“Science and Technology Innovation Board Daily”, May 21 — On Monday local time, Microsoft introduced “CoPilot+PC” at the press conference, symbolizing that AI PCs have reached a new level.

According to reports, compared to many AI PCs launched earlier, Microsoft has now fully introduced its AI assistant Copilot into the Windows system, and has built in OpenAI's GPT-4O model. It is reported that its hardware AI computing power can perform more than 40 trillion operations per second. According to official Microsoft tests, the multi-threading performance of the first CoPilot+ PC released, the new Surface Pro, is 58% higher than Apple's MacBook Air.

Influenced by this news, AIPC concept stocks are gaining strength today. As of press release, Yingli shares have risen or stopped, Siquan New Materials has risen by more than 12%, Thor Technology has risen more than 10%, Zhidi Technology has risen more than 7%, and Gelin Precision, Huaqin Technology, and Longyang Electronics have continued to rise.

Notably, the NPU (neural network processor) computing power of the Surface Pro released this time reached 40 TOPs (trillion processing times per second). In fact, Microsoft earlier regarded 40 tops as the entry threshold for AI PCs, and Intel also revealed earlier that the prerequisite for running Microsoft Copilot locally is “an NPU requiring at least 40 tops of computing power.”

What is the equivalent of an NPU that performs more than 40 trillion operations per second? The iPad Pro, which was introduced not long ago, has an NPU computing power of 38 TOPs. Intel's current Meteor Lake chip combination power can reach 34 TOPs. Currently, the vast majority of AI PCs on the market have a minimum NPU computing power of only 10 TOPS, so there is no possibility of running Microsoft Copilot locally. In this sense, Microsoft's move to introduce CoPilot+ PC is tantamount to redefining the perception of AI PCs.

So what does CoPilot+PC do?

The first is the new function Recall. With the support of GPT-4o, Copilot+PC can recall and capture everything that has ever appeared on the display.

Second, with the help of powerful computing power and language models, Copilot+PC can translate around 40 languages in real time and generate or optimize AI images in real time.

Furthermore, CoPilot+ PC has sufficient real-time interaction capabilities. The Microsoft press conference demonstrated the game “My World”. The results showed that AI provided players with basic information equivalent to a game strategy, including how to build items and avoid zombies.

In any case, the advent of Copilot+PC symbolizes an advance in AI PC. 2024 is known as the first year of AI PCs. The development of AI is driving the boom in products such as AI PCs, AI servers, and AI mobile phones in the market. It is reported that in the future, computer manufacturers including Lenovo, Dell, HP, Samsung, etc. will also launch AI PCs that can run Microsoft Copilot. According to Canalys, the market share of AI PCs will reach 19% in 2024.

Returning from a domestic perspective, the Donghai Securities Research Report points out that currently, various consumer electronics manufacturers are releasing AI PCs one after another, and AI models are rapidly iterating. Competition for AI to land on hardware will tend to heat up in the future. It is recommended to focus on the AI chip-related industry chain. Toubao Research Institute pointed out that the development of AI PCs has driven PCs to carry out comprehensive upgrades and iterations from software to hardware, showing a changing trend of hardware first and then software. Currently, AI PCs have entered the first stage, mainly hardware-side changes. It is expected that in the mid to late 2024, they will enter software-side changes such as personal models and personal agents.

From an investment perspective, Debon Securities suggests that demand for AI PC superimposed switches is expected to drive a sharp rise in the upstream industry chain. It is recommended to focus on targets related to the PC upstream industry: 1) complete machines and ODM: Lenovo Group, Wingtech, Huaqin Technology, and Yidao Information; 2) Components: Everbright Tongchuang, Chunqiu Electronics, Hanbo Hi-Tech, Huibao Hi-Tech, Huichuangda; 3) Cooling: Siquan New Materials, Feirongda, and Zhongshi Technology; 4) IC design: Xinhai Technology, Zhongxin Technology.