The development status of the artificial intelligence chip industry
Artificial intelligence is an important driving force for a new round of scientific and technological revolution and industrial transformation, and has become the focus of global scientific and technological competition. Artificial intelligence chips (AI chips are the core hardware foundation of artificial intelligence technology and have a huge impact on the development of artificial intelligence. Broadly speaking, AI chips refer to modules that are specially designed to handle a large number of computing tasks in artificial intelligence applications, that is, chips for artificial intelligence are called AI chips. AI chips in a narrow sense refer to chips that have been specially accelerated for artificial intelligence algorithms. "No chip, no AI", the computing power realized by AI chips is an important measure of the development level of artificial intelligence.
The development of AI chips has gone through hardships, but the current status of industrial development is good. Before 2006, there were no breakthrough AI algorithms, and access to data resources was limited. Traditional CPUs have been able to fully meet the computing needs at that time, and there is no special demand for AI chips in academia and industry, so the development of the AI chip industry has been relatively slow. After 2006, the emergence of deep learning algorithms promoted the development of the AI chip industry. In 2012, AlexNet achieved great success in the ImageNet competition, marking the arrival of the era of deep learning and putting forward higher computing power requirements for AI chips. In 2016, AlphaGo defeated the top human chess players, further promoting the development of the AI chip industry.
The integration of AI chips and integrated circuit industries
As a branch of chips, AI chips have their specificity and universality, dedicated to the field of artificial intelligence, and at the same time, like other chips, they are inseparable from the development of the integrated circuit industry itself. The development of AI chips is restricted by the development level of the integrated circuit industry, and at the same time provides a new direction for the development of the integrated circuit industry. Artificial intelligence, especially deep learning, has developed explosively in recent years, largely thanks to the accumulation of integrated circuit technology for many years. If it weren't for the fact that integrated circuit technology has developed to a certain level and can provide enough processing power for large-scale machine learning, AlphaGo would not have defeated the top human chess players.
The integrated circuit industry chain can be divided into four major links: design, manufacturing, packaging and testing, and equipment materials. The upstream of the AI chip industry chain is chip design, the midstream is chip manufacturing, and the downstream is system integration and application development. At present, China has made some progress in chip design, and the AI chip design capabilities of Huawei HiSilicon, Horizon and other enterprises have reached a high level. However, in the chip manufacturing process, the mainland is still facing a large shortcoming. Huawei and research institutes in China have begun the research and development of high-end lithography machines, but the technical level has yet to be verified. In the midstream manufacturing link of the industrial chain, the barriers to entry are high, the level of localization is low, and third-party foundries such as SMIC still need to improve their process support for AI chips.
The development of the AI chip industry is strong and the competition is fierce
In the process of the development of the AI chip industry, traditional chip companies, IT giants and emerging AI chip startups are actively deploying and fierce competition.
Traditional chip companies have an obvious dominant position
Traditional chip manufacturers such as Qualcomm, NVIDIA, Intel, and AMD have rapidly entered the field of artificial intelligence by virtue of their leading positions in the chip field for many years, and are currently in a position to lead the development of the industry. When it comes to GPUs and FPGAs, they're basically in a monopoly position. For example, Nvidia has launched the Tesla series of GPU chips, which are specifically designed for deep learning algorithm acceleration.
IT giants have increased the R&D and customization of AI chips
Since 2015, international Internet and IT giants such as Google, Microsoft, IBM, Meta, Apple, and Amazon have carried out cross-border research and development of AI chips, trying to break through the bottleneck of computing power and control the core components in their own hands. For example, in 2016, Google released a chip TPU developed specifically for the open-source framework TensorFlow to accelerate machine learning workloads.
Domestic enterprises have accelerated the pace of independent innovation
In China, companies such as Huawei, Horizon, Cambrian, and Yuntian Lifei have carried out a lot of independent innovation in the field of AI chips and have made certain achievements. For example, Huawei's Ascend 910 is the first AI training chip based on an autonomous instruction set architecture in China, and its performance is close to NVIDIA's top products. The "Rising Sun" series of edge AI chips launched by Horizon have been widely used in scenarios such as intelligent driving and intelligent security.
The field of brain-like chips is on the rise
In addition to traditional digital chips, analog chips are also shining in the field of AI. The analog chip borrows from the working principle of biological neurons, and has the advantages of high energy efficiency and low latency, and is called "brain-like chip". Domestic companies such as Horizon and Zhongtian Micro have made breakthroughs in the field of brain-like chips and launched a series of innovative products, bringing new vitality to the AI chip industry.
The development trend of the AI chip industry is good, but due to the involvement of multiple links, the mainland faces different challenges in different links of the industrial chain, and needs to make continuous efforts, increase independent innovation, and narrow the gap with the international advanced level.
Nvidia sees Huawei as its biggest competitor
In the development process of the AI chip industry, NVIDIA has long occupied a leading position, but recently regarded Huawei as its biggest competitor for the following two reasons:
Changes in the global pattern of AI chips
With the rapid development of AI technology, the global landscape of advanced process chips driving AI technology is changing. In its filing with the U.S. Securities and Exchange Commission, Nvidia identified Huawei as its top competitor for the first time in a number of major categories, including artificial intelligence chips. Nvidia positions Huawei as a cloud services company that can design its own hardware and software to improve AI computing. "Huawei may even have more marketing, financial, distribution, and manufacturing resources than we do, and may be more able to adapt to customer or technology changes," Nvidia noted. ＂
Huawei chips became a replacement
Affected by a series of US AI chip export restrictions, Nvidia has been unable to export advanced chips to China, and Huawei's products have become an excellent alternative to its products in the Chinese market. Shi pointed out that Huawei's Ascend series of AI chips, especially the Ascend 910B launched last year, is seen as a replacement for the A100 chip launched by Nvidia three years ago in the Chinese market. Previously, Chinese technology giants Baidu, Tencent, Alibaba, etc. have long been important customers of Nvidia in the past, but there is news that Baidu has turned to Huawei to place orders, which indicates that Chinese companies have begun to get rid of their dependence on American technology under the continuous increase in export controls to China by the United States.
Nvidia CEO Jensen Huang has publicly stated that although Huawei is limited by its own semiconductor processing technology, it can still build a very large system by aggregating many chips together, so Huawei is a "very difficult to deal with" AI chip competitor. Huang also specifically mentioned that even so, Nvidia is still about a decade ahead of Huawei.
Nvidia regards Huawei as its biggest competitor, reflecting that the pattern of the AI chip industry is undergoing profound changes, and the independent innovation capabilities of Chinese enterprises have been internationally recognized, but there is still a certain gap with the international advanced level, and it is necessary to continue to increase innovation.