The key change of COMPUTEX 2024 in the communication industry: ease of use and end-to-end diffusion of computing power
Core point:
With the theme of "Connecting AI", Comuptex 2024 covers many topics such as AI computing power, forward-looking communication, future mobile, XR, green energy sustainability, among which AI computing power concept edge update led by Nvidia and AMD is the most obvious. We summarized two key changes for the development strategy of AI computing power manufacturers and products on Computex.
(1) Nvidia began to think about the long-term strategy of computing power iteration while raising the upper limit of GPU performance. First, Nvidia continues to gain insight into user needs on the software side and provide optimization solutions. Nvidia announced that NIM (an inference microservice) can be used at the Computex pre conference speech, and we summarized its idea with "out of the box": Nvidia agrees with the business model of open source models such as Llama and Mistral, and NIM uses more than 40 open source models as the "core", so that developers can deploy AI applications "within minutes" (rather than weeks) after installing the system. Second, Nvidia actively adapts to changes in hardware technology and seeks the optimal solution in the changes. NVIDIA has laid out Ethernet products. According to its plan, 1.6T (physical bandwidth) port IB/Ethernet switch products are expected to be available in 2026, and the scale of computing power cluster is expected to be expanded to the level of million GPU interconnection. The demand for supporting 3.2T (logical bandwidth) optical modules is expected to rise accordingly. Third, Nvidia launched GB200 NVL2 to position small embedded computing servers, which is conducive to accelerating the penetration of computing servers into traditional data centers. We believe that Nvidia is making software, entering Ethernet and launching small servers to make computing easier to access and use.
(2) Microsoft and Google lead software and Qualcomm lead hardware. 2024 is expected to be the first year of AI PC/AI mobile phones. The reason why 2024 is called the "first year" is that we have observed two changes in AI PC/AI mobile phones: first, the head manufacturers' plans for device side AI are clearer than last year.
Before the opening of Computex 2024, Microsoft updated the surface product line with the theme of "Copilot+PC"; Google has a clearer plan for Android AI this year, that is, building AI applications by combining the cloud (Gemini Pro) and device (Gemini Nano).
Second, AI PC is close to Arm architecture. The new products of Microsoft's Surface series all use Qualcomm Snapdragon X series processors based on the Arm architecture. Microsoft believes that the AI efficiency and work efficiency of ARM PCs have increased by 20 times and 100 times respectively, and the Arm ecosystem is constantly improving. At present, 87% of users' use time is based on native Arm software. We believe that in the AI PC era, Arm is expected to gradually penetrate the mobile PC market with its advantages of low power consumption and high customization.
Investment advice. Looking at the optical module from the perspective of development, the variables are the diversification of technical paths, the shortening of iteration cycle, and the concentration of supply patterns to the head. What remains unchanged is that the GPU always needs to be connected, and optical connection is the optimal solution in most scenarios; If the AI PC is scaled up, the demand for cloud reasoning computing power will be pulled accordingly under the scheme of cloud computing power cooperation on both sides. Continue to recommend Zhongji Xuchuang, Tianfu Communication and New Yisheng.
Risk warning. Risk that AI PC penetration rate is lower than expected; The risk that AI PC is used less frequently than expected; The risk of increased sanctions imposed by the United States on domestic technology enterprises; Risk of intensified industry competition
