Statistics show that AI chips are forecast to account for up to 20% of the total semiconductor chip market by 2025. Compared with other traditional chips, AI chips reduce the cost of operations, promote efficiency, and minimize potential risks in various industries. Nevertheless, high development costs and a lack of skilled workforce impede the AI chip market. The global AI chips industry was valued at USD 9.29 billion in 2019 and is estimated to reach USD 253.3 billion by 2030. The demand for AI chips is rising dramatically due to their widespread adoption, including image recognition, recommendation engines, natural language processing, and autonomous vehicles (AVs).
AI Chips Basics
AI Chips, also known as AI hardware or AI accelerators, are function modules that incorporate machine learning capabilities and process massive computing tasks in AI applications.
With respect to the technical architecture, AI chips can be categorized into Graphics Processing Units (GPUs), Field- Programmable Gate Arrays (FPGAs), Application-specific Integrated Circuits (ASICs) that are specialized for AI, and brain-inspired chips. According to the location in the network, AI chips can be specified by cloud AI chips, edge and terminal AI chips. Moreover, AI chips can also be divided into training chips and information chips in terms of its goals in practice. With the growing maturity of AI hardware acceleration technology, the new technological innovations are demonstrated by the combination of circuit and device-level technologies, such as brain-inspired computing chips.
General-purpose chips, such as Central Processing Units (CPUs) can also be used for some simpler AI-specific tasks. Although increasing processor frequency can elevate the performance, the high frequency will result in huge consumption and overheating of the chip. As AI rapidly iterates and advances, CPUs can no longer meet the requirements of on-chip performance and computing power.
Similar to General-purpose CPUs, AI chips can complete more computations per unit of energy consumed by incorporating huge numbers of smaller transistors, gaining speed and efficiency. AI chips possess other AI-optimized design features which will dramatically accelerate the identical, predictable, independent calculations required by AI algorithms. Due to its unique features, AI chips are tens or even thousands of times faster and more efficient than CPUs for training and inference of AI algorithms.
Different categories of AI chips are used for different tasks. GPUs are often used for initially developing and refining AI algorithms, known as "training." FPGAs are mostly used to apply trained AI algorithms to real-world data inputs, which is considered as "inference." ASICs can be designed for either training or inference. While GPUs, FPGAs, and ASICs still dominate the AI chip market, researchers are pursuing alternative models and architectures to produce systems that can deliver improvements in speed and efficiency.
China's AI Chip Industry: What is happening
The development and production of AI chips is considered as an important building block in the development of AI-related technologies and application scenarios. It is a rapidly evolving industry that has attracted a large number of investments, allowing rapid development and deployment at scale. The increased demand for AI chips is projected to contribute significantly to the industry's overall growth. The AI chips industry will stimulate innovation in chip design and present a considerable number of opportunities for new players. Many Chinese enterprises capitalized and invested heavily in AI chip design and manufacturing.
Key Policy Developments Around AI Chips Industry
AI-related companies are encouraged to strengthen their innovation capacity by the government through a myriad of support policies and incentives. In 2015, China announced its Made in China 2025 initiative. This program intended to upgrade China's entire manufacturing industry and set an ultimate goal for China to produce USD 305 billion worth of chips annually and meet approximately 80 percent of the domestic demand for chips by 2030.
The Development Plan on the Next Generation Artificial Intelligence (referred to the Development Plan), released by the State Council in July 2017, provides a detailed overview of an increasingly integrated AI ecosystem. The Development Plan demonstrates the government's intent for China to become the global leader in AI theories, technologies, and applications by 2030. It is also emphasized that China has stepped up to indigenously develop "energy-efficient, re-configurable brain-like computing chips." As an indispensable part of Made in China (2025), the Development Plan indicates that AI is considered as a catalyst for transforming China's manufacturing and service industries. The primary intention of the transformation is to improve China's economic competitiveness and catch up with the United States and other advanced economies.
The proposal Guiding opinion on promoting the deep integration of artificial intelligence and the real economy (hereinafter referred to as Guiding Opinion) was adopted at the seventh session of the central committee in 2019. The Guiding Opinion aims to accelerate the development of the advanced manufacturing industry, with the real economy as its core, and to deepen the integration of the internet, big data, artificial intelligence, and the real economy. The deep integration of information technology with the real economy is key for China to achieve industrial innovation and optimize the industrial structure. In 2021, the 14th Five-Year Plan (2021-2025) pointed out that China's new generation AI industry will mainly focus on accelerating innovation and transformation of critical and core information technology, such as critical chips, on promoting the sector towards the higher end of the global value chain.
Future Trends and Prospects
The future trend towards chip industries in China is driven by four dimensions.
First, China's market share in AI chip technology will grow in the next few years if China continues to emphasize specialized ASICs development for AI applications. According to McKinsey, ASICs will account for about 70 percent of the edge inference market while GPUs will account for 20 percent. China's leading chip-making companies heavily focus on ASICs; six of the top ten companies specialized in producing ASICs in 2018. Chinese local enterprises have fabrication capability for chips at trailing technology nodes. ASICs designed at these older nodes can provide performance advantages over chips designed for various applications.
Second, leading-edge AI applications require state-of-the-art chips for economic and efficiency purposes. Since AI algorithms iterate faster than traditional chips, Chinese AI chip startups should leverage AI chip's both cloud and edge functionality and address the latest algorithms. More importantly, they should develop AI chips progressively to strengthen existing portfolios and enable the expansion of an open ecosystem.
Third, the Chinese government advocates AI and AI-powered chips as a strategic area supported by high-level policies. Following the 2017 release of its AI strategy, China ranked number one globally in 2020 in terms of research papers on AI and the number of AI-related patents. Moreover, the Chinese government strongly supports its domestic companies in strategic industries. The government and technology giants have maintained a strong collaborative relationship. The national or local programs provide financial returns on investment to AI companies and imply the government's preferences on AI development and application.
Fourth, the application scenarios of AI chips have attracted a growing amount of awareness in the Chinese market. AI-related applications will be widely used in developing health care, education, media, finance, and customer service. Specifically, manufacturing and transportation are two main sectors with significant application scenarios of AI.
US-China technology conflict: How will it re-shape China's AI Chips Industry?
Although China invested billions of dollars in establishing a domestic semiconductor industry, it still faces tremendous challenges in becoming a leading global producer of AI chips. First of all, China's technological breakthrough in AI lacks a robust foundation in leading-edge AI chips. Until recently, AI applications run by leading-edge major Chinese technology firms were powered by exported chips from US semiconductor firms. Furthermore, as AI chips need to be advanced and specialized, the steady up-front investment costs and performance requirements have hindered China's ability to excel in the global market.
Some domestic companies have advanced their design capabilities. For instance, Huawei's Kirin 980 AI system-on-a-chip (SoC) incorporates Chinese-designed neural network processing units (NPUs). However, these chips mainly target niche market areas, such as cellular devices. Two US companies, Navidia and AMD, still play a dominating role in the international design market for GPUs. In comparison, the domestically produced GPUs provided by the leading Chinese GPU company are significantly less competitive.
Recently, the Biden administration plans to impose restrictions on semiconductors used in artificial intelligence and chip-making instruments on US exports to China. The growing tension between the two parties is disrupting China's access to advanced chips from the United States. Although China's AI chip industry is mounting vigorous and progressive development and its technological breakthroughs are well received by experts, China continues to lag significantly behind the United States concerning its fragmented domestic AI chip value chain and also in terms of its international reach. Confronted with rising US technology restrictions, China's access to leading-edge AI chips is under great threat. More seriously, the COVID-19 pandemic has further impeded China from international trade and technological flows.
Developing a robust AI chip industry at home is arguably China's most immediate and intractable concern. China's ability to lead AI chip development under the US restrictions will depend on whether or not Chinese chip developers have the sufficient technical capability to fill gaps in China's AI chip production chain. In order to achieve this, it will require Chinese tech firms to develop a large domestic and foreign customer base to attract technical expertise, making the investment economically sustainable. It is anticipated that China will be able to converge into the technological frontier in AI chip development despite US restrictions, as it has huge potential to attract huge government-supported investments, a large engineering pool, and dozens of semiconductor fabs under construction.
More significantly, the unprecedented US restrictions may act as catalysts for China to strengthen its R&D capabilities in AI, focusing on fundamental technologies. Under tremendous pressure from US export restrictions, the Chinese authorities are exerting efforts to mitigate the fragmentation of China's AI innovation system. Ironically, US-imposed technology restrictions are leading to a reform of China's technological investment and innovation policy. Chinese AI firms are forced to innovate and upgrade in terms of US competitive advantages.