In 2016, the AlphaGo computer program beat South Korean Go player Lee Sedol 4:1, shocking the world with the power of artificial intelligence (AI). Today, AI has penetrated everyday life and traditional industries, impacting everything from voice assistants and autonomous driving to unlocking smartphones with face recognition.

AI has also become a mainstay in China's top-level national planning. Ever since it was first mentioned in the Chinese government's annual Report on the Work of the Government, AI has become a keyword mentioned year after year.

A wide range of Chinese companies, including tech companies, internet giants, and startups, have all invested in AI, which has also attracted a large influx of capital. However, traditional chips can no longer meet the functional and computing requirements of today's AI industry.

To meet the demands of future AI, companies must find a way to build high-performance AI chips that will effectively combine chip and AI technologies.

China seizes AI chip development opportunities

A new generation of AI was born with breakthroughs in deep learning in 2006. Along with algorithm development, AI technology and its applications continued to progress in various fields including image, voice, and natural language. This round of AI industry development was inextricably linked to the improvement of computing power - increasingly complex deep neural networks, which trigger the migration of AI computing from CPUs to GPUs.

Additionally, as AI chips mature, the industry continues to grow. The global AI chip market is expected to reach US$72.6 billion by 2025 at a compound annual growth rate (CAGR) of 46.14%, according to data from Tractica.

With such good market prospects, leading international companies, including Nvidia, Google, Intel, and Xilinx, are developing new products in this field.

In addition to their own development, leading traditional semiconductor manufacturers and tech giants rely on their capital advantage to carry out merger and acquisition (M&A) financing overseas, as well as attract quality capital and innovative teams to promote their growth in the AI chip industry. For example, Intel acquired AI chip companies Altera, Nervana, Movidius, and Mobileye.

In terms of Chinese companies, major internet and ICT companies are vying to enter the AI chip market. China's top three internet companies (Tencent, Baidu, and Alibaba) are developing their own AI chips. Internet startups Meituan and ByteDance are also entering the market.

Among ICT companies, in 2018, Huawei released the Ascend series AI processors, the world's first AI IP and chip series. Since then, the Ascend 310 product and cloud service has been widely used in applications such as smart industrial parks and autonomous driving.

As the market grows, more and more AI startups with their own AI chips have emerged in China, including Cambricon Technologies, Enflame Technology, Iluvatar CoreX, Horizon Robotics, and Black Sesame Technologies (BST).

The popularity of the market has also attracted significant capital. According to EqualOcean (EO) Intelligence, as of January 2022, there were 92 financing events in AI-chip-related fields in China in 2021, totaling approximately CNY30 billion.

Based on a report by Questel, chip patents in China have grown rapidly over the last decade. In fact, China has become the country with the most patent applications.

AI chips offer an excellent opportunity for the development of Chinese companies. Chinese companies that already have the technical and intellectual property (IP) foundations to compete with and work with international companies have the potential to flourish in the AI chip field.

Autonomous driving a major growth sector for AI chips

In 1939, General Motors (GM) used its Futurama exhibit at the World's Fair in New York to display its vision for self-driving cars. It was not until 2015 that the auto industry realized the hope of autonomous vehicles and began exploring the implementation and industrialization of autonomous driving technology.

Today, the industry has shared the consensus that smart driving lies in the fusion of chips and vehicles.

The automotive electronics industry has grown by leaps and bounds from the earliest electronic control units (ECU). Today, automotive chips can be roughly divided into three categories: AI chips, microcontroller units (MCU), and IGBT power components. Among these, AI chips are the key to transforming traditional vehicles into smart vehicles - they are responsible for self-driving, as well as human-computer interaction computing and processing.

The auto industry is currently moving from mechanical-defined vehicles to smart-defined ones, and electric vehicles (EV) need a larger amount and higher quality automotive chips than fuel-powered cars. Furthermore, as the penetration rate of new energy vehicles (NEV) substantially increases and smart driving draws nearer, automotive AI chips are at the core.

The advent of high-level intelligent driving means that smarter vehicles need to process even more unstructured image and video data. Traditional MCU chips alone cannot meet these computing requirements, but AI chips can achieve fast, accurate, and smart computing.

The global automotive AI chip market is expected to grow at a CAGR of 31% through 2025 and reach US$23.6 billion, according to data from Gartner. The Chinese automotive AI chip market is expected to reach US$6.8 billion by 2025, growing to US$12.4 billion by 2030 at a CAGR of 28.14%. The automotive industry is expected to be a main growth sector for AI chips.

Integrating more AI units is a major trend in the development of smart chip technology. Currently, the main paths for AI chips are GPUs, CPUs, ASICs, FPGAs, and NSoC.

CPUs are good at logic control and general-purpose data calculations and are irreplaceable. GPUs are good at large-scale parallel computing. FPGAs have high computing power and are suitable for small-scale customized development and testing. ASICs also have high computing power and are highly energy efficient. NSoCs refer to chips integrated with more neural network units and can achieve fast convolutional neural network (CNN) computing; however, they can only support a small number of algorithms.

At present, there is still a lot of controversy over the use of general GPU, FPGA, and ASIC chip solutions for future vehicle AI chips. Adding neural network units to AI chips will allow them to perform AI calculations more efficiently.

Although newly emerging NSoCs are not an ASIC fixed algorithm, they have the advantage of lower cost and lower power consumption; however, their adaptability to different scenarios is still relatively weak. In the automotive sector, both future performance and cost will be equally important.

Automotive AI chip market to standout

Since 2015, China has ranked first in the world in terms of production and sales of NEVs for seven consecutive years. Data from the China Association of Automobile Manufacturers (CAAM) revealed that the market share of domestic brand passenger vehicles reached 44.4% of the Chinese market, driven mainly by NEV brands. In fact, China has become an important driving force in the transformation of the global auto industry to battery electric vehicles (BEV).

The Chinese auto industry achieved decent results in the first half of the automotive revolution. Along with the development of the four new modernizations for vehicles (electrification, networking, intelligence, and sharing), power systems have switched from internal combustion engines to battery electric; control systems have evolved from distributed to centralized; auto brands have gone from closed to open systems. These once-in-a-century changes are now entering the second half.

So what will be at the heart of the competition in the second half? What is required of automotive smart technology? The core competition will be in smart chips, the "brain" of smart cars.

The shortage of automotive chips has worsened since the second half of 2020. Many new automakers have had to delay deliveries and postpone announcements of new models, underscoring the importance of automotive AI chips.

Automotive chips in China, much like chips for consumer electronics, rely heavily on imports. Data from the China Automotive Chip Industry Innovation Strategic Alliance shows that Chinese-manufactured automotive chips only accounted for 4.5% of the global market in 2019, with the country's dependence on overseas automotive chips as high as 90%.

Foreign chip manufacturers dominate the automotive chip market, relying on their established advantages in the traditional MCU chip sector. The chip shortage has further highlighted this issue.

Traditional automotive chip companies such as Renesas Electronics, NXP Semiconductors, and Texas Instruments (TI) are currently the leading mass producers with extensive experience in automotive chip design. The deep link between the embedded processor field and automotive software and systems development allows for better coordination of vehicle control and control function safety requirements.

Additionally, Nvidia, Qualcomm, and Intel have made large-scale deployments of automotive core control chips in recent years due to the rapid development of automotive intelligence. They now rank in the top 25 global automotive semiconductors.

The low-level driver assistance segment (L1-L2) is dominated by Mobileye, Intel subsidiary, and Xilinx. At one point, Mobileye had a market share of more than 70%. In 2020, Mobileye shipped nearly 20 million units, while Xilinx shipped more than 7 million.

Mobileye's strength is in vision technology, while Xilinx excels at perceptual computing. Automakers generally use Mobileye's vision solution and Xilinx's millimeter wave (mmWave) radar chip for L2-level autonomous driving.

In March 2017, Intel acquired Mobileye for US$15.3 billion. In February 2022, AMD completed its acquisition of Xilinx for US$35 billion. As a result, in the future, Intel and AMD are expected to become leaders in the field of low-level driver assistance chips.

Nvidia and Qualcomm have joined the high-level segment (L2+ and above) and have adopted different strategies.

Qualcomm is focused on smart cockpits, applying its experience and technological advantages in the consumer electronics field. The majority of chips for mainstream smart cockpits currently come from Qualcomm.

Nvidia is a latecomer to the field, relying on its technological strengths. Its Xavier, Orin, and Atlan chip series, as well as Hyperion and Drive AGX system platforms can support L2-L5 levels of autonomous driving, helping Nvidia lead the way in the high-level autonomous driving sector.

Tesla, which has taken the "complete link" route for its self-developed chips, is the most successful auto company to have both software and hardware. Its self-developed Full Self-Driving (FSD) chip was mass-produced and used in the Model 3. Tesla's FSD business brought in US$1 billion in 2020 and future revenue from FSD is expected to exceed vehicle sales. In August 2021, Tesla announced the Dojo supercomputer, which has a processing power of up to 362 trillion operations per second (TOPS) and uses the 7nm process; it is expected to go into mass production in 2022.

Foreign tech giants are racing against each other in the automotive AI chip sector while Chinese automotive chip development is still in the early stages. However, industry insiders believe the environment in China is good for automotive AI chips, noting that the largest market for autonomous driving is China. This is a good opportunity for Chinese suppliers, but they must build the foundation now while the competition is just getting started.

By DIGITIMES