There are different voices in the Chinese industry and economics industry as to whether this century transaction should be released. According to media reports, on September 27, at the "Fourth Information Security Industry Development Forum", Ni Guangnan, an academician of the Chinese Academy of Engineering, said when talking about NVIDIA's acquisition of ARM, "I believe our Ministry of Commerce may reject this acquisition." .
To understand the significance of this transaction, we need to start with understanding ARM. With this original intention, this report will give an in-depth understanding of ARM:
ARM has shaken Intel's dominance by streamlining its instruction set. It has become the biggest winner of the mobile Internet, and it is very likely to continue to dominate the Internet of Things era. Its equity has also become the focus of competition among major technological powers.
What is more worthy of vigilance is that IP core (Intellectual Property core, intellectual property core) as a certain link in the chip field, we are basically in a blank, the industry and the market pay more attention to IP, far from enough.
British ARM company, its full name is Advanced RISC Machines, translated into Chinese as An Mou, is a world-renowned semiconductor IP provider.
As far as the Chinese public is concerned, the understanding of ARM may not be due to obscure and deep chip technology, but due to various "lace news." Since 2016, ARM has made as many as 3 news on global technology headlines:
◆In July 2016, taking advantage of the time window of Britain's Brexit, Japan's Softbank Group proposed to acquire ARM for 24.3 billion pounds (32 billion US dollars) in cash. After the completion of the acquisition, ARM was delisted. This transaction is also considered to be a representative work of Sun Zhengyi's bet on the Internet of Things.
◇In 2019, ARM made headlines again because of the tense trade relationship with Huawei. In May 2019, the Bureau of Industry and Security of the U.S. Department of Commerce included Huawei in the list of export control entities, and then ARM confirmed that it would suspend business cooperation with Huawei.
If Huawei cannot use ARM's chip design IP, then HiSilicon will be shocked on the spot. Fortunately, Huawei has the latest ARMv8 architecture-level permanent authorization, which can ensure that Huawei develops new products on this basis, with limited short-term impact.
◆In September 2020, Nvidia (NASDAQ:NVDA), the first share of semiconductor AI, announced plans to acquire the entire equity of ARM from Japan's SoftBank Group for US$40 billion. The industry's interpretation is that SoftBank will get the cash that is desperately needed, and Nvidia will use ARM's technology to reinforce the shortcomings of the CPU, and it will be the best in the Internet of Things era; and the market represented by China may encounter the fate of being monopolized. .
ARM, which stands at the top of the human science and technology tree, was first founded as a joint venture by companies such as Apple, Nokia, Acorn, VLSI, and Technology. Later, due to the ARM architecture used in Apple's mobile phone chips, it gradually moved towards the mainstream market.
ARM succeeded in defeating the unbeatable Intel in the era of mobile Internet and completely monopolized the chip design architecture of mobile terminals.
Therefore, ARM has naturally become the focus of the world's major technology countries. Every equity dispute surrounding ARM has become an important matter in the industry. The competitors include the United Kingdom, the United States, Japan, China and Europe.
Starting from ARM, we must first understand its business model.
In a word, what ARM does is IP authorization in the field of chip design, which can be simply understood as making a living by selling intellectual property. But in order to deeply understand why ARM can stand on top of the chip industry, we cannot avoid the discussion of technology.
Everything starts from the processor.
As the brain of a device, the processor gives the device intelligence; we go deep into the processor, generally including the kernel, debugging system, peripherals, memory and other components, and the kernel is the most important part of the processor.
Further disassembly, the kernel is composed of an instruction set and a micro-architecture; the instruction set is a collection of all instructions, and the instructions specify the operations that the processor can perform, as the interface between the software and hardware, the instruction set and the upper operating system, application programs, etc. Closely connected; micro-architecture is a hardware circuit design used to complete instruction operations, so that the processor can complete operations as efficiently as possible after receiving instructions.
What ARM does is sell IP core business, namely instruction set + micro-architecture.
Through the above introduction, we know that the IP core is the most difficult part of the chip, but the more familiar Intel sells the entire chip, while ARM only sells the core, or it is done in an authorized manner, which is more flexible , Light asset model.
Because ARM's IP core is based on reduced instruction set (RISC), compared to Intel X86's complex instruction set (CISC), RISC can simplify the entire processor and is suitable for application scenarios with low power consumption, cost-sensitive, and relatively fixed tasks in.
And this happens to be the pain point of mobile terminals represented by smart phones, so ARM seized the opportunity to become the most successful semiconductor company in the mobile Internet, while Intel (NASDAQ: INTC) encountered the "lost decade".
When it comes to ARM, what cannot be avoided is its huge advantage of "technology + ecology". Because the design of the instruction set is very complex, and the high-precision link of the architecture design is superimposed, leading to the fact that ARM always maintains the generational lead over competitors; besides technology, another advantage of ARM lies in its provision from chip design, manufacturing to sales All kinds of support, create customer stickiness with innovative strength + deep cooperation.
The strengths of ARM, in addition to the absolute technological leadership and huge ecological stickiness, the Damask researcher believes that the innovation of its business model is the focus of attention. This kind of business model that only authorizes no physical goods is strange and difficult to understand for Chinese companies whose intellectual property rights are still in their infancy.
ARM does not manufacture or sell any physical chips. It just designs its own IP, including instruction set architecture, microprocessors, graphics cores, and interconnect architectures, and then earns revenue by licensing these IPs.
The fee collected includes two parts: the pre-licensing fee for IP authorization, and the royalty based on the price of each chip. It also cooperates with customers through three different levels of authorization modes.
And the birth of this miraculous business model has profound industry development laws behind it:
As smart terminal products become more and more complex and diverse, the difficulty of chip design is rapidly increasing, and the design cost rises exponentially. In order to reduce design risks and costs, chip design companies are increasingly seeking to use verified IP, so the chip design industry Separate the chip IP industry.
This is very similar to the field of chip manufacturing. In the past, it was a vertically integrated IDM model. Later, TSMC established a business model that only did chip foundry. So we can think of ARM as the "soft version" of TSMC.
The predecessor of the ARM company was established in 1978, which is regarded as the first batch of semiconductor companies established; but the company became the prototype of today, it needs to be traced back to the establishment of ARM Holdings in 1990. At that time, the most important shareholder was Apple, which could be described as holding the golden key. Origin.
From 1993 to 2008, 15 years of rapid growth of the company, the company quickly became invincible in the low-power embedded processor market.
What everyone talks about most is that ARM has become the chip IP overlord in the mobile Internet era. In the OS+CPU system, the ARM+Android system defeated the Windows+Intel system in the PC era.
After the ARM architecture relied on Apple and Samsung to open the market, barriers gradually formed. Due to an open attitude to developers, rather than a closed attitude of Intel, it almost monopolized the mobile chip market. It is no exaggeration to say that the intelligence of every smart phone we use is given by ARM.
Without enumerating the dazzling financial data, we can use a picture to illustrate ARM's cattle x. According to ARM's own IP licensing-related chip shipment data published by ARM, as of 2019, the shipments of ARM-based chips have exceeded 1,660. Billion, accounting for 34% of the global total chip market shipments.
Naturally, a question that is easy for everyone to think of is, as the annual shipment of smart phones reaches the bottleneck of 1.4 billion units, will the growth of ARM come to an end? actually not.
As mentioned above, ARM is a streamlined instruction set (and the content of the instruction set is continuously updated), so it is naturally suitable for a large number of IoT devices, and mobile phones are just one of them. This is also the essential reason why the management of AI's first stock NVIDIA is so excited about the acquisition of ARM.
ARM's broad stage, we can see the micro-reading from the server chip market: The current global server annual shipments exceed 11 million, but the server chip is basically monopolized by X86. With a close to 100% monopoly, the server has become Intel’s cash machine. Nearly 50% of Intel's annual profits come from servers.
We see that everyone is trying to challenge Intel’s monopoly:
Huawei released the first server chip based on the ARM architecture—Kunpeng 920 in January 2019; Apple recently released the development of Mac chips based on the ARM architecture;
At present, ARM has clarified its iterative upgrade strategy of Neoverse architecture for data centers. The performance improvement of each generation is more than 30%, far exceeding the performance improvement of each generation of X86 architecture CPU. The future performance gap between ARM and X86 Will keep shrinking.
The server is just an appetizer. According to ARM's own judgment, its future markets include mobile terminals, ICT infrastructure, automobiles, embedded and other scenarios.
The market that ARM can participate in as a whole is expected to reach 232 billion U.S. dollars by 2029, and ARM, which currently earns less than 2 billion U.S. dollars, has endless space from the ceiling.
This is obviously not a idiotic dream. ARM’s simplified instruction set has low power consumption, low price, high concurrent processing efficiency, and fast upgrade speed. It is born with the Internet of Things genes. Taking the current 5G network infrastructure market as an example, ARM’s market share The rate has reached 28%.
Under the background of the global technological war, everyone’s focus is on semiconductor manufacturing and equipment. For example, TSMC cannot produce for Huawei HiSilicon.
But in fact, the software part of both sides of chip design and manufacturing is still close to blank in China. This field mainly refers to the IP of the underlying architecture and the EDA software that must be used in the field of chip design, especially after the acquisition of ARM by NVIDIA. The possibility of getting stuck in this link will increase significantly.
We know that the mainstream instruction set in the current market is the complex instruction set (CISC) represented by x86 and the simple instruction set (RISC) represented by ARM. They are basically in a completely monopolized seller's market. Most domestic CPU companies choose to buy from foreign countries. Authorized to design the chip, even the most advanced Huawei HiSilicon.
With the investment boom in the domestic chip industry, a large number of chip design companies have been established. By 2019, the number of domestic start-up chip design companies will reach 1,780, which is twice the number in 2015. If the core IP is missing, these companies will eventually become lighter. He can only make wedding dresses for others, but he will be cut off at any time.
Considering that most of the chips in my country are based on the IP or architecture authorization of foreign companies, promoting the localization of IP and chip infrastructure is a market choice and a national strategy requirement.
We see that most domestic chip design companies flock to which market in which downstream application fields are hot, and are unwilling to touch the hard-to-eat underlying core things.
Take the IP core industry that is not currently valued by everyone in the chip field as an example. Through reviewing the growth process of ARM, we know that ARM is because of a single mind to simplify the instruction set, breaking Intel's dominance of the complex instruction set, rather than avoiding the most important A subdivided chip design track.
Back in China, there are too few hard-core companies such as Huawei HiSilicon and China Microelectronics (SH:688012), and the market has given these targets (Huawei HiSilicon is not listed) expensive valuations due to scarcity.
As the chip investment boom cools, companies that only tell stories will lose support. What is worth looking forward to in the future is companies that do bottom-level breakthroughs like ARM, but possible investment opportunities will be concentrated in the primary market.
Of course, because the instruction set has a natural monopoly similar to the operating system, it is almost impossible to directly challenge ARM's position, just like ARM, in the first 30 years of development, it will not shake Intel's position in the server market at all.
But with the advent of the Internet of Things era, the demand for the instruction set will become more fragmented, so the concentration of the IP core industry should be more dispersed, which instead gives domestic companies entrepreneurial opportunities. Currently, there is only one company listed on the Science and Technology Innovation Board, VeriSilicon (SH: 688521) engaged in IP licensing-related business, and whether it can successfully break through requires follow-up observation.
As mentioned above, with ARM's technology generational leadership, the natural ecology of the instruction set, and the possibility of a strong alliance with NVIDIA in the future, it is basically impossible for domestic start-up companies to replace smart phones and server IP cores.
However, with the fragmentation of chip demand in the Internet of Things era and the deepening of the division of labor in chip design, what domestic companies can still do is differentiated competition.
Differentiated competition is reflected in two markets:
In processor chips, the number of IP applications is increasing. Taking mobile phones as an example, current SoCs are designed based on a variety of different IP combinations. With the evolution of advanced manufacturing processes, the number of IPs that can be integrated on a single SOC chip has also increased significantly.
In other fragmented scenarios, IP will continue to emerge. The trend of separating IP from chip design is still advancing, and future storage, GPU, and analog IP requirements will gradually be reflected.
Intuitively from the data, according to IBS statistics, the chip IP market will grow from US$4.6 billion in 2018 to US$10.1 billion in 2027. The processor IP market is still the largest single market and is expected to reach US$6.3 billion in 2027; But other market segments should not be underestimated. It is estimated that by 2027, the digital-analog hybrid IP market is expected to reach 1.3 billion US dollars, and the RF IP market is expected to reach 1.1 billion US dollars.