From 2010 to 2020, China’s mobile internet and new business flourished. In terms of digital economy size, China has surpassed the United States, with a ratio of 1.35:1 in digital economic output between China and the US, while the key transition from a “networked nation” to a “cyber power” is chips.
Chips have always been the largest import item for China. According to publicly available customs data, in 2019, China’s chip imports amounted to $304 billion, surpassing the combined total of crude oil, iron ore, and grain at $301.6 billion.
For every three chips produced globally, one is used in Chinese industries. The United States controls the development of the global technology supply chain through its “chip + OS” strategy. China is the largest customer of American chip companies, becoming the main payer of the “chip tax”.
SIA data shows that in 2019, the US held a 47% market share in the global semiconductor industry, especially leading in photolithography equipment, EDA software, and CPU/GPU computing chips.
American semiconductor companies hold a 48% market share in China, with companies like Qualcomm, Microchip, Micron, and Qorvo generating over 50% of their revenue from China, while the gross profit margin for American semiconductors is 54%.
Although China has rapidly developed in the manufacturing of servers, mobile phones, PCs, laptops, and wearable devices, it faces the predicament of being “choked” by American chips and operating systems, paying a high “chip tax” and “OS tax” to the United States every year.
With the US ban on chip sales to leading Chinese companies, it is expected that in the next three years, US semiconductor companies will see a revenue reduction of $37-40 billion, with global market share dropping from 48% to 40%. The reduced market share will mainly flow to South Korea (possibly surpassing the US) and Japan, while also bringing China’s semiconductor self-sufficiency rate to over 25%.
Figure 2019 Revenue Share of Major US Semiconductor Companies in China
The US supply chain has three leading advantages: microprocessors, EDA design software, and photolithography machines, which are the three major mountains that Chinese chip companies must overcome:
Microprocessors: In 2019, the US market share reached 98%, especially in wireless communication chips, analog chips, logic, MCU, and memory, which are highly competitive, but can consider replacing American products with semiconductors from South Korea, Japan, and Europe.
EDA Software: This is a set of software tools for the functional design, synthesis, verification, and physical design of integrated circuits. The global market share of two American EDA companies, Synopsys and Cadence, exceeds 54%, with advantages in full-process design solutions and integration with foundries and IP companies.
Photolithography Machines: The core equipment for chip production, using technology similar to photo development to imprint fine circuit diagrams from masks onto silicon wafers.
In 2019, the global photolithography machine market was dominated by ASML of the Netherlands, which held 81% of the market, followed by Nikon (5.9%) and Canon (11%). ASML is already controlled by the US, with Morgan Stanley Capital International and BlackRock being the first and second largest shareholders.
On the other hand, the ASML photolithography machine, which consists of 50,000 parts, has its core component supply chain monopolized by the US. Due to its agreement with the US Department of Energy, 55% of the fine components must be supplied by US companies, including masks, extreme ultraviolet light sources, lasers, micro-laser systems, and electromechanical devices.
In the 40 years since the popularity of PCs, the first half was dominated by “Windows + Intel”, while the second half saw the rise of “MacOS + ARM”.
In 1981, IBM, the PC leader, launched the first generation of personal computers using the Intel 8088 chip. With the hot sales of IBM PCs, Intel entered the “Fortune 500” and began the rapid development path of Moore’s Law.
In 1980, the operating system development for IBM PCs chose the startup Microsoft as the software supplier, eventually forming the “Win-Tel Alliance” that dominated the compatible machine market and overturned IBM’s PC leadership.
Another PC era leader, Apple, adopted its self-developed Mac operating system and mainstream manufacturer chips on Mac models, using Motorola CISC chips in the 80s, PowerPC chips jointly developed by Motorola and IBM in 1994, Intel X86 chips from 2005 to 2015, and Apple’s self-developed Silicon chips (graphics card, CPU) based on ARM architecture in 2020, with TSMC as the foundry for 5nm process. In the last ten years, Apple has delivered 2 billion system-level chips (SoC) and tens of billions of other chips.
He who has the chip holds the OS, he who has the OS holds the applications, and he who has the applications (ecosystem) holds the world.
The MacOS and computer applications supported by the X86 instruction set, and the iOS and App mobile applications supported by the ARM instruction set, have split Apple developers into two segments, requiring re-compilation for application migration.
In the mobile era, the Mac’s share in the Apple ecosystem has fallen below 7%. Therefore, on November 11, 2020, Apple officially released the new generation MacBook, embedding its self-developed M1 chip based on ARM architecture, achieving a new SoC integration (160 billion transistors) with CPU + GPU + Neural Engine + T2 through 5nm technology.
From now on, all Apple computers, laptops, tablets, and phones will achieve a unified ARM instruction set, iOS, and application ecosystem, with application development running across all terminals, reflecting the powerful vitality of Apple’s self-developed chips.
Considering that the chip computing power of the iPad has surpassed that of the X86 architecture Mac, future Apple wearable devices (including AR glasses) are likely to build an “edge intelligence center” around the iPad (or iPhone).
From another perspective, 99% of the world’s mobile phones and tablets use ARM architecture, regardless of whether they are from Apple, Samsung, or Huawei, and they will compete for market share in desktop computers in the future.
The “iOS + ARM” and “Windows + Intel” camps will compete in personal computers, data centers, and other fields. If Nvidia successfully acquires ARM, it will dominate the GPU and CPU markets on the personal computer side.
The development of the chip industry is a process of continuous specialization and division of labor. The “Fabless + Foundry model” (chip design + foundry) in the mobile internet era has replaced the “IDM model” (Integrated Design and Manufacture) of the PC era, with ARMdroid architecture rapidly popularizing, while major mobile manufacturers self-develop chips to enhance technological barriers and product competitiveness.
According to Investopedia, in 2019, over 95% of smartphones and tablets worldwide adopted ARM architecture.
However, unlike the decline of PC manufacturers and the rise of chip companies in the PC era, in the mobile internet era, mobile OS manufacturers dominate self-developed chips, even offering free open-source operating systems, leading to a transformation in the “chip + OS” ecosystem.
As early as the 3G feature phone era in 2010, smartphones with the Android operating system already held 70% of the global market, and the ARMdroid architecture (ARM + Android) became the market mainstream.
Since ARM belongs to the Fabless model, it does not sell chips but provides IP licensing for semiconductor design. Therefore, Google (Whitechapel), Apple (A series), Samsung (Exynos), and Huawei (Kirin) have obtained ARM architecture instruction set licenses, integrating customized requirements into chip development, transforming the ARM architecture and self-developing mobile chips and server chips.
For example, Apple obtained instruction set licensing, expanding swift, Typhoon, Twister, and other architectures on the ARM architecture, forming the globally leading A series chips used in all Apple products including iPhone, iPad, and Mac.
Huawei purchased permanent licensing for the ARMv8 instruction set, developing the Da Vinci architecture for high-end mobile models.
Qualcomm, based on ARM instruction set licensing, self-develops architectures such as Scorpion, Krait, and Kryo. The sales volume of mobile phones is positively correlated with the shipment of self-developed chips, making Qualcomm, Huawei HiSilicon, and MediaTek the top three chip manufacturers for Android phones.
Self-developing the complete set of chips has become a standard strategy for leading mobile manufacturers, with Apple entering the 5G market with self-developed chips, while Huawei HiSilicon addresses its shortcomings.
By 2018, Apple had sold 2 billion iOS devices. In September 2020, the number of active iPhone users exceeded 1 billion, with an annual increase of 20-30 million new users.
Thus, the A series chips based on ARM architecture will see rapid growth, with the ARMpple architecture (ARM + Apple) and ARMdroid architecture (ARM + Android) fiercely competing for ecosystem resources, including users, developers, application markets, and new markets (such as autonomous vehicles, smart home appliances, and wearable devices).
Currently, Apple’s A series computing chips supply all terminals (Mac + mobile), while communication chips still rely on Qualcomm, but it is expected that from 2023 to 2025, Apple will replace Qualcomm with its self-developed 5G baseband chips.
On the other hand, ASML of the Netherlands announced in October 2020 a large-scale sale (700 units) of DUV deep ultraviolet photolithography machines to China, which are previous generation products capable of producing chips above 10nm, while the most advanced EUV photolithography machines (3-10nm chips) are still banned from sale to China.
Therefore, Chinese chip companies like the Chinese Academy of Sciences and Huawei still need to face challenges independently to fill the gaps in photolithography machines, EDA design software, and production processes.
Figure Chip Design Costs (unit: million USD)
ARM controls the underlying architecture of the IoT ecosystem through licensing. Nvidia’s acquisition of ARM from SoftBank has a significant impact on the global chip industry chain, significantly enhancing American influence. ARM may become a leverage for the US to constrain (or ally with) European and Asia-Pacific countries.
It is expected that by 2035, there will be over 1 trillion smart electronic devices interconnected globally, including sensors, access cards, mobile phones, wearable devices, smart home appliances, autonomous vehicles, industrial machinery, communication base stations, medical devices, data centers, and cloud servers. Chips and AI-driven 5G IoT will become the infrastructure and “industrial lifeline” for every country and industry.
ARM meets the characteristic needs of the IoT through three series of chips:
Cortex-A series chips (A: Application): Designed to meet the wireless entertainment needs of consumer electronics, featuring high computing, frequent interaction (video images), and thick operations (systems), including smartphones, automotive entertainment systems, smart TVs, etc.
Cortex-R series chips (R: Real-Time): Designed for applications requiring real-time operations, including automotive power systems and large-capacity storage controllers for embedded real-time applications.
Cortex-M series (M: Microcontroller): Designed for microcontroller applications, characterized by low cost and low power consumption, including human-computer interface devices, automotive and industrial control systems, smart home appliances, and smart medical devices.
ARM controls the global chip industry chain through licensing. According to SenseTime’s Intelligent Industry Research Institute, in 2020, global ARM licensed chip shipments will exceed 340 billion units.
Chip design is upstream in the industry chain, and ARM, as a representative of IP licensors, is at the upstream of the chip design process. Thus, ARM controls both mobile manufacturers and chip foundries through licensing to enhance control over the entire industry chain.
With the development of Moore’s Law, the process technology continues to delve into 5nm and 3nm, and the cost of chip IP design is rapidly rising, requiring larger global production and sales volumes to share the IP design and R&D costs.
4G has brought standardized chip demands for consumer internet, while 5G has brought customized chip demands for massive industrial internet, further driving the geometric increase in chip IP usage and further refining the division of labor in the chip industry chain.
On the other hand, although ARM has long maintained its leading position, its global IP market share has dropped from 50% to 44%, leaving opportunities for innovative Chinese enterprises.
Data Center Chips: Controlling “All-domain Chips”
Global chip manufacturers have entered a “super Matthew cycle”, competing in “all-domain chips” through mergers and acquisitions, including “cloud-edge-vehicle” chip products.
AMD is the chip leader in the mobile internet, Nvidia leads in desktop computers (graphics cards) and artificial intelligence (GPU), while Intel leads in CPUs. The three are all invading each other’s fields through mergers, accelerating the strong expansion of leading chip manufacturers under the “super Matthew effect”:
1. Nvidia: In 2019, Nvidia fully acquired Mellanox, a company that provides InfiniBand and Ethernet communication (including virtual networks) and lightweight AI chip (NPU neural network chip, SoC) technology services.
In 2020, Nvidia’s $40 billion acquisition of ARM aims to conquer the future massive smart IoT (AIoT) market, evolving from the notebook and data center market to the autonomous driving and IoT market, thus completing the “GPU + NPU + CPU + SoC” family of all-domain chips.
2. AMD: To counter the ecological offensive of Nvidia + ARM, AMD plans to spend $35 billion to acquire Xilinx, expanding from the existing notebook and server (CPU + GPU), and gaming console (Xbox + PlayStation) chip markets into 5G wireless communication, data centers, automotive aerospace, and industrial control FPGA / programmable SoC chip fields.
Because of the semi-custom characteristics of FPGA, it is very suitable for AI deep learning and neural network calculations. From the perspective of the global AI acceleration chip market in cloud computing centers, Nvidia’s GPU accounts for 86%, AMD’s GPU and Xilinx UltraScale account for 9%, Intel’s Arria10 accounts for 4%, and Google’s TPU and Amazon’s AWS Inferentia account for 1%.
The future consumption and development space for autonomous driving vehicle chips are enormous, so the current mergers and acquisitions aim for tomorrow.
3. Intel: In 2016, Intel acquired Altera (the world’s second FPGA company) for $16.7 billion and acquired Mobileye for $15.3 billion in 2018 to enter the advanced driver assistance vehicle chip market.
4. Other terminal manufacturers: Apple and Huawei are advancing into the smart home, wearable devices (including AR/VR glasses), and data center (GPU-like chips) markets based on self-developed ARM chips, with Huawei moving towards IDM chip manufacturing.
Alibaba’s Tsinghua Unigroup is also developing the RISC-V architecture-based Xuantie series chips, Wu Jian SoC, and Ling Guang 800 (AI chip), following a similar IP core design licensing model as ARM.
Figure X86 Chips Remain the Mainstream Configuration for Global Servers
The global chip market is developing sequentially according to PCs/laptops, data centers/cloud computing, and mobile/IoT/automotive, and is currently at the intersection of the second and third waves.
Nvidia (over $300 billion), Intel (over $200 billion), and AMD (about $100 billion) are all competing for market share in data centers and exploring the blue ocean market for autonomous driving chips.
Global server shipments exceed 11 million units annually, with an industry scale surpassing $70 billion. Intel’s X86 server chips account for 85% of the global market (with Intel’s X86 server market share at 98.7%).
Due to international technological threats and cost pressures (processors account for 40% of server costs), Chinese cloud computing companies like Huawei and Alibaba Cloud are increasingly self-developing ARM architecture and RISC-V servers, which severely threatens Intel’s monopoly in the server market.
Figure The Four Major Cloud Computing Data Centers Globally Use Nvidia Tesla GPUs
AI chips are becoming a blue ocean market for data centers and supercomputing centers.
Data centers are in the process of AI transformation, with the GPU market showing a two-horse race between Nvidia and AMD, with Nvidia accounting for 70%. The top four global cloud computing data centers (Amazon AWS, Microsoft Azure, Alibaba Cloud, Google Cloud) extensively use Nvidia Tesla GPUs.
According to data analysis from McKinsey, Allied Market Research, and ABI Research, the scale of the AI chip (training + inference) industry in data centers will reach $14 billion in 2023, with Nvidia almost monopolizing the training market and continuously penetrating the inference end, competing with ASIC and FPGA.
The foundation of AI is massive data processing, and heterogeneous computing has already posed challenges in performance complexity and cost expenditure.
In October 2020, Nvidia launched the DPU (Data Processing Unit) chip product, integrating up to 125 CPU chips (BlueField DPU) and plans to add GPUs in 2021 (BlueField-2 DPU). The development ecosystem based on DPU + DOCA will require medium- to long-term cultivation, and the next three years will present a limited time window for domestic AI chip companies to compete.
Where is China’s Chip Industry Heading?
Faced with the complex international technological competition, China’s chip industry has three breakthrough methods:
Open Alliances (Changing Ecosystems): In the short term, learning from Intel’s formation of the “Extreme Ultraviolet Alliance (EUV LLC)” with the entire American chip industry, China should form a new generation chip R&D industry chain with non-sensitive countries. For example, leveraging the RCEP free trade framework, integrating industrial chains with countries like Japan and South Korea to form a multilateral R&D alliance in Asia comparable to the EU and North America, investing in shared costs and benefits.
Jumping Position (Changing Battlefields): In the medium term, the smartphone industry has matured, while the 5G era’s smart connected vehicles, AR glasses, smart home IoT, and smart sensors are on the rise. Therefore, targeting the next generation of AIoT terminal’s “algorithm + chip” demand for R&D will gain the era’s dividends of “curve overtaking”.
For instance, Huawei selling Honor, developing full-stack smart automotive technology, and researching automotive chips with higher power tolerance will shift the chip competition to the future battlefield, which is a clever strategy worth emulating for more Chinese mobile, new energy vehicle, and AI companies.
R&D Shift (Changing Technologies): In the long term, TSMC recently announced plans to trial 2nm chips in 2023 and conduct research on 1nm processes, indicating that silicon-based chips are nearing the “ceiling”.
Choosing new materials for future chip R&D will likely lead to “ultimate competition”. Carbon-based chips at 90nm can achieve the level of silicon-based chips at 28nm, bypassing the limitations of precision instruments.
In October 2020, the Shanghai Institute of Microsystem and Information Technology launched domestically produced 9-inch graphene monocrystalline wafers. Meanwhile, the Peking University research team plans to complete the 90nm carbon-based CMOS process R&D in 2-3 years. However, the significant upgrade and transition from theory to mass production face new challenges. For example, increasing the purity of carbon nanotube semiconductors to six nines and solving engineering problems related to chip design equipment and production process management could take ten to several decades.
(Tian Feng is the director of the SenseTime Intelligent Industry Research Institute, and Guo Junxiang and Liu Jun are assistant researchers at the SenseTime Intelligent Industry Research Institute)
Written in the 1000th issue of Economic Observer: This is indeed the time to bloom.
Looking to the future in 2020: How to maintain rationality and optimism in the “no man’s land”? Li Lu discusses crises, innovation, individuals, and the world.
Zhao Cen, secretary of the board and financial director of Guosheng Jin控, was dismissed after raising objections to the third quarterly report. The Shenzhen Stock Exchange issued a letter of concern.
