Following Google, Apple, and Amazon’s own design and manufacture of AI chips, Facebook recently decided to join the camp of independent research and development of AI chips. In addition, Alibaba and Baidu in BAT have also tested water AI chip manufacturing. Giants said that independent research and development will reduce In addition to the cost, according to the chips that are customized by their products, they can be better adapted and tuned.
The outbreak of 'self-produced core' has brought threats to chip makers such as Intel, Qualcomm and Broadcom. At the same time, chip manufacturers are facing pressures such as slower growth and rising costs. Under pressure, some chip makers tried to pass Mergers and acquisitions, while cutting new technologies, cut manufacturing and development costs. Other companies, such as Nvidia and Intel, have transferred most of their chip manufacturing expertise to AI and emerging industries, and are preparing to make money by serving technology giants.
Self-produced AI chip outbreak
As AI technology continues to move into reality, AI chips have become the focus of competition in the chip industry.
Last week, Bloomberg reported that Facebook is recruiting project managers responsible for creating AI chips to reduce the dependence on chip giants such as Qualcomm and Intel. However, there is a craving for AI chips not only for Facebook, Google, and Apple. Amazon is also among them.
Back in 2006, Google considered building an application-specific integrated circuit (ASIC) for neural networks. Until 2016, Google launched a machine learning processor TPU that has a 7-year advantage over its peers, including greater latitude, and can be used more. Complex and powerful machine learning models, etc., to enable users to get smarter results more quickly.
Apple also began planning to establish an engineer team to independently design chips for the iPhone and iPad 10 years ago. Last year, Apple's iPhoneX and other three mobile phones introduced at the autumn conference adopted the independently developed AI chip, the A11 bionic chip. Replaced the chip supplier Imagination Technologies, using Apple's own design of the chip, graphics processing speed compared with the previous generation up to 30%.
Amazon, according to research institute The Information, reported in February this year, the company has begun to design and manufacture AI chips. Dedicated AI chips can also help its smart home products to reduce dependence on the overall remote server, so that smart speakers Echo and control the door The connection of home appliances such as locks and security cameras is more secure and reliable.
Analysts in the market entered the age of artificial intelligence. Because the general-purpose chips do not adapt well to the requirements of deep learning algorithms, they are inefficient, consume large amounts of power, and cost high. They eventually gave birth to the giants of science and technology trying out water cores. In order to meet the needs of AI software, it is a general trend in the industry.
On the domestic front, BAT technology giants also set foot in the AI chip business. On the 20th, Alibaba.com acquired a wholly-owned domestic chip maker, Zhongtianwei. Prior to this, Ali also invested in the Cambrian, Barefoot Networks, Shenzhen, and China. Kneron), ASR and other chip companies. In the past few years, Baidu has made considerable progress in the field of deep learning. Last year Baidu released the AI cloud computing chip called XPU.
According to the Ministry of Industry and Information Technology, the compound annual growth rate of China's IC industry from 2013 to 2017 is 21%, which is about 5 times of the global growth rate over the same period. The high-tech industry represented by the chip industry has a strong driving force for development. The international market According to market research company Report Linker, by 2023, the global AI chip market will reach 10.8 billion U.S. dollars, with a compound annual growth rate of 53.6%.
The chip manufacturer is attacked by the enemy
Organizations such as Morgan Stanley predicted at the end of last year that chip prices rose or peaked soon, and that technology giants tried to make water cores and chip manufacturers ran into trouble. In order to cope with the impact, the road to hunt for chip industry mergers and acquisitions has continued. For a long time.
In order to expand the business map to improve competitiveness and reduce manufacturing and development costs, in 2015, Avago spent USD 37 billion to acquire Broadcom. Similarly, in order to expand chip categories and expand business scope, Qualcomm announced in 2016 that it would acquire NXP for US$47 billion; In 2017, Broadcom also proposed to acquire Qualcomm with peers for US$103 billion in order to withstand the industry competition risks in the 5G era.
The International Semiconductor Industry Association predicts that in the next decade, the semiconductor industry will move from horizontal integration to upstream and downstream vertical integration. The overall strength of chip manufacturers will become stronger and stronger, and the degree of industrial concentration will increase. The oligopoly structure will be further strengthened.
In addition, some chip makers have been affected and attempted to seek scale advantages through mergers and acquisitions. At the same time, there are also leading chip manufacturers that have transformed themselves and become beneficiaries of the AI era. Intel and NVIDIA have transferred most of their chip manufacturing expertise to the company. AI and emerging areas, through the service of technology giants to profit. Among them, Intel announced last year, is working with companies such as Facebook to develop an AI chip called 'Nervana neural network processor'.
It is worth noting that Google, Facebook, and other proposals for building independent chips are still in their infancy. In fact, Google and Nvidia are still partners. Nvidia's GPUs and TPUs are on the Google cloud platform. Still compatible. And Alibaba also stated that after the launch of Ali NPU (embedded neural network processor), the Nvidia GPU will still be running on the cloud computing platform.