority" in the field of cutting-edge technology to promote development. The second is market confidence. Tencent CEO Pony Ma Huateng said at the shareholder meeting: "AI is an opportunity that the Internet has rarely encountered in hundreds of years, similar to the industrial revolution that invented electricity."
Just when everything seemed to be moving in a better direction, there were differences in attitudes in the A-share market. On the evening of April 20, a group of listed companies involving the concept of AI, such as Cambrian, Wondershare Technology, Shanghai Film, Zhongke Suguang, Southern Media, and Dianhun Network, collectively disclosed announcements about abnormal stock transactions. Entering May, the stock prices of AI large-scale companies also fluctuated. For example, iFlytek plunged sharply in the intraday market, and Bioland, 360, and Gao Weida followed suit. The stock prices of Kunlun Wanwei and Cambridge Technology once fell by more than 5%.
The capital movements in the past two months may indicate that the market has become alert to bubbles. As Zhang Jun, chairman of China Europe Capital, said in April this year, capital, as one of the important external factors affecting the development of the AI industry, should return to calmness.
This reminds people of Metaverse, another popular concept before the AI model, which was once sought after by capital and is no different from today's artificial intelligence. Facebook founder Zuckerberg even renamed the company "Yuan". Under his leadership, the attention and commercial value of the Metaverse soared. However, the good times didn't last long. According to the "Google Trends" website, the search traffic of the Metaverse will drop by about 80% in 2022. The "Industrial Metaverse Team" established by Microsoft for only four months was disbanded in February this year. department……
Some analysts pointed out that the reason why the Metaverse has been "extinguished" in just a few years is that it has been trapped in concept hype for a long time and consumer demand is relatively vague. At present, artificial intelligence is actually facing the same situation. The matching between market demand and cutting-edge technology is not high, and commercialization exploration is still in the initial stage. It is quite necessary to be vigilant against the emergence of capital bubbles and return to the perspective of value investment.
The long-term growth of AI large model returns to value investment still lacks a clear target
From the perspective of value investing, the company's fundamentals and long-term growth are the core. At present, it has become a consensus that the future prosperity of the AI large-scale model track will be high. However, if we focus on specific companies, we can find that there is still a lack of clear investment targets for long-term growth.
This has something to do with the fact that there are too many subdivided tracks in the current AI large model. It is reported that there are currently multi-modal general-purpose large models (such as GPT-4 developed by OpenAI), industry-specific large models for specific industries or fields (such as Google’s development of Med-PaLM 2 focused on healthcare), and vertical models for specific tasks or scenarios. There are four types of large models (the basic image segmentation model Grounded-SAM developed by the Guangdong-Hong Kong-Macao Greater Bay Area Digital Economy Research Institute), and further in-depth customization of exclusive large models.
Under many tracks, enterprises have accelerated their deployment. In the domestic market alone, there are more than 30 large-scale models so far. Some people in the industry pointed out that the number of enterprises participating in the research and development of large-scale models has actually exceeded 60.
However, although there are currently many AI large-scale model companies, their respective businesses are actually quite different. There are innovative software productivity tools, such as the artificial intelligence product matrix "360 Zhinao" developed based on the 360GPT large model for the search field; "Thousand Questions", adopting the same logic that OpenAI introduces plug-ins for ChatGPT. In the field of games, many companies even proposed "all in AI". For example, Nvidia launched Avatar Cloud Engine (ACE) for Games, which provides customized AI model services for games. NetEase's self-developed AI technology is also applied to the entire process of game industrialization. Key links The work efficiency can be increased by up to 90%. It is not difficult to see that it is difficult for each player in the track to form a clear benchmarking relationship at the business level, and there is a lack of specific valuation anchors.
In fact, under the background that the industry has been in a long-term upward period and the pattern has not yet been determined, there is a certain deviation in measuring its investment value based on the current valuation of the corresponding business. its short-term valuation.
The emergence of business inflection points often means changes in the company's business model. In other words, the long-term vitality of an emerging technology is inseparable from the widespread implementation of commercialization. Therefore, it is a good way to judge the value of the current development of large-scale AI enterprises from the perspective of application scenarios. Ding Lei, CEO of Netease, also said that in the AI large-scale model competition in the future, the real winners will be companies that can choose good application scenarios. Netease's primary task is to explore the application of AI large models in different product scenarios, and to develop high-quality products that are more in line with user experience.
This may not only be what NetEase needs to do, but also the future development direction of the entire industry.
Competing in trillions of market application scenarios laying logic emerges
AI large-scale models are undoubtedly a huge market. According to the IDC report, the global AI market will grow to 221.87 billion US dollars (about 1.5 trillion yuan) in 2025, and China is an important market for global artificial intelligence. With the AI With the continuous implementation of applications, it is estimated that the scale of China's artificial intelligence core products will grow at a CAGR of 26.8% from 2019 to 2025, driving related industries to grow at a CAGR of 27.8%. Kai-fu Lee, chairman and CEO of Sinovation Ventures, said at the 2023 Zhongguancun Forum "Development of Large-scale Artificial Intelligence Models" sub-forum that large-scale AI models contain business opportunities worth tens of trillions of dollars.
However, if you want to compete in this trillion-dollar market, you first need to solve the problem of the current difficulty in the widespread application of AI large models. Many industry experts believe that the implementation of AI large models is a process that is easier said than done. The lack of standardized data and the high cost of computing power restrict the commercialization and scale of AI large models. For example, the computing power cost of GPT-3 training exceeds 4 million US dollars. However, the lack of fusion between the cutting-edge technology of AI and commercialization also means that whoever can take the lead in completing the generalized application of AI will win the race. Judging from the current market situation, the logic of application scenarios is gradually emerging, and a group of potential companies will accelerate the release of value.
At present, the application scenarios of AI are laid out mainly from four aspects. The first is to optimize products and services through artificial intelligence. For example, Meitu uses AI as technical support to launch a beauty camera and hairstyle manager for users' real beauty needs. The second is to improve user experience, such as the AI BUY launched by Taobao to help consumers improve shopping efficiency. The third is to reduce operating costs, such as SF’s AI ARGUS, which improves the ability to automate operations and helps companies reduce costs and increase efficiency. The fourth is to innovate business models. For example, the decision-making platform WAYZBANK launched by Wisdom Technology has improved the decision-making efficiency of financial scenarios such as banks.
The logic of scene laying actually coincides with the essence of AI big model. Kai-fu Lee believes that AI large models are most likely to be empowered by those industries that already have a large amount of data and the data is structured and integrated, such as games and finance. In manufacturing, retail, health, medical and other fields, data aggregation and landing Building is not done overnight, it still takes some time.
Among them, games have always been called "AI training grounds" in the industry. From the application of AI large models in the game industry, we may get a glimpse of its future path to empower other industries.
On the one hand, the current AI large model has mature text and image processing capabilities. According to the keywords provided by developers, the AI large model can quickly generate a large amount of materials, improve efficiency, and optimize development costs. For example, NetEase’s net income of 25 billion yuan and net profit of 7.6 billion yuan in its financial report for the first quarter of 2023 exceeded market expectations, which is related to the dozens of AI efficiency improvement tools it launched. On the other hand, many game companies use AI large-scale model-generated dialogue technology in games to provide players with a better sense of interaction and game experience. For example, the smart cute pet "Jiujiu" in the game "Jiuji: Qifeng Journey" is a generative dialogue smart elf designed using AI large models, which can conduct natural, personalized and interesting dialogues and interactions with players, which improves the game experience. Interesting.
The AI large model can empower game development, which is related to its easy standardization of development process, less difficulty in data accumulation and scene learning, and easy large-scale reuse. Based on this, in order to realize the large-scale reuse of AI large models in more industries, it is necessary to reduce the application threshold and marginal cost, and the development of pre-trained large models can be said to be a necessary measure.
In this regard, many companies have already begun relevant deployments. For example, NetEase undertook the construction of the Zhejiang provincial pioneer project "ultra-large-scale pre-training model cloud platform". Model, Meta has open sourced the SAM model, which is an in-depth step towards the basic visual model.
It is foreseeable that with the growth of data volume and computing power, the accuracy and generalization capabilities of AI large models will be further enhanced, and will continue to shape new formats and new scenarios, creating huge commercial value, which belongs to the "iphone moment" of the AI industry Expected to actually come. However, the long-term development of the industry needs to squeeze out asset bubbles. The recent partial correction of AI concept stock prices may mean that capital is shifting from short-term hype to long-term value. This is undoubtedly good news for the development of large-scale AI models.
Author: Good blue is not good
Source: Hong Kong Stock Research Institute