Who does the Chinese version of ChatGPT feed?

The take-off of the ChatGPT sector, which companies and segments are feeding.

Text / Ba Jiuling (WeChat public account: Wu Xiaobo channel)

The Chinese version of Generative AI Smash Bros. officially sounded the gong.

On April 11, Alibaba Cloud released ChatGPT's similar product "Tongyi Qianqianqian". Up to now, in addition to Baidu, which said Wen Xin's words, and Ali, which released Tongyi Qianqian, Tencent, Byte, 360, Kunlun Wanwei, SenseTime, and Baichuan Intelligent have all held loudspeakers and announced: "Yes, we have our own big language model of the same model as ChatGPT." ”

The magic of telling the market, "We have a big language model," is just too great. After the announcement, Kunlun Wanwei's share price has risen by more than 80% since March. Even because the stock price rose too fast, it also received the "concern" of the Shenzhen Stock Exchange. After Baidu's release of Wen Xin's words, nearly 300 ecosystem companies have received sunshine-like benefits.

This also gives investors a new way of thinking – in addition to plugging the language model into their own products and using the technology, ancestral supply chain thinking can also analyze this new industry to see which companies and segments the ChatGPT sector has taken off and which companies and segments have been fed.

Therefore, our Wu Xiaobo channel report team also followed this idea, sorted out ChatGPT and the industrial investment logic behind it, and found three major investment directions under the ChatGPT sector and a segment worthy of attention.

It soared 60% in 2 months

The ChatGPT sector has walked out of the "overlord step"

Before digging into these opportunities, let's briefly sort out the trend of this round of ChatGPT sector.

On the trading day after the Spring Festival in 2023, ChatGPT officially began its stock market adventure, and the capital market began to understand this new technology from the surface to the inside.

Data source: Wind, Snowball @ Neumann Feng Note: The data is as of March 29, 2023

On the whole, the process of the capital market constantly realizing the importance of ChatGPT can be roughly divided into three gradual stages of "application-data-computing power". At different stages, the market also throws a series of questions at ChatGPT — or generative AI products. Whoever can answer these questions will be "rewarded" by the capital market during the cycle.

The ChatGPT sector moves from January 30 to April 10, 2023

The first stage that ChatGPT has experienced in the A-share market, the theme is "what can it do", and content companies are one of the most concerned targets in this stage.

From the end of January to the middle of February 2023, the index of the ChatGPT sector of the A-share market rose from 1579 points to 2169 points, showing an increase of 37.4%.

During this period, most investors understood ChatGPT as "a conversational AI" and wondered what tasks the product could accomplish. Hanvon Technology stood up and told the market that "generative AI trained with a large amount of data can help produce graphic content" while telling everyone that "the company is already using this technology".

As a result, Hanwang Technology, which hitched a ride, gained 5 consecutive limit boards in early February, becoming a microcosm of the early stage of the ChatGPT sector galloping A-shares.

The second stage that ChatGPT goes through in the A-share market is "why is it so strong". Concepts such as chips and computing power are stronger at this stage.

During this time period, investors discovered that the power of ChatGPT came from a large training sample. With the rapid growth of related application users, the computing power requirements for computing infrastructure are getting higher and higher, and the demand for related chips is also put forward. Therefore, while the market demand for the flagship graphics processing chip A100 has soared, the manufacturer of this chip, NVIDIA, also represents the concept stock of computing power and stands at the center of the market.

In the third stage that ChatGPT has experienced in the A-share market, the theme has gone a step deeper, becoming "how to become stronger".

The semiconductor sector has become the focus of the market. In the second stage, the market recognized the influence of chips, computing power and other foundations on the field of ChatGPT and artificial intelligence. So in the third stage, the market began to focus on the more upstream semiconductor industry.

As the basis of chip production, the nature of semiconductor materials and manufacturing processes determine the performance and function of chips. So when the market realized the gap in computing power, the semiconductor industry was pushed to the center of the market. From March 16 to April 10, the A-share semiconductor index rose from 5522.05 points to 6326.23 points. In less than a month, the increase reached 14.6%.

Cut open the ChatGPT industry chain

Several new tracks surfaced

In addition to clarifying the market's investment logic for an industry, investors also need a clearer investment combing.

From the upstream and downstream of the industry, we can split the related industries of ChatGPT into four dimensions: algorithm, computing power, data and application. Under the four dimensions, it can be spun out of chips, data analysis, intelligent transportation and other subdivided industries.

Essence Securities mentioned in the "Digital China Panorama Investment Manual" that under the huge and complex industrial system, there are three directions that are worth paying attention to in the upstream and midstream links of ChatGPT:

1. NLP technology

Natural language processing (NLP) is a branch of artificial intelligence that enables computers to understand, generate, and process human language, allowing users to ask machines for data using natural language text or speech. To put it bluntly, humans no longer need to learn code to communicate with computers, type or talk like chat, and make computers understand their needs.

Since ChatGPT is mainly based on natural language processing, enterprises with a large number of precipitations in the NLP field are expected to be the first to realize partial reproduction of functions.

2. Data labeling

In the process of ChatGPT generation, data annotation is more basic than NLP. Data labeling is to label unprocessed images and text, such as taking out a picture to let AI understand what is a car, what is a person, and what is a traffic light.

In the training process of ChatGPT, the strength and accuracy of manual labeling are increased. In the future, in the field of artificial intelligence, high-quality data sources and powerful labeling capabilities will become the infrastructure of the industry, which is beneficial to artificial intelligence data labeling enterprises.

3. Computing power facilities

The biggest difference between the various versions of GPT products is the ultimate improvement in the size of the model. For example, ChatGPT3.5 has 175 billion parameters, and by GPT4.0, there are more than 100 trillion parameters. The process of iteration of ChatGPT products is the process of major manufacturers throwing money to buy sky-high computing power. In the future, data centers and related supporting industries are expected to achieve faster growth.

AI that can read pictures

Make manufacturing go more smoothly

From the perspective of the birth logic of a product, the above three investment directions are in the early stage of the birth of ChatGPT products and are the bottom of the pyramid. One level higher, such products will play a role at the practical application level. For example, the field of vision applications.

Through learning, machines can recognize images to obtain information, a function similar to how the human brain processes visual signals. Mature algorithms can allow machine vision to eliminate fake news on the Internet and obtain more accurate results, so as to reduce manual operations and deal with complex and diverse usage scenarios.

Images are the most informative type of data, accounting for more than eighty percent of data collection. So now electronics, automotive, battery, semiconductor, packaging, food, medicine and other industries are iterative machine vision through image data collection.

Among the major industries, 3C electronics is the largest application market, accounting for 25% of the market; followed by the flat panel display market, accounting for 12.15%; The market share of automotive, battery, robot, semiconductor and other scenarios is more than 5%.

The scale of the software and information industry has increased from 1 trillion yuan in 2010 to 8 trillion yuan in 2020, and has maintained rapid growth, and its contribution rate to the economy has increased from 3% in 2010 to 10% in 2021.

From the analysis of the domestic market, the demand for the machine vision industry mainly comes from two aspects:

1) Industrial upgrading. The comprehensive transformation of China's manufacturing industry requires precision and high-end. The accuracy and objectivity brought by machine vision are expected to play a "push" role on the road of China's manufacturing development.

2) Cost reduction and efficiency increase. To put it more bluntly, the most practical role of artificial intelligence, including ChatGPT products, for the manufacturing industry, is to save costs. Especially in the era of transition from demographic dividend to talent dividend.

Although in the week of April 4-April 10, the ChatGPT sector suffered another round of correction, falling by more than 10%. However, short-term fluctuations and fierce competition do not prevent large manufacturers from squeezing their heads to join this front.

This also proves that an AI that can communicate with humans without barriers can show the value brought by the product itself in multiple dimensions such as time, industry, and enterprise. At the same time, investors should also see the direction of investment from the life cycle of an industry and a product.

In this process, what the Xiao Report team can do is to present these industry opportunities to you one by one through perspective extraction, data summary and other methods.

Author of this article | Lu Hong | Current value edit | Sesame sauce

Editor-in-Chief | He Mengfei | Image source | VCG