Huang Yue: Artificial intelligence is constantly catalyzing, how to invest chips and games?

First, chips are the cornerstone of artificial intelligence

Host: Recently, there was news that NVIDIA's first-quarter performance and second-quarter outlook actually exceeded market expectations. This also verifies the trend of continuous growth of artificial intelligence computing power. Can you briefly introduce the position of chips and their specific role in the development of artificial intelligence.

Huang Yue: Okay. It can be said that in the field of artificial intelligence, chips are a very core component. Because AI mainly involves several elements, including computing power, algorithms, and how to build and debug large models on top of computing power, there are also applications, and data is also one of them. These are probably the four main points of artificial intelligence.

Computing power can be said to be very core, without computing power, everything else is useless, so chips in the entire field of artificial intelligence can be regarded as the cornerstone of the foundation.

NVIDIA can be said to be the core of the field of artificial intelligence computing. Because everyone knows that artificial intelligence needs to use computing power such as supercomputing and cloud computing. This computing power is fundamentally different from the personal computers or mobile phones we normally use.

Many servers or our own computers mainly use the computing power of the CPU, and although it is very powerful in handling complex tasks, it takes time to switch between tasks and is not very good at handling multiple simple tasks at the same time. Therefore, at present, graphics cards are mainly used as chips in artificial intelligence training. Nvidia was originally a company that mainly produced graphics cards, and investors familiar with personal computers knew this.

So why is NVIDIA able to get to the forefront of artificial intelligence? This is because the core part of AI is currently dominated by neural networks. Now people engaged in artificial intelligence can basically be divided into two categories, one is engaged in neural networks, and the other is engaged in non-neural networks. Most of today's large models are based on neural networks. As a model for artificial intelligence, neural networks mainly handle many less complex but very large tasks through parallel computing. In the process, the graphics card has instead gained an advantage in computing.

Although the graphics card is not as powerful as the CPU in its ability to handle a single task, it is well able to support less complex tasks such as parallel computing. Therefore, at present, graphics cards are mainly used as chips, and among domestic manufacturers, objectively speaking, there are almost no chips that can meet the needs of computing power such as artificial intelligence training and preprocessing. Therefore, the correlation between the field of artificial intelligence chips and domestic manufacturers is very low, and only one or two listed companies' chips may be able to meet the needs of some subsequent reasoning links, of course, this is also a large increment.

In general, chips are the core foundation of artificial intelligence, but the relationship with domestic manufacturers is very limited, because the domestic chip industry is basically limited.

Plus, it's not just a matter of technology. For some domestic core chips, its computing power and system parameters may also be very powerful. NVIDIA has a range of technologies throughout the system, including hardware and software. In addition, their many years of experience in the field of artificial intelligence and cooperation with various manufacturers allow their engineers to be deeply involved in AI debugging, identifying problems and diagnosing problems. This is something that domestic manufacturers do not have.

However, this also provides an opportunity for domestic manufacturers to catch up, especially for some manufacturers, although they may not be able to train at present, but may be barely feasible in terms of reasoning. So for them, it's an opportunity.

Second, the chip industry investment logic interpretation

Moderator: Okay, we just mentioned chips. Can you elaborate on the current investment logic of the chip industry? Because there are many types of chips, including the previously mentioned AI chips and the memory chips that everyone has been paying attention to, these are the hot areas in the market. Which chips are optimistic about next? Or what are the investment opportunities worth our layout?

Huang Yue: The chip industry chain is mainly divided into several links, due to limited time, we will not discuss the entire industrial chain in detail today, we mainly focus on its downstream applications.

In the past, chips were mainly used in consumer electronics such as servers, computers and mobile phones. In general, the chip is located upstream of consumer electronics, similar to the components of consumer electronics, just like lithium ore in the new energy vehicle industry chain, it plays an important role in this chain.

For domestic chip manufacturers, the first is consumer electronics-grade chips. There were some domestic manufacturers in the first echelon of the world, such as Huawei, but they were also affected by some interference. In addition, there are some manufacturers with relatively low technical difficulties, such as low-end storage and power semiconductors, including a large number of semiconductors for new energy vehicles. In general, the technical level of the country is roughly at the level of the quasi-first or second echelon.

Then for some core mitigation measures, such as CPUs, memory chips and computing chips, most of the design levels of domestic chips are at the level of the second and third echelons, so overall, the road to catch up is still very long.

Another most critical link is chip manufacturing, especially upstream chip manufacturing, equipment and materials, which are currently seriously constrained. Therefore, when referring to constraints, it mainly refers to the upstream link.

In some mature markets, such as 28nm process, China already has relatively good autonomy and mass production capabilities.

It is generally believed that it is necessary to break through to the high-end, but objectively speaking, it is very difficult to break through the high-end, and the high-end demand is actually not as great as people think. Mature processes such as 28nm, or up to 14nm, can basically meet the needs of traditional servers and personal laptops.

Even artificial intelligence chips, which sound very high-end, basically only need a process of 7 nanometers. Like NVIDIA's current core chips, they basically use this process, and its most advanced process is mainly used in mobile phones. For mobile phones, it is basically necessary to use the most advanced processes like TSMC and Samsung, but the vast majority of other applications can be satisfied using mature processes.

Therefore, in China, the chip industry can be mainly divided into several categories:

1) The first is some chip design companies driven by new energy vehicles in the past few years, which have experienced a wave of independent markets;

2) The other category is companies related to consumer electronics, including low-end memory chips and other design chips closely related to mobile phones. These companies are still in a downward cycle in the short term, have not completely cleared out of the market, and are still in the bottom-building stage;

3) There are also some companies related to domestic substitution, such as domestic equipment and materials, which are mainly affected by the independent market and the expansion of domestic equipment manufacturers.

Therefore, the investment targets of the domestic chip industry can be mainly divided into these categories, each with its own logic.

For consumer electronics-related targets, I think it may take a while longer, and there may be a rapid inflection point in the second half of the year or so. If the development is relatively slow, it may not usher in a phased inflection point until the first half of next year. However, stock prices are usually ahead of the curve, and they tend to be about a quarter or so ahead of the fundamental inflection point. Therefore, I think that the trend of consumer electronics-grade chips in the second half of the year will show the characteristics of shock and bottoming. The targets related to domestic substitution are mainly affected by policies and capital flows. Overall, I think this year's chip industry investment belongs to the bargain hunt. You can pay attention to the chip ETF (512760).

Much of the opportunity in the chip industry depends primarily on the consumer electronics sector. Overall, if the consumer electronics market is on the rise, it will bring a lot of incremental demand to the chip industry. Of course, the field of artificial intelligence also provides many growth opportunities for the chip industry, but the correlation with the domestic market is not as high as everyone thinks. More related to overseas NVIDIA companies.

Third, when will the inflection point of the chip industry appear?

Host: Just mentioned that the demand in the field of consumer electronics accounts for a relatively large proportion of the overall demand for chips. So how should we think about chip demand? What exactly is the current demand, and will there be an inflection point delay?

Huang Yue: Regarding chips, I just introduced several downstream links of chips, and there are two main aspects to consider.

1. Periodicity

The first is cyclical, i.e. based on our existing consumer electronics, especially mobile phones.

If you want to see when chips recover, you need to pay attention to the sales of mobile phones. When mobile phone sales recover, we can say that the chip has reached a phased inflection point. However, in the industry chain, chips may be subject to the inventory overhang and inventory clearance process of mobile phone manufacturers, which adds a certain complexity. There may also be problems with the conduction from the terminal to the inventory.

Overall, to talk about the cyclical recovery of consumer chips, it depends on when mobile phones recover, and there is obviously no recovery.

2. Increment

In addition to periodicity, focus on increments. In the past few years, new energy vehicles have brought a lot of incremental demand, and the incremental demand that can be seen in the future mainly comes from artificial intelligence.

However, at this stage, the relationship with the domestic market is not so close, as just mentioned. More correlations may exist in the field of consumer electronics, such as VR, MR and other devices. These devices have been around for years, and the key in the next few years is whether their cost can be reduced, and once the cost is reduced, it is likely to lead to mass adoption.

These are incremental demand, and after the incremental demand appears, the entire chip sector is also expected to have a big market. Therefore, basically the cyclical recovery depends on mobile phones, and new increments can focus on devices such as artificial intelligence and MR/VR.

Fourth, the domestic deterministic opportunities of artificial intelligence are concentrated in upstream computing power

Host: Recently, investors have paid more attention to the artificial intelligence sector, and we know that artificial intelligence has a troika, that is, computing power, algorithms and data. Can Mr. Huang benchmark it with new energy, and combine the upstream, midstream and downstream to answer the relevant investment opportunities in detail?

1. Upstream

Huang Yue: Regarding the computing power or infrastructure of artificial intelligence, it is similar to the lithium mine in new energy vehicles, which is in the most upstream position, and it cannot be promoted without this thing. This is the most central part, and it is also the most certain part at this stage, because it is uncertain who will succeed downstream.

2. Downstream

The downstream application of artificial intelligence can be compared to the whole vehicle in new energy vehicles, which should be the most valuable part in the future.

For example, let's look at Tencent's WeChat, WeChat's technical content may not be as high as the operating system or office software, and may not even be as difficult as the technical difficulty of domestic operating systems or office software. However, the commercial value of WeChat is certainly much higher than that of domestic operating systems or office software.

So, overall, the real greatest value of AI in the future will certainly be in downstream applications, but these applications are mostly short-term, and it is impossible to predict which companies will stand out among them. Therefore, at present, the funds in the market are still mainly concentrated in the link of computing power.

3. Domestic upstream computing power can pay attention to communication

So about the computing power, the core I mentioned earlier is the chip. Chips are at the top in terms of added value. However, as mentioned earlier, at this stage, computing power chips are basically not much related to China.

However, there are also opportunities with a relatively high correlation with computing power in China, such as optical communication and optical modules in the field of communications. This is because between the chips, it is necessary to connect and integrate through the network, and a large number of optical communication devices are required in this process. Including photoelectric signal conversion chip and data connection cable, similar to our commonly used network cable. In these fields, domestic manufacturers are in a leading position in the world, and most of their customers are overseas customers, such as Microsoft, Facebook, Google and so on.

Therefore, in general, in this year's computing power environment, the core plate in China is not chips, but communications. This can also be seen in the share prices of optical communications and operators, which are mainly driven by cloud computing.

4. In the future, the real big track in China is the application field

However, in the future, domestic manufacturers may focus more on downstream applications. In other words, one or two leading domestic companies may be able to build large-scale models, or they will buy large-scale model licenses abroad, or enter into long-term agreements with foreign countries to call their large-scale models and develop small-scale models of vertical industries themselves. For example, the pharmaceutical industry, the financial industry, etc.

These small-scale models are perfectly achievable, for example I want to know where the stock price is moving today, or I want to observe if there is an island reversal in the technical graphic. The existing technology is perfectly achievable, but I might want to talk to him and ask what the market was like when there were island reversals in history, and review periods when stock prices performed better. These are all things that existing technologies cannot do, because existing technologies cannot talk and communicate with artificial intelligence, and can only obtain information through their own queries.

In the future, in the vertical field, domestic application scenarios are very rich. And the country can circumvent our disadvantages in computing power and algorithms. In terms of application, you only need to call other people's models, as long as the application scenario is done well and you have relevant data in the vertical field, you can build a model suitable for that field.

Therefore, I think that in the short term, the domestic more certain areas are communications, especially optical communications and operators.

Of course, some areas of benefit are the assembly of consumer electronics. Nvidia recently released an assembled chip. Simply put, NVIDIA connects its own CPU and GPU in series, provides it to customers in the form of a server, has both computing power and parallel computing capabilities, and packages storage functions together.

In this case, the requirements for packaging technology are very high. Therefore, for consumer electronics packaging companies, they will also benefit from the development of computing power.

However, I think the real big track in China in the future is still the application field. At present, I think the application field is still in a relatively early stage. Therefore, there is no need to rush to act, in the future, you can use index funds, such as computer software and other related index funds, to layout related application targets in a one-click way. You can pay attention to the Communications ETF (515880), Computer ETF (512720), Software ETF (515230).

Fifth, the impact of artificial intelligence on the game industry

Moderator: Next, let's discuss the recent hot game industry. Recently, driven by artificial intelligence, the gaming industry has boomed again. We see news that NVIDIA has announced a custom AI model foundry service for games. So I would like to ask Mr. Huang, can you briefly analyze what kind of far-reaching impact this will have on the game industry?

Huang Yue: Okay, the field of AI vision is very important for the entire field of AI. At present, what we can do is that, for example, our surveillance cameras, they must have applied artificial intelligence algorithms such as face recognition. But the technology we have today is largely based on recognition, not creation. However, with the advent of large models, many technical challenges related to generating images have also been broken. Therefore, a large model is essentially a language model, but it seems to work well in any field. Therefore, the field of artificial intelligence has developed to architectures based on large language models, combined with small models of our own subdivided industries. In this way, the demand for computing power and other requirements is not so high, and we can better meet specific business needs.

1. Reduce costs

As far as game production goes, images are one of the biggest business scenarios that AI can replace. Many industries have similar characteristics, such as lawyers, financial analysts, etc., and they all have a big brother to guide a group of people to get the job done. A similar situation exists in advertising, media, and gaming. Big brother acts as an architect or leader, providing creative ideas, including advertising, visual effects, sound effects, and storytelling.

In the process, there will be experts who focus on various segments to provide ideas. Once the idea is determined, a group of coders need to realize it and put the idea into practice. In the process, there are constant iterations.

In advertising, we often hear jokes about customer versions such as version 1.0, version 2.0, version 3.0, and maybe even modified until version 10.0 and finally back to version 1.0. These links can be replaced by artificial intelligence, but the big brother is irreplaceable, because he is responsible for providing ideas and has a lot of experience. However, artificial intelligence in the big brother's creative landing link can greatly reduce costs, and artificial intelligence will not write some satirical customer or joke-making content.

In the game, visual creation and story creation can be done with the help of artificial intelligence. Big Brother may provide some scene needs, such as gloomy tones and sad atmosphere, while requiring sad music. At this time, artificial intelligence can create corresponding visual and sound scenes, and further processing and modification through dialogue, this communication cost is very low. Therefore, in the field of games, artificial intelligence is mainly reflected in reducing costs and improving research and development efficiency.

2. Improve the game experience

In addition, AI has the potential to enhance the experience of certain games, such as non-player characters (NPCs) and dialogue mechanics in games.

The NPCs and dialogue mechanisms in existing games are basically based on preset scripts, but in the future, we can fully use artificial intelligence to interact, which will bring a creative interactive experience, in line with the concept of the metaverse.

The core idea of the metaverse is to be able to further create new virtual scenes in virtual scenes, not just based on preset scripts. In the future, we may achieve a more vivid interactive experience through artificial intelligence, which will be a fundamental difference, that is, the difference between predictive and non-predictive.

Therefore, I think the most mature and obvious application in the current gaming space is cost reduction. As for the further applications I mentioned earlier in the interactive or other areas, I think it remains to be explored, but we can also see some of their long-term goals.

Host: So will there be a situation where game companies accelerate involution after reducing costs and improving efficiency, similar to the situation in the photovoltaic industry, including the problems caused by cost reduction? Which gaming companies are more competitive and better able to cope with this situation?

Huang Yue: I think there are two aspects. First, companies that can accelerate the adoption of new technologies will have a competitive advantage in the short term because their R&D efficiency and costs are greatly reduced. So these companies will be more competitive in the short term.

But more importantly, I think the core part of creative industries like games and film is still creativity. That is to say, the core is people, and as I mentioned before, people cannot be replaced by artificial intelligence.

Including our domestic game technology, such as visual technology and game engine and other core fields, of course, there is a gap with the international leader, but the gap may not be as big as everyone thinks. Although the domestic game industry has not yet appeared like the international AAA game giants, except for some unlisted companies that have recently launched some excellent games. But I think the gap between us and foreign game giants is mainly in terms of creativity and talent, which cannot be replaced.

So I don't think artificial intelligence needs to worry too much, just like the competition in the photovoltaic industry has intensified, I think what will intensify in the short term is the competition for efficiency, that is, reducing costs. For example, in the past, I may only be able to support 3 or 4 projects at the same time, but after the cost is reduced, I may support 10 or 20 projects at the same time, and there is no doubt that I have my own advantages.

However, in the end, which project will succeed, in fact, still depends on factors such as core talent, creativity, details and even operations, which are difficult to replace by artificial intelligence, which is the most essential part of the industry.

6. How to invest in the game sector

Host: At present, investors are more concerned about the game industry has risen relatively much, can it still enter now? In addition, are there other areas of artificial intelligence that can allow us to lay out more calmly?

Huang Yue: Okay, the whole artificial intelligence-related sector, I think the rally is relatively strong. If a sector doesn't rise much, it's not a core sector, there's no doubt about it.

We can see that the first hype in the field of artificial intelligence is computers, because everyone is focusing on the application side. But it turned out that no application can really be made, and most applications may be combined with artificial intelligence in less scenarios than everyone thought. Therefore, the market began to hype computing power (chip field), but soon found that domestic chips did not work, so it began to hype the communication sector, and then returned to the application side. Because games are the most observational in their field of application, AI has been successful in gaming applications.

So I think the market response is very effective, if a sector does not rise much, it is certainly not the most core sector. We can also understand why games and other AI-related sectors have risen so much.

First of all, I think they are definitely not industries that can complete the hype in one year, and you can compare them with new energy vehicles. They basically take three or four years of big markets. Second, we need to consider our strategy, that is, in such a big market, how do we enter and layout?

1) We need to see if we already have a position before, and if we already have a position, I think you can wait and see what happens, and you don't need to make adjustments. This is the first point, because although the short-term increase is very large, if you look at the monthly chart, you will see that it has been down for 8 years since 2015, and now it has only ushered in a very large rebound in a quarter. You can look at the monthly chart, or look at the annual chart, compared to 2015 or even 2016, 2017, its height is far from recovering, so I think you are more worried about the highs on the daily chart. However, this is also because after 7 to 8 years of chips clearing, the entire valuation declined, in exchange for such a sharp rebound. Therefore, if it is a long-term investor, and there is already a position, there is no need to toss, this is the first category.

2) The second type of investors are those who do not have a position or have a high cost of holding a position. For these investors, I don't think they need to be in too much of a hurry and can wait a bit.

Because in this quarter, the industry has risen 90%, and if you adjust 30% to 40% or more, that's normal. Even if the future expectation is a 3x or 5x increase, if a 1x increase has already been reached in a quarter, it is very normal to adjust by 30% to 50%.

But if you buy at a high level and adjust by 50%, then you need to rise 100% to get back, and the rise and fall are asymmetric.

So I think if you're bullish on these sectors, you can buy them in batches and don't buy them when they're already going up a lot. For example, the computer and communications sector has a relatively large pullback from the top, while the gaming sector has adjusted relatively small from the top, probably only in the range of 10% to 20%.

From a medium- to long-term perspective, I think you can buy, but in the short term you should buy it in batches. We can buy a part of the water first, because after all, the long-term market needs to test the water first, and then gradually increase the position.

Then regarding the period of adding positions, you can set it to a quarter, or 3 months, or 6 months. Funds are purchased in batches over a period of 3 or 6 months, either a little every month or a little every week, and the purchase is divided equally. Don't try to pinpoint its inflection point, because I find it very difficult. So if you're not in the market yet, or you're stuck and want to cut costs, I think it's more appropriate and not too aggressive. You can pay attention to Game ETF (516010) and Game Shanghai-Hong Kong-Shenzhen ETF (517500).

Seventh, is there still a chance for the new energy track?

Moderator: Thank you Mr. Huang for your answer. At present, there are three hot sectors in the market, namely artificial intelligence, China Special Valuation and new energy. These three plates are interrelated and can be said to be each other's blood packs.

Many investors may have previously allocated relatively more new energy and relatively little in artificial intelligence. In the second half of the year, whether we need to gradually improve the configuration of artificial intelligence, and whether the allocation of new energy needs to be reduced.

Huang Yue: It can be seen from the first quarter report that the proportion of new energy allocation is declining, while the proportion of artificial intelligence-related allocation is rising, and this trend of capital flow may be a medium- and long-term trend, rather than a short-term trend.

At present, the valuation level of new energy is relatively low and has a relatively high cost performance. It belongs to the price-performance ratio that fell out. Although new energy may have a certain cyclical nature, it is still a high-growth sector in the cycle.

In the entire A-share market, it is very difficult to find sectors that grow by 20 or 30% or more per year, so in fact, the target of new energy is still relatively scarce. Moreover, the new energy sector has good prosperity and policy support, and this sector is very scarce. Including its internal energy storage and photovoltaic fields, future demand is still good. And with Musk's cooperation in China, energy storage is likely to become an important area of cooperation with the Chinese government in the future.

In general, after a round of sharp decline, I think new energy can be bought and allocated. However, it is also more difficult for the new energy sector to get out of the trend market, not because this sector is not good, but because from the perspective of the game, its chip structure is not good.

At present, public funds still have a high allocation ratio in the new energy sector. I think if new energy wants to get out of the next wave of strong trend market, it may need to wait for the allocation ratio of public funds in the new energy sector to be reduced to at least the ordinary level, before a strong trend market may appear. But the absence of a trending market does not mean that there are no opportunities in this sector.

For the new energy vehicle sector, it is often an opportunity to over-fall rebound. Just like the liquor sector, baijiu also faced a lot of negative information when it reached the index high around 2021, including negative information such as policy and inventory. Although after 2021, the overall downward trend of baijiu, the space and time for each rebound is quite long.

Therefore, I think that new energy vehicles are still a target with strong operability. However, it is more difficult to get out of the trend opportunity, it is more likely to try it gradually after a low valuation, a long period of decline and a large decline, and then gradually reduce the position after the rebound. You can pay attention to the new energy vehicle ETF (159806).

8. How to configure the special valuation plate?

Moderator: Finally, let's talk about this section.

Huang Yue: For the special valuation sector, I think it mainly reflects two dimensions.

First, it benchmarks ESG (environmental, social and corporate governance) concepts overseas to some extent. Overseas ESG development includes social, corporate governance and environmental aspects, and although it does not seem to be directly related to corporate profitability, it is actually a concept formed based on shared values.

After the formation of the concept, a large number of institutional funds began to practice this concept, such as insurance funds and pension mandatory allocation, and also issued a large number of index funds and purchased related targets. After purchase, it becomes an alpha factor and becomes an effective analytical tool.

Therefore, I think that whether the valuation system with Chinese characteristics can truly become a valuation system with Chinese characteristics in the future may need to refer to overseas ESG concepts. For example, will our pension and insurance institutions mandate allocations according to this value, or at least screen targets according to this value.

In terms of passive index funds, the number of index funds issued by China Special Valuation has been quite a lot this year, and there have been many in the past few years. However, whether the special estimate can really become a special estimate, I think there are two aspects to be observed: one is consensus, and the other is whether there is an incremental inflow of real money.

In the short term, when economic uncertainty is high, people are more concerned about the profitability of companies. On the whole, the profitability of state-owned enterprises is more resilient to pressure during the downturn of the cycle than that of private enterprises, so they have their own certain value.

In the medium special valuation, I suggest that you pay attention to some medium special valuation companies with better profitability, such as refining and operators, which belong to the medium special valuation companies with outstanding profitability. However, I think it may be necessary to be a little cautious about some mid-range estimates that have a gaming component or where earnings sustainability is less obvious. Because these companies may not be as reliable unless their valuations have gone up, and if their valuations are not going up and they can't see earnings or sustainability of earnings.

Therefore, I think it is still important to focus on the sectors that are profitable in the special estimate. You can pay attention to the central enterprise win-win ETF (517090).

In addition, there are some growth sectors, you can pay attention to the military industry. The military industry is essentially a sector of China Special Valuation, but the military industry has not performed well this year. Recently, there are also some index funds of China Special Valuation and Central Enterprise Technology Growth to be issued, which contain 40% of the military component stocks, so it is bound to bring a certain amount of incremental funds. You can pay attention to the Military ETF (512660).

In general, I think you can pay attention to the mid-range valuation of the technology category, including sectors related to operators, computing power, and data assets, which have performance expectations. In addition, the military industry is also a sector that may attract growth funds.

In the field of traditional industries, I think China Special Valuation can pay attention to refining, and the asset quality is relatively good. In terms of refining, our purchase of crude oil from Russia is discounted relative to the market price, and the cost is reduced, which transfers the benefits to our refiners. Therefore, I think it may be advantageous to focus on these aspects.

The content and data are for reference only and do not constitute investment advice. AI technology strategy is provided for the cloud.