The popularity of ChatGPT has ignited the enthusiasm of the market for the AI industry, and the demand for hardware such as servers and switches has increased significantly. As chip carrier, PCB also embraces great increase in order quantity. Thus, what opportunities PCB indutry enables take to develop under the hottest discussion on ChatGPT in public?
What is ChatGPT?
ChatGPT is the chatbot released by OpenAI, which is a natural language processing tool driven by artificial intelligence technology. The GPT series model is a deep learning model for text generation based on data available on the Internet. It does not require supervised learning for model fine-tuning to complete tasks such as article generation, but requires a large amount of data, parameters, and powerful computing support.
GPT-3 is trained on the Azure AI supercomputing infrastructure (a high-bandwidth cluster composed of V100GPUs) provided by Microsoft, and the total computing power consumes about 3640 PF-days (that is, one quadrillion calculations per second, running for 3640 days). The continuous rise of ChatGPT has brought a new increase to AIGC, and put forward higher requirements for the computing power support required for AI model training.
ChatGPT highly replies on computing power capability
ChatGPT’s has reached 100 million users in two months after launching, making it the consumer application with the fastest-growing monthly active users. However, ChatGPT’s frequent downtime due to the surge in users recently reflects ChatGPT’s high dependence on the computing power.
Market attention has gradually spread from the application aspect and model aspect to the increasingly insufficient computing power aspect. The computing power layer mainly includes core logic training chips GPU, CPU, FPGA, as well as PCBs, connectors and other components.
In terms of market space, ChatGPT puts forward high demand for computing power in the process of model training and operation, driving the construction of computing power infrastructure. ChatGPT may drive hundreds of billions of ICT hardware investment demand in the follow three years.
ChatGPT stimulates server chips growing
In 2021, China’s server shipments will reach 3.91 million units, a year-on-year increase of 11.7%. In the next few years, as demand in the downstream market picks up, it is expected to speed up the construction of new infrastructure in the seven major areas of 5G, industrial Internet, big data centers, and artificial intelligence.
China’s server market will continue to grow steadily. The widespread application of deep learning in the field of artificial intelligence has exponentially increased the demand for computing power. According to the data, the market size of artificial intelligence chips in China is growing at an astonishing rate. By 2024, the market size will reach 78.5 billion yuan.
ChatGPT greatly increases the amount of PCBs and related components
The analysis pointed out that the ChatGPT has brought about a sharp increase in computing power requirements. For example, the GPT3 model requires thousands of a 100 chips for over a month of training time, including 175 billion parameters and 45TB of data. As the carrier of computing power and transmission hardware, servers and switches will greatly increase the consumption of components such as PCBs and connectors.
ChatGPT promotes the PCB toward high frequency and high speed
The sharp increase in computing power poses new challenges to data transmission and brings growth to the PCB industry. Core parts such as motherboards, power supply backplanes, hard disk backplanes, and network cards in servers need to adopt PCBs. While the computing power revolution has increased the amount of PCBs used, it has also promoted the development of PCBs toward high frequency and high speed.
General server or storage PCB requirements are mainly six to sixteen-layer boards and packaging substrates. With the development of 5G, cloud computing, AI, big data, the requirements for server computing power are getting higher and higher.
The demand for high-speed, large-capacity, cloud computing, and high-performance servers will increase, and the design requirements for PCBs will also continue to upgrade, such as the number of layers, large size, high aspect ratio, high density, application of high-speed materials, lead-free welding applications, etc.
The application of PCB in high-end servers mainly includes backplanes, high-level digital line cards, HDI cards, GF cards. Its characteristics are mainly reflected in the number of high-level layers, high aspect ratio, high density and high transmission rate.
In addition, the ultra-high-rise boards used for server products are too large and heavy, and the high-temperature and impact-resistant protection for such a large precision computing board will cost a lot of money. Therefore, it is the best choice to directly use HDI PCB to reduce the computing core, which is expected to bring huge market demand for HDI PCB and high-frequency PCB.
The development and application of artificial intelligence technologies such as ChatGPT and AIGC will increase the demand for high-speed computing power and storage and will promote the innovation and progress of electronic information industries such as IC substrates and PCBs.
It is reported that the output value of the global PCB industry will reach 74.76 billion US dollars in 2023. In recent years, the focus of the PCB industry has gradually shifted from Europe and the United States to Asia, forming a new pattern centered on Asia (especially mainland China) and supplemented by other regions.
While the high computing power demand of artificial intelligence brings opportunities for the development of the PCB industry, it also puts forward higher requirements for the technological innovation of PCBs. Chinese enterprises should seize development opportunities and continue to pay attention to the development of cutting-edge technologies.