标签归档 夜西安论坛

The growth rate of global server industry’s shipments this year is approaching the "Guarantee 1" war. However, with the rise of self-driving, AIoT and edge computing and the explosion of artificial intelligence (AI) chat robot "ChatGPT", AI servers will become a pioneer in the growth of server niche this year. It is estimated that the annual growth rate of shipments will reach 8%, and Taiwan Province supply chains such as TSMC, Weiying and Quanta are expected to benefit.

According to the report of TrendForce, a research institute, under the leadership of emerging application themes in 2018, many large cloud operators began to invest heavily in the construction of AI-related equipment. By 2022, it is estimated that the annual shipment of AI servers equipped with GPGPU(General Purpose GPU) accounts for nearly 1% of the total server.

It is estimated that in 2023, with the application of chat bots, it is expected to stimulate AI-related fields again. It is estimated that the annual growth rate of shipments will reach 8%, and the compound growth rate from 2022 to 2026 will reach 10.8%.

In 2022, the proportion of AI server procurement is dominated by Google, AWS, Meta and Microsoft, the four major cloud providers in North America, accounting for 66.2%. In recent years, with the intensification of localization in mainland China, the wave of AI construction has warmed up, and ByteDance, Tencent, Alibaba and Baidu are the providers with large purchases.

In terms of computing specifications, ChatGPT is mainly based on NVIDIA A100 and exclusively uses Microsoft Azure cloud service resources. If other Microsoft applications are merged, the annual demand for Microsoft AI servers is 25,000. Baidu Wenxin adopts NVIDIA A800 in a word and combines other applications. The annual demand for Baidu AI servers is about 2,000 units.

Taiwan Province AI server supply chain includes TSMC, IC design factory Creative and Purui -KY, server foundries Quanta, Weiying and Inventec. The market is promising. With the contrarian growth of AI servers, the supply chains of related factories are expected to benefit.

Among them, whether it is AI computing or data center, most of the back graphics cards are manufactured by TSMC, especially on the AI chips led by Nvidia, including GPU A100/H100. Related accelerators and chip design applications are mainly produced by TSMC’s 5-nanometer family process.

Wei Ying pointed out that in 2022, AI servers will be less than 20% in terms of shipment or revenue. At present, more than 50% of newly opened cases are AI-related servers.

Quanta has invested in the field of AI servers many years ago. Although the revenue proportion of AI servers has not been announced, the legal person pointed out that the unit price of the main board of AI servers is twice as high as that of standard products, and it is difficult to manufacture, and the gross profit margin is rising, which is beneficial to the improvement of product portfolio.