Ma Yiwen: How to cast artificial intelligence computing power? Detailed explanation of chip and communication investment opportunities

Per editor: Zhao Yun

(Live Guest: Ma Yiwen, Assistant Fund Manager of Cathay Quantitative Investment Department, Live broadcast time: June 27)

First, the supply and demand relationship in the field of artificial intelligence computing power continues to be tense

Moderator: Today we will focus on investment opportunities in the artificial intelligence sector. Although the sector experienced a certain correction before the Dragon Boat Festival, some of these sub-sectors, such as memory chips, are still highly concerned by the market. Therefore, please tell us about the role of computing power in the entire field of artificial intelligence?

Ma Yiwen: In fact, in the development of artificial intelligence, computing power plays an important role. The large model generation of artificial intelligence is mainly divided into two application aspects: training and inference.

During the training phase, the AI accepts labeled data, which can come from corpus resources such as the Internet, print newspapers and magazines. By labeling this data and feeding it to a large model for training, you can determine the parameter settings of the model.

The inference stage is mainly when the user uses the large model, asks a series of questions to the artificial intelligence, and the model calculates the corresponding answer and outputs it to the user.

It can be seen that both aspects of the composition and use of large models require computational steps. The computing power required for the entire large model is not limited to the training phase, but also includes the inference phase. Both of these links require enormous computing power.

According to the analysis of overseas institutions, like the current popular ChatGPT, the number of GPU chips used in the training stage has exceeded 10,000, and the inference stage has exceeded 20,000. As a result, the demand for computing power remains scarce overall.

Since 2013, the demand for computing power in data science represented by artificial intelligence has been greatly promoted. About every 3 or 4 months, the scale of computing power doubles, and the growth rate is very fast.

In addition to this, we can take a look at the development of large models. Take ChatGPT as an example, it released the first generation of GPT in 2018, with about 117 million parameters and about 5GB of pre-training data. By the latest GPT and this year's fourth-generation GPT, the amount of parameters has reached hundreds of billions of levels, and the amount of pre-training data may have reached tens of terabytes. Judging from the burst of model parameters and pre-training data in the training stage, the demand for computing power has increased significantly.

In addition, from the inference side, it can be found that the demand and supply relationship of the entire computing power is relatively tight, taking the fourth generation GPT as an example, roughly only a dozen to two dozen questions can be asked in three hours, which also reflects the state of insufficient computing power.

Considering these two factors comprehensively, we believe that in the development stage of artificial intelligence big models, especially in the field of computing power, the relationship between supply and demand continues to be tense. From an investment perspective, this is also an important point to watch.

Second, the interpretation of investment opportunities in the field of computing power

Moderator: Mr. Ma also mentioned just now that computing power is very important in the field of artificial intelligence. As an investor, if you want to participate in the investment in the field of computing power when investing in the stock market, what sectors are involved?

Ma Yiwen: In fact, we can discuss the field of computing power from two perspectives.

1. GPU chip

The first is the GPU chip, which is part of the core in the computing chip. In the chip market of artificial intelligence large models, the market share of GPU chips can reach more than 90%. This shows that in parallel computing, GPU chips have clear advantages, both in terms of energy consumption and efficiency. Therefore, the usage of GPU chips is relatively large.

However, in the investment in the chip field, the independent research and development rate of domestic GPU chips is relatively low, especially in the field of artificial intelligence GPU chips, the market share of the two overseas leading manufacturers can reach about 96%, which highlights its absolute dominance.

From the perspective of domestic manufacturers, although there is a trend of domestic substitution in the short term, the current technical level of more advanced domestic manufacturers is basically the same as the technical level of overseas leading manufacturers around 2018. Therefore, there is still a lot of room for domestic substitution.

With the continuous intensification of Sino-US technology friction in the past few years, some overseas advanced computing chips have actually adopted embargo measures against China, so the localization of domestic artificial intelligence computing chips is imminent.

As far as the scale growth rate of the entire industry is concerned, under the general trend of artificial intelligence development, the demand for GPU chips has ushered in an outbreak. According to the forecast of overseas market institutions, in the 10-year period from 2021 to 2030, the GPU market size will grow at a compound annual growth rate of nearly 30%. Such a growth rate is quite fast, whether from the overall growth rate of the market or from the perspective of domestic substitution, the future GPU chip field will provide better investment opportunities.

2. Server

On the other hand, we can also focus on the server field, especially artificial intelligence servers. A server may carry dozens of GPU chips, and thousands of servers make up the entire AI computing cluster. There are many interconnection needs between GPU chips and between servers.

From this point of view, there is actually an increase in demand for communication equipment in the communication sector, such as switches, optical fiber cables and optical modules. In particular, the field of optical modules has received high market attention this year.

The optical module needs to use optical signals in fast transmission application scenarios, and photoelectric signal conversion at the connection end, and the optical module is the component used. As far as the current global competition pattern is concerned, the domestic head optical module manufacturers have strong competitive strength. Last year, among the top ten optical module companies in the world, Chinese companies accounted for about 7, so there was not much problem in terms of localization rate.

With the development of artificial intelligence, there is actually a large demand increment for some high transmission rate optical modules. Especially the 800G optical module that everyone is concerned about this year, since the second half of last year, domestic head manufacturers have begun to ship small quantities. This year, it is expected to gradually enter the supply chain of some large overseas manufacturers, especially leading companies in the field of artificial intelligence such as Microsoft and NVIDIA. Therefore, from the second half of this year, we hope to gradually see the landing of these orders. So the investment opportunities in this part of the communication can be focused on the communication ETF (515880).

For the expectations of the market as a whole, the second half of the year will also provide some good investment opportunities. In general, in the field of computing power, GPU chips are relatively core and have high value. However, at present, the country is relatively weak in terms of technological strength, so in general, investment opportunities are more inclined to the medium and long term, that is, the opportunities brought by domestic substitution and technological strength progress.

Third, the interpretation of investment opportunities in the chip sector

Host: Just now, Mr. Ma also emphasized GPU chips, and many investors are familiar with this field. Two years ago, we began to pay attention to domestic chips, when the hot spots in the market mainly revolved around the chip field, triggering a wave of large-scale markets. So, in the field of computing power, how much does the chip account for the overall investment? After a period of correction, what investment opportunities will there be in the chip sector in the future?

Mu Yiwen: The chip is the core of AI computing power and the basic computing unit that provides computing power. In AI servers, dozens of GPU chips usually need to be used, so the value of chips in AI servers is very high, and in the long run, it is also a direction with high growth prospects.

From the perspective of investment opportunities in the entire chip industry, since the chip industry entered the boom cycle in 2019, investors have paid more attention to the chip sector. However, from the perspective of the industrial chain cycle, it is still in the stage of gradually building the bottom.

1. Consumer electronics cycle

First, let's take a look at the proportion of downstream demand in the chip industry. It can be found that the consumer electronics field represented by mobile phones and PCs accounts for more than 60% of the demand. Emerging fields, such as new energy and artificial intelligence servers, account for less than 10% of the demand. Therefore, overall, we believe that the consumer electronics sector, especially mobile phones and computers, is still the most demanded direction in the chip sector.

However, shipments in the two fields of mobile phones and computers have been in a stage of negative year-on-year growth since the beginning of last year, both at home and abroad.

From the historical data, whether it is consumer electronics demand or the shortage cycle of the chip itself, its downward cycle is about 18 months, that is, about one and a half years. Therefore, from the beginning of last year to the middle of this year, the chip industry has gradually experienced this downward cycle. From this point of view, we believe that starting in the second half of this year, whether from the commodity prices in the market, or from the inflection point rate of the industrial chain, as well as the performance guidance of listed companies, there is hope to gradually see the inflection point of the chip industry cycle.

2. Domestic substitution cycle

On the other hand, we can observe that the domestic substitution cycle of the chip sector may be longer, and the deduction cycle of investment opportunities will be longer. From the perspective of domestic substitution, since last year, Sino-US science and technology friction incidents have gradually intensified. With the breakthrough of domestic substitution, coupled with the upward trend of the chip industry boom cycle, we believe that the investment opportunities in the chip sector in the second half of this year are still worth paying attention to. Relevant investment opportunities can be captured through chip ETF (512760).

Fourth, the chip plate can pay attention to the upstream equipment material link

Host: You gave a detailed explanation of the correlation between chips and consumer electronics, at present, the consumer electronics sector is in the bottom-building stage, as Mr. Ma mentioned, the beginning of a new cycle is gestational. As an investor, if we are going to invest in the chip direction, which specific area do you think has a greater investment opportunity?

Ma Yiwen: From the perspective of the chip industry chain, it can be divided into upstream material equipment, midstream design and manufacturing, and downstream packaging and testing. We believe that the long-term investment space and prospects of the chip industry mainly come from the improvement of the localization rate.

As far as the domestic technological breakthroughs in all links are concerned, the current downstream packaging and testing has a high strength and can basically enter the first echelon of the world.

Midstream design links, such as Bluetooth headsets and smart phone chips, can basically achieve domestic substitution, and even enter the supply chain of overseas manufacturers.

At present, the relatively scarce link is actually upstream material equipment and midstream manufacturing.

From the perspective of the upstream material equipment field, the localization rate of some links is less than 10%, especially the photoresist in the material and the lithography machine in the equipment, which is particularly low in the field of advanced processes. Since last year, the US Department of Commerce has imposed some equipment embargoes on the country, which has also led to some impact on the expansion plans of some fabs in the domestic manufacturing field this year.

From the logic of the comprehensive industrial chain, the bottleneck of upstream materials and equipment has led to a large gap between the technical strength of the current Chinese mainland fab and the world's most advanced technical level. At present, this year's domestic equipment factories are carrying out technological breakthroughs and product introduction. Next year, the affected fab expansion plan is expected to land, the main reason behind it is that domestic equipment manufacturers have made some progress in technological breakthroughs and product introduction.

Therefore, in general, from the current perspective, from the perspective of localization rate, it is recommended that everyone pay attention to upstream material equipment manufacturers. With the increase in localization rate and the rising expansion cycle of downstream fabs, demand growth is still obvious.

Then especially the fab, there are about three major fabs in China, and there are about dozens of relatively small fabs. Their capital expenditures and expansion plans are fairly robust over the long term. For upstream material and equipment manufacturers, on the one hand, they are facing the growth of overall demand, and on the other hand, there are opportunities for domestic substitution. Especially for equipment manufacturers, their value accounts for about 40% of fab capital expenditure, so it is basically a fairly important part. Therefore, we believe that the long-term growth space in this area is relatively certain, and we recommend that you pay attention.

On the other hand, although the localization rate of downstream packaging and testing has been quite high, there are still some investment opportunities for overtaking on curves. Including the current popular advanced packaging technology, it stacks some relatively backward chips, so as to realize the functions of some advanced process chips. This is also one of the investment opportunities for future domestic replacement or corner overtaking.

In general, we believe that the current chip industry chain, if you want to pay attention to the investment opportunities in its segments, it is recommended to pay attention to upstream materials and equipment, which have a low localization rate and relatively certain long-term demand prospects.

Fifth, the driving factors of the communication sector

Moderator: Mr. Ma actually emphasized the logic of domestic substitution of chips, and suggested focusing on the upstream. In the artificial intelligence discussion just now, optical modules and communication sectors were also mentioned, which are also hot areas this year, and have accumulated a large increase since the beginning of the year. For these sectors, is there any specific investment logic that can prove the sustainability and long-term growth of their market?

Ma Yiwen: Regarding the investment opportunities in the communications sector, we believe that there are two main aspects. The first is support at the policy level, followed by technological breakthroughs and demand growth at the industry level.

1. Policy dimension

From the perspective of policy, we believe that the driving force of the current market mainly comes from the change of traditional concepts. In the past, data was often not considered an important element. However, with the release of 20 data policies, a series of problems including data rights confirmation, circulation and transactions have been solved, and the mainland's data basic system has been gradually constructed. For data elements, it is expected to gradually become a production factor in the future, which releases the market's attention and confidence in the data level.

This year, we see three important policy catalysts. The first is the overall plan for the construction of digital China released at the beginning of the year, one of the key concerns is to include the construction of digital China in the assessment scope of the party and government system, which will provide a strong guarantee for the implementation of digital China construction.

In addition, the plan mentions that social funds should be guided to participate in the investment and financing system of digital China's construction, which solves the market's concerns about insufficient funds.

The third policy is related to the plan, which also mentions that the digital economy should fully empower social and economic development. In this goal, several areas are involved, including agriculture, industry, finance, education, health care, transportation and energy. This means that the space for the digital economy in the development and empowerment of the real economy will be effectively expanded. The establishment of the National Data Administration reflects the strategic importance of the digital economy for national security and economic drivers. Combined with the aforementioned promotion of eastern data and western computing power projects, on the whole, an indispensable part of the construction of the digital economy is actually the demand for communication equipment.

We know that with the combination of eastern data and western computing power, a series of links in the communication industry chain such as optical communications and satellite communications will inevitably face certain demand growth and investment opportunities. From a policy perspective, breakthroughs in the dimension of the industrial chain have become more important. The field of artificial intelligence just mentioned is an example. The origins of this field can be traced back to the release of ChatGPT last year, and at first the market did not pay much attention to it, mainly due to the lack of confidence in its commercialization. However, with the release of the fourth generation of GPT this year and the launch of large-scale models of artificial intelligence by major Internet and technology giants, we have found that it can achieve many functions, which has greatly helped its commercialization.

For example, the fourth generation of GPT has seen significant growth in logical reasoning and word processing capabilities. It can participate in human law, accounting, and other exams, and can achieve the top 10% or so scores. It can process tens of thousands of words. Such capabilities are very helpful for its commercialization.

Overall, the advancement of artificial intelligence large models is quite fast. Therefore, we believe that the application of artificial intelligence large models in various fields such as finance, games, medical care, and law is expected to be gradually realized. From the perspective of investment opportunities, especially in the upstream computing power field, the certainty of demand for communication equipment still exists.

2. Industrial dimension

Another focus of the industry dimension is the field of innovative information and innovation. We know that Xinchuang is actually the concept of information technology application innovation, which is not very new. The last round of information innovation began in 2019, when domestic communication equipment manufacturers were sanctioned overseas, and China began to recognize the importance of domestic basic software and hardware replacement.

Looking back at the last round of information creation market, its development context is basically from the second half of 2019 to 2020. The landing of the policy has gradually promoted the formation of the market, and by the second half of 2020, the landing of orders began to appear, and in 2021, it basically reached the stage of performance cashing. The entire Xinchuang market lasted for about two years. Since last year, especially since September last year, the State-owned Assets Supervision and Administration Commission has issued a guiding document requiring domestic central state-owned enterprises to replace domestic software and hardware.

Different from the previous round, the space for this round of information creation is greater, because the previous round of domestic substitution mainly occurred within the party and government system, while this round spread to a series of central state-owned enterprises under the State-owned Assets Supervision and Administration Commission, involving finance, communications, agriculture, energy and other industries with a high proportion of central state-owned enterprises, and the space for domestic substitution is relatively large.

These substitutions include replacements for database servers, operating system hardware, software, and hardware. We believe that the current round of information innovation is still at an early stage, from the issuance of policy documents in September last year to the gradual emergence of a series of bidding in the fourth quarter. In the first and second quarters of this year, it is still in the early stages of order landing, and by the third and fourth quarters, we hope to gradually see the realization of performance.

Therefore, from the perspective of performance, the communication sector also has a good outlook this year. Based on the three policies of the policy dimension just mentioned, including the promotion and development of information innovation and artificial intelligence in the industrial dimension, the growth space of the communication sector, especially in the field of communication equipment, is relatively large. Interested investors can pay attention to the Communications ETF (515880).

6. Interpretation of investment opportunities in the game sector

Moderator: Mr. Ma mentioned two directions: one is the prospect and demand of the communication industry driven by information and innovation, and the other is the driving force of artificial intelligence that we are most concerned about. In the field of artificial intelligence, the market is currently very hot on AI technology, discussing its application in all walks of life. Especially the games we will talk about next, the game industry can be said to be quite rapid in this regard, and has also experienced significant growth recently. Mr. Ma, what is your opinion on the gaming sector? Do you think this field will have greater opportunities for updates and iterations in the future?

Ma Yiwen: Actually, for investment opportunities in the game industry, it is not just about artificial intelligence.

1. Restoration of version number issuance

First, we can look at the industry from its fundamental and policy perspectives. Over the past few years, the gaming industry has undergone some degree of adjustment, so valuation levels were low before the start of the year. The main reason is that from a fundamental point of view, from the middle of 2021 to 2022, there were 8 months when the version number of the game company was in a state of discontinuation.

It mainly stems from the strong regulatory cycle of the game industry, which is oriented towards the protection of minors. The requirements of the game industry such as content creation and the protection of minors impose restrictions on the issuance of game version numbers. We know that the game version number is a vital lifeline for new games to go live, so this aspect is very important.

Since the end of last year, game companies have gradually resumed the issuance of version numbers, especially some of the leading manufacturers listed in the second half of last year gradually obtained their own game version numbers. Including the first 6 months of this year, there are basically game version numbers issued every month, and the number of game version numbers issued so far this year is basically the same as last year. After a game company gets a version number, it usually takes about 1 to 2 quarters to launch and test a new game. Therefore, in the second half of this year, we are expected to see a steady flow of water contribution.

Therefore, from the fundamental perspective of game companies, with the recovery of performance and the improvement of growth rate in the second half of this year, the prosperity of the entire industry will rise.

2. Policy support

So from a policy point of view, we mentioned that since 2021, the domestic game industry has entered a strong regulatory cycle oriented to the protection of minors.

In the past year, this regulation has achieved quite good results in the protection of minors in the gaming industry. Whether it is the play time of minors or the contribution of game revenue, the proportion is already relatively low. From a policy point of view, strong regulation has entered a relatively stable stage.

On the other hand, since the second half of last year, state media have published articles emphasizing the importance of the gaming industry in terms of technological and cultural values. It affirmed the contribution of the game industry to domestic culture, such as the spread of Chinese culture after games go overseas, and the technological value of the game industry for virtual reality (VR), augmented reality (AR), GPUs, graphics cards, etc.

So, overall, a major investment logic for the gaming industry this year is the valuation repair brought about by its fundamental repair and improved policy environment.

3. Artificial intelligence catalysis

In addition, we can see that the game industry is also one of the hottest concepts in the field of artificial intelligence this year. Including in the first half of this year, the gains of the game ETF (516010) doubled at one point. The main reason behind this is that the gaming industry has a good combination with the field of artificial intelligence.

We believe that the changes that AI has brought to game companies can be seen in two ways.

The first is to reduce costs and increase efficiency. AI can play a role in the game development process, including planning, development, and art. During the planning phase, AI can help curate and build plots and create content such as text. In the development phase, artificial intelligence has already realized the application of functions such as code assistance and code comments overseas. Especially in terms of art, art costs account for as much as 50% to 60% in the game industry. With the development of overseas artificial intelligence, the realization of multimodal functions, such as large-scale models can realize tattoo designs and other functions, can significantly reduce the overall time and cost of game art creation, and can save about 20% to 30% of the cost. Therefore, the gaming industry can clearly benefit from the perspective of reducing costs and improving efficiency.

On the other hand, the game industry also has innovative ways to play. We know that some smart PC interactive games have been implemented at home and abroad, mainly to enable non-player characters (NPCs) to interact intelligently with players, which will significantly improve the player's game experience.

In addition, the artificial intelligence in the game industry just mentioned in terms of creating plots and text creation is also landing. For gamers, this may lead to a personalized gaming experience that can be enjoyed by thousands of faces. From the perspective of innovative gameplay, the entire game industry will achieve more innovation in the future, which will play an important role in helping the valuation of the entire industry.

On the whole, we believe that the investment opportunities in the game industry this year not only come from the cost reduction and efficiency improvement and innovative gameplay of artificial intelligence, but also from the fundamental repair and the recovery of the policy and regulatory environment after the resumption of the release of the version number, as well as the upside potential of valuation. Overall, the gaming industry remains a sector to watch.

In the short term, as mentioned earlier, the gaming sector has seen a larger increase in the first half of the year, but there has been a certain correction recently. This may be affected by several factors, one is the recent reduction of some listed companies in the game field, and the other is the short-term uncertainty of overseas interest rate hikes. These factors have led to short-term volatility in the sector.

However, in the medium to long term, including several factors mentioned earlier, it is expected to drive both fundamentals and valuation growth in the gaming sector. Therefore, if there is a significant correction in the sector, you can consider gradually paying attention and opening a position in the game ETF (516010) according to the opportunity dip or in batches.

7. Pay attention to the long-term investment value of artificial intelligence

Host: In addition to the gaming sector, the AI sector has also experienced some pullbacks recently, but there are still many investors who have been paying attention to this area. Since the previous increase was very large, many people missed the opportunity to get on the bus. Now, in the context of the pullback, will it bring a second chance to get on the bus? If we want to participate in the investment of the artificial intelligence sector, what direction do we need to pay attention to?

Ma Yiwen: First of all, the reasons for the recent adjustment of the artificial intelligence sector can be attributed to factors such as the reduction of holdings by listed companies, overseas interest rate hikes, and capital repositioning or profit taking near the end of the second quarter.

Overall, we believe that the reason for the correction of the AI sector is not based on the fundamentals of the entire market, but more likely to be a disturbance at the emotional or financial level.

In the medium to long term, we believe that the market for the artificial intelligence sector is still in a relatively early stage. Whether it is the upstream computing power performance cashing, or the downstream application field, and even the model development and vertical model landing of the midstream computer software sector, it is in a relatively early stage.

For the market situation in the field of artificial intelligence, we believe that we should pay attention to it from a more medium- and long-term perspective. In the short term, including the TMT sector of computers, media, communications, electronics, etc., in terms of growth is relatively large. Especially in the first half of the year, the A-share large-cap index is still in a volatile stage, and the excess return of the artificial intelligence sector is relatively obvious.

Therefore, there is indeed a certain amount of profit taking in the short term, especially in July, which is about to face the test of the second quarter earnings report. As we mentioned earlier, public companies in areas like gaming communications may not be able to cash out as quickly. At this stage, we believe that there is still some volatility and adjustment possibility for sectors or listed companies in the entire field of artificial intelligence. Therefore, the advice for investors is to focus on the long-term value of their investment.

In the medium and long term, looking back at the historical investment context of the technology sector, including new energy and 3G/4G cycles, investment is often accompanied by an increase in prosperity and penetration at the beginning. This phase is usually accompanied by rapidly increasing valuations and large swings. In the performance cashing stage, the overall Davis double-click effect is realized, and the market will gradually weaken. Usually it is necessary to wait until the prosperity gradually declines or the performance growth rate reaches an inflection point, which may mean the cyclical inflection point of the gradual weakening of the market.

Therefore, in general, the current AI-related computer, communications, games and other sectors are actually in the early stages of performance fulfillment. Investors should be patient and not be too pessimistic about short-term market fluctuations. If there is a relatively large correction, you can also consider bargain hunting and buying in batches.

Risk Warning:

Investors should fully understand the difference between regular fixed investment and fractional savings methods such as regular fixed investment of funds. Regular fixed investment is a simple and easy way to guide investors to make long-term investments and average investment costs. However, regular fixed investment does not avoid the inherent risks of fund investment, does not guarantee investors to obtain returns, and is not an equivalent financial management method to replace savings.

Both stock ETFs/LOF funds are securities investment funds with higher expected risk and expected return, and their expected return and expected risk level are higher than hybrid funds, bond funds and money market funds.

The Fund's assets invested in stocks on the STAR Market and ChiNext Board will face unique risks arising from differences in investment targets, market systems and trading rules, which investors are reminded of.

The short-term rise and fall of sectors/funds is only used as auxiliary material for the analysis of the article, is for reference only, and does not constitute a guarantee of the performance of the fund.

The short-term performance of individual stocks mentioned in the article is for reference only and does not constitute a stock recommendation, nor does it constitute a forecast or guarantee of the performance of the fund.

The above views are for reference only and do not constitute investment advice or commitment. If you need to purchase relevant fund products, please pay attention to the relevant regulations on investor suitability management, make a risk assessment in advance, and purchase fund products with matching risk levels according to your own risk tolerance. Funds are risky and investment needs to be cautious.

Daily economic news