The growth rate from 0 to 1 is always the fastest
The recently released Google Gemini has sparked global enthusiasm, once again igniting market enthusiasm. Meanwhile, the TPU V5P version was announced along with Gemini.
Although Google appears to have achieved a phased victory in the field of AI, its success has definitely not been able to bypass one thing — its long-term investment and bet on infrastructure and TPU resources.
Previously, Google revealed that more than 90% of its artificial intelligence training tasks were completed through Google's self-customized TPU chips.
Since ChatGPT was launched, Nvidia's stock price has been breaking, and GPUs have also stolen the limelight in the chip industry, overshadowing the power of other important AI chips (such as ASIC chips).
However, with the rise of Google's large multi-modal model, Gemini, the industry or investment market is gradually heating up, and the development process of various AI chips and related chip companies have also returned to the spotlight.
01
ASIC chips: a key stronghold in the battle for AI
In the ChatGPT craze, “computinghunger” (computing power hunger) has become a major problem in AI technology innovation — computing power hunger may lead to an efficiency crisis and a cost crisis.
According to Open AI estimates, since 2012, the global computing power requirements for head AI model training have doubled every 3-4 months, and the computing power required for head training models has increased by as much as 10 times each year, while the total computing power consumption during the ChatGPT training stage is about 3640 PF-days (that is, 1 petaflop/s efficiency runs for 3640 days).
We generally think that the AI server provides computing power; in fact, in this case, the chip provides the core computing power support.
Under simple attribution, it can be said that “there are no big models without chips,” so there is no current AI revolution. It is because of this that beautiful countries have changed their tricks over and over again, trying to “control” advanced computing power chips to seize the “neck” of AI development in other countries.
In detail, current mainstream computing power chips include traditional chips such as CPUs, GPUs, and FPGAs, as well as ASIC chips specially designed for the field of artificial intelligence.
In the context of the emergence of various large models, GPUs and ASICs were shipped the most. Compared to GPUs, ASIC chips have different advantages — through algorithm solidification, they often achieve high parallel computing power, extreme performance, and energy efficiency. Simply put, ASIC is characterized by being cheap, easy to use, and dedicated. Google's TPU is representative of excellent ASICs.
In addition to Google, the world's leading technology companies are also rushing to develop AI chips in the ASIC field.
For example, in 2019, Intel acquired Habana Labs, an artificial intelligence chip company, and launched Gaudi 2, an AI ASIC chip with excellent performance in 2022; at the end of 2022, IBM Research released an AI ASIC chip called AIU, which is expected to be launched in 2023; and Samsung's first AI ASIC chip, the WarBoy NPU chip, began mass production in 2023.
Driven by technological progress and demand explosion, ASIC chips have been released rapidly in recent years.
According to KBVResearch report data, the global ASIC chip market is expected to reach 24.7 billion US dollars in 2019-2025. A few years ago, ASIC has been proven in the blockchain field with its high parallel computing power and excellent low power consumption and efficient computing performance, and has achieved widespread success in rapid development.
Of course, the former requires computing power, while the development of artificial intelligence requires more computing power.
Generally speaking, wherever computing power will appear in the future, chips should appear. Furthermore, it can be discovered that ASCI chips are expected to reach a higher level in the future.
According to McKinsey Analysis data, the share of ASICs in AI chips will increase dramatically: on the data center side (that is, in the cloud), ASICs accounted for 40% and 50% of reasoning and training applications respectively in '25. In marginal fields where it is expected to be more widely used and will have a significant impact on real life, ASIC accounted for 70% and 70% of reasoning and training applications in '25, respectively.
Driven by the wave of AI, ASICs are rising at a speed “visible to the naked eye.” As Google's big model and TPU once again ignite market enthusiasm, this is also quickly recognized by the market.
Since last week, the Philadelphia Semiconductor Index has shown a relatively rapid rise. Nvidia, the representative of GPUs, has risen 2.9%, and the Chinese securities stock ICG has risen as high as 28%.
Judging from market trends over the past three months, ICG was significantly ahead of Nvidia and the Philadelphia Semiconductor Index for most of the time or in terms of overall growth, further confirming that ICG has become the leader in ASIC chips.
02
A rapidly rising “arrow through the clouds”
Among China's ASIC chip manufacturers, the rapid rise of ICG is like an arrow through the clouds.
In the past three years, the company's revenue has increased more than tenfold, or the highest in the ASIC field.
What does ICG do, what fields is it in, and why is it rising so fast? Does it have the potential and advantages for future development? Maybe you can use a chart to answer that.
The Fabless model (also known as the fab free model) adopted by the company means that chip companies only design and develop chip products, and leave manufacturing and packaging testing to foundries and packaging factories. This business model can effectively reduce costs and product development cycles.
The Fabless model, which has the advantages of both low cost and high efficiency, has made it a concentration camp for the market to find high-multiplier bull stocks in the chip industry.
ICG focuses on the high-performance computing ASIC chip track with high computing power and excellent performance, as well as supporting software and hardware solution services.
There are a few places to focus on. The first is that the company only makes advanced process chips, only high-value digital chips rather than analog chips;
Second, the ASIC chip circuit where the company is located provides customized special chip products and services, so its customer loyalty and the degree of bundling between the company and customers will be far higher than that of general purpose chips. Technological and reputable ASIC chip companies can obtain relatively better conditions and rewards from customers, which can reflect the advantages of pricing power. These companies will have a better profit level, and therefore more valuable.
The third point is that what ICG will enter next is the main channel with future development potential, that is, the AI chip track.
It can be seen that every foothold chosen by ICG, without exception, has found the shortest path to implementation due to the massive migration caused by the development of the times, and has initially verified that the company and team do have the right vision of judgment and the ability to seize big opportunities at the right time. This is also the main reason why an unknown chip company has been able to stand out in just a few years.
Insisting that “choice is better than effort”, perhaps the rise of ICG in the early stages shows that the market is well aware of the importance of this.
In addition, ICG, a company that places special emphasis on technology, has provided a highly experienced technology research and development team and a leading “Xihe” proprietary technology platform. Most of the company's R&D personnel graduated from well-known universities such as Fudan University, Shanghai Jiaotong University, Wuhan University, and the Chinese University of Science and Technology. A “pure-blood domestic” team of local chip talents, isn't this the self-developed chip company we urgently need?
If the R&D team with a leading scientific research background forms the company's strong “brain,” then the “Xihe” proprietary platform forms the company's powerful “toolbox.” The platform covers all aspects of chip design, from initial concept to final manufacturing, and provides comprehensive and complete design support for R&D personnel.
Through the “Xihe” platform, the company is able to launch products with a shorter time to market, lower overall cost, and relatively higher gross margin compared to industry competitors.
With the innovative power of the R&D team and the support of the “Xihe” proprietary technology platform, the company was able to maintain competitiveness in the ASIC field and promote the development and application of advanced technology.
As of September 30, 2022, ICG has completed 8 streams using the “Xihe” platform's 22-nm ASIC chip, with a 100% success rate of all films.
The chip industry is no stranger to streaming.
The so-called flow film is to manufacture a chip through a series of process steps like an assembly line. This process is in the middle stage of chip design and chip mass production, and is a key part of chip manufacturing. In order to test whether a chip design is successful, it is necessary to carry out streaming. This is why chip design companies generally need to invest a lot of money in the early stages.
When the chip is fully designed, it is necessary to etch it on the wafer according to the drawing. What kind of process is used, the size of the wafer, and the complexity of the chip will affect the success rate and cost of this chip. Moreover, many chips are not able to be successfully filmed at once, so it is often necessary to perform multiple flow films to obtain the desired results.
In the industry, the success rate of a high streaming film is a “hard indicator” that measures the technical strength of a chip company, and a 100% streaming success rate, whether among chip companies that have risen rapidly in recent years or in the growth process of industry leaders in the past, is extremely rare.
According to publicly available statistics, for a chip with a 14 nm process, the flow sheet costs about 3 million US dollars at a time; for a 7 nm process chip, the flow film costs 30 million US dollars at a time, while for a 5 nm process chip, the flow film reaches 47.25 million US dollars at a time. As can be seen, streaming is a huge expense and burden for chip design companies. Controlling the flow rate has become the only viable path for chip companies under the Fabless model to save costs, improve profitability, reduce business risks, and seek more high-quality customers.
For streaming films that can easily cost tens of millions of dollars, the success rate of streaming is a key indicator for measuring the overall strength of chip companies. The success rate of high streaming films can almost be equated with high-value scarce chip companies.
At present, ICG has successfully developed a variety of efficient and expandable ASIC chips, and has quickly occupied the market with this. According to ICG's prospectus, the company has ranked first in several market segments.
According to Frost & Sullivan data, in terms of cumulative sales power in 2020 and 2021, the company has leading market share in ASIC chips designed by various algorithms such as Blake2BSHA3, SHA512MD160, Cryptonite V4, Eaglesong, and Blake2s.
Thus, it is easy to see that ICG's technical strength in the ASIC field has been fully recognized by the market.
However, by deeply cultivating the design experience and customer barriers accumulated in the industry over many years, ICG can easily be extended to the field of AI ASIC chips. These logics are common.
For example, blockchain ASIC processing involves a large amount of parallel computation, and generative AI is also a typical parallel computing application. The higher the degree of parallelism, the higher the efficiency of model training and inference tasks; blockchain ASICs and AI ASICs both involve acceleration of specific cryptographic algorithms. Although the specific algorithms are different, they all use common hardware acceleration technology. Furthermore, the two have strong similarities in terms of high-bandwidth storage and memory access, power optimization, etc.
Because of this, chip companies that have achieved success and rapid rise in the previous stage, such as growing companies such as ICG, are fully expected to replicate past successful experiences and then “shine brightly” in the golden age of AI.
03
Chip bull stock code: we need to balance win rate and odds
Currently, the market is fluctuating, and the degree of risk aversion of capital has changed greatly. How exactly can we find chip bulls in the golden age of AI?
The answer is to balance odds and odds — something that meets the “double high” characteristic, has a margin of safety, and has a huge gap in expectations.
It is not easy to find targets with high returns in the chip industry, because making integrated circuits is a very expensive thing. For the time being, the profitability of most listed ASIC companies is still not in an ideal state.
However, there are some chip companies that have achieved leadership in technology research and development and commercial implementation at the same time. ICG is one example.
As can be seen, ICG is in a stage of rapid growth. Its operating income has increased more than 11 times in the past three years. By the end of '22, the company's gross margin was as high as 80%, far exceeding that of other year-on-year companies. ROE was also significantly higher than the comparable reference target for A-share listings during the same period, and its profitability was outstanding.
The author has always believed that for chip companies, the high gross profit margin in financial indicators is a concentrated expression of the company's competitiveness, business model, track, and market position in the industry, etc., and an expression of advantages.
A higher gross profit margin means that the chip company can actively compete for “volume” competition. Over time, chip companies with insufficient profit points will voluntarily relinquish market share without strong capital expenditure.
The method for betting from the perspective of investment success rate is simple. It is to prioritize the identification of chip companies with high growth, high gross profit, and the ability to maintain a “double high” over a long period of time, and be able to find the answer behind “double high.” For example, in the case of ICG, what we are interested in is the success rate of its high-profile film, which is almost overpowering.
Driven by high growth and high gross profit, it can also be seen that ICG has a rare and very healthy financial report for chip companies - from 2020 to 2022, the company's net profit increased from 8.247 million yuan to 355 million yuan, an increase of 34.3 times! This growth rate is far higher than the growth rate of operating income, demonstrating a very prominent positive cycle effect and scale effect. The astonishing increase also increased the company's profit position and placed it at the top of the segmentation circuit.
At the same time, the company has demonstrated strong hematopoietic ability. In 2022, the company's net operating cash flow reached 327 million yuan. The value of competitors during the same period was mostly negative.
Under the combination of these positive factors, the company has a good asset structure that most companies in the industry do not have - the company has cash of 700 million yuan, and the balance ratio at the end of 2022 is only in single digits.
Adequate cash flow and reasonable asset structure can easily meet future capital requirements for AI chip development and provide a strong financial guarantee for future strategy implementation.
Let's take a look at our understanding of odds. The odds of an investment come from two main directions. The first is the margin of safety, and the second is whether there is a huge gap in expectations.
ICG's margin of safety can be said to be a gift from “shifting gears” in the era of computing power.
When the company went public this year, at a time when the ASIC industry was being dragged down by blockchain, it was only able to depress the valuation and “tearful” listing. After listing, it was further greatly dragged down by previous US stock digital currency or blockchain market conditions.
Currently, ICG has been gradually reclassified by the market as a chip company developed in-house in China.
Although its stock price has been drastically revised in recent times, it still shows a high cost performance ratio and valuation advantage from a valuation perspective.
Based on the closing price on December 15, corresponding to 2022's profit, ICG's PE is still only about 13 times, far below the valuation of benchmark chip companies by peers that can easily be tens or hundreds of times. The company's PS is only in single digits. It is in a “depression” in the valuation of Chinese chip companies, and the margin of safety is sufficient.
As to whether there is a huge gap in expectations, etc., this answer appears to be “discerning”. For example, the author believes that the reason ICG has been ignored by the market is mainly the “black under the light” problem — the de-inventory process in blockchain-related downstream applications is nearing its end. ICG is expected to be driven by the recovery of the blockchain market and ushered in a fundamental reversal and valuation repair. You may wish to keep an eye on the recent performance of the digital currency market, so you can know the general recovery process.
Finally, the second gap in expectations comes from changes across industries: Currently, ICG is rapidly developing the AI ASIC field. In the AI field, in addition to the huge computing power demand generated by the cloud, both the end side and edge side are on the eve of an explosion. Every once in a while, there will always be quite a few breakthroughs and new products. There will always be quite a few breakthroughs and new products. There will always be enough demand for dedicated ASIC chips, whether dedicated algorithms or special systems.
The interpretation of the trend of chipization of algorithms and chipization of systems is clear enough. We are always looking forward to a Chinese chip company with know-how and good potential to lead in markets and fields where the AI chip track has yet to be conquered, bringing more encouragement and stimulation to the Chinese semiconductor industry.
Leaving aside these hopes, from a realistic investment perspective, ICG is expected to return to the Davis double click state after the difficult situation is reversed, yet the company's own growth is relatively certain, and it also has full advantages in profitability and valuation position. Once it is recognized by the market and the compounded differences in expectations are gradually realized, its valuation flexibility will be shown.
Because the growth rate from 0 to 1 is always the fastest, it is far superior to the growth logic of other stages. This is also the key reason why the company's stock price has recently been able to lead ASIC and even outperform the AI chip sector as a whole. (End of full text)