AI PC is changing the PC supply chain and ecosystem

What is an AI PC?

AI PC, as the name suggests, is a personal computer that can perform AI operations. Traditional computer computing uses the CPU (Central Processing Unit) as the core. The CPU has a small number of cores, but each core has a complex structure and is suitable for complex control and operations. The new generative AI is like teaching children to recognize cats or dogs.

The problem is not complicated, but it involves the calculation and processing of a large amount of data for “training” and “inference”.

At this time, NPU (neural network) unit) is more suitable for use than the CPU. Because the characteristics of NPU and CPU are reversed, it has a very large number of cores, but each core has a simple structure and is suitable for performing a large number of repetitive general operations, which is commonly known as parallel operations.

How big is the market?

How big is the AI PC market size?

MIC estimates that a large number of new AI PC products will begin to enter the market in the second half of this year. It is estimated that the penetration rate of AI PC in the global PC market will be about 15.7% in 2024, double to 33% in 2025, and increase to 48.4% in 2026 , growing to 65.9% by 2027, which means that for every three new computers by then, 2 will be AI PCs.

Microsoft’s definition

Microsoft also proposed the latest definition. Computer models that can support its Copilot AI functions must meet the following minimum criteria:

The computing power must be at least 40 TOPS (the unit of NPU, 1 TOPS means that the processor can perform 1 trillion operations per second).
If you use Copilot’s model offline, the computing power must be more than 45 TOPS.
The memory capacity must be at least 16GB, the storage capacity must be at least 256GB, and battery life also has relevant requirements.

Please note: Windows 11 AI PC must be able to provide AI services even when not connected to the Internet.

Microsoft, which has just announced the formation of the “Copilot+ PC” ecosystem, is preparing to decentralize more Copilot AI application services to the PC side. Microsoft has already announced that it aims to sell 50 million AI PCs in the next year.

Requires small language model (SLM)

At present, most LLMs (large language models) are trained in the cloud. He believes that this wave of AI will gradually move from the cloud to edge devices, driving the AI PC and AI mobile phone markets to take off.

For small language model (SLM), please see my post of “The Importance of Small Language Models (SLM) in AI, how is Apple AI different from others?“

PC supply chain

New PC Players

Qualcomm

In mid-May 2024, Microsoft chose Qualcomm for the first time as its AI PC. Microsoft’s Copilot+ PC will be the first partner. In addition to Microsoft’s own Surface series (Snapdragon version), other major computer manufacturers such as Asus, Dell, Lenovo and HP also They have released their own Windows 11 AI PCs, but without exception, they are all equipped with the Snapdragon X Elite/X Plus chip series PC platform. This is the first time that Qualcomm’s PC platform has received such extensive support from PC ecosystem partners.

For details on this part, please see my other post “Qualcomm PC might bring it back to glory” for details.

Nvidia

In May 2024, Nvidia stated that there are already more than 200 RTX AI PCs on the market, with computing power up to 700 AI TOPS and capable of running generative AI. The newly released RTX AI notebooks will be equipped with up to GeForce RTX 4070 GPU and energy-saving system-on-chip with Windows 11 AI PC functions. These Windows 11 AI PCs will receive free updates when the Copilot+ PC experience is launched.

Nvidia released RTX Tensor core GPU and DLSS technology in 2018, ushering in the era of AI PC. Now with Project G-Assist and NVIDIA ACE, NVIDIA will open a new generation of AI experience for more than 100 million RTX AI PC users.

Microsoft is also collaborating with Nvidia to help developers use new generative AI capabilities in Windows native and web applications. The cooperation between the two parties will provide application developers with easy-to-use APIs so that they can obtain GPU-accelerated AI. Small language model and enable the Retrieval Enhanced Generation (RAG) function supported by Windows Copilot Runtime that can run on the terminal device.

Traditional PCs aren’t sitting still

Intel Core Ultra

Intel’s Core Ultra processors have built-in AI acceleration technology, so you can now enjoy a variety of experiences, including enhanced collaboration, productivity, and creativity, directly from your desktop.

At the same time, Intel also announced that it will launch the latest processor code-named Lunar Lake in the third quarter of 2024. The AI performance can be more than three times that of the previous generation processor, exceeding 40 TOPS. It aims to ship more than 40 million units this year. AI PC processor.

Apple’s Mac series of computers

Apple’s Mac series computers, which account for about 10% of the personal computer market, have been recognized in terms of performance, software, and user reviews since they ditched Intel processors and replaced them with Apple’s own M-series processors. Apple’s M-series processors contain GPU and NPU, so artificial intelligence capabilities are not a problem.

The problem with Apple’s Mac series computers is basically not about the hardware, but when Apple can integrate artificial intelligence functions into the operating system and software of Apple’s Mac series computers. Without software, everything is empty talk.

Alphabet Chromebook

Alphabet Chromebooks that can run on x86 and ARM hardware platforms have made small gains after nearly ten years of efforts by Google and Intel, Qualcomm, MediaTek, Acer and other manufacturers.

However, the influence and market share of Chromebooks are not only far from those of Microsoft Windows, but their functions and performance are also really limited. The market has already positioned it as a computer for students, only for surfing the Internet and doing simple paperwork, and cannot be used in offices.

There is currently no improvement plan or schedule for Chromebooks on AI PCs.

Conditions to success

The artificial intelligence personal computer (AI PC) market will not explode immediately, because for artificial intelligence personal computers to be popularized, hardware, software and various application services must cooperate with each other for this to happen, otherwise it will be in vain.

Network computer

Do you remember why the network computer proposed by Oracle failed more than 20 years ago? No one disputed that it was a good idea, but the time was not yet ripe, Oracle itself had no influence in the PC world, and the dot-com bubble was the reason for the failure of the Internet computer.

ARM Windows Computer

What potential does ARM Windows Computer WOA (Windows on ARM) have, which appeared more than ten years ago? , using Arm’s processor instead of Intel or AMD’s x86. Moreover, Microsoft and the hardware platform vendors of these Arm processors have not given up. In the past ten years, they have continued to launch Arm version of Windows PCs. Even Microsoft’s Surface family has sold Arm version of Windows PCs, but It was not successful for only one reason: without software cooperation, the system performance was very poor.

For WOA, please see my post of “How much potential does ARM’s WOA (Windows on ARM) has?“

Native software

Without software, it’s all empty talk

It is true that Windows can now run on non-Intel or non-Supermicro non-x86 instruction set hardware platforms (such as Qualcomm or Apple), but the performance is very poor. It can be used for display and showoff, but no one will ask for trouble. Use it to go to work and handle official business. Such a scheme existed more than a decade ago, but now only a handful of people use it. No software vendors are willing to rewrite their software for the ARM version of Windows PCs. Why? In business, if there is no market, who would want to invest double resources?

Some people will object, it can be implemented! Of course I know that I can use the simulator to execute it. The problem is that the speed and performance of the simulator are very poor. For example, would you like to use a personal computer you bought thirty years ago to run the latest version of complex software? Practical?

Current native software support

Microsoft is more fully prepared at the application level. When it releases Copilot PC products in May 2024, the entire ARM Windows Computer WOA (Windows on ARM) ecosystem will also be unveiled.

According to Microsoft, when the first batch of Copilot+ PCs are launched next month, the Chrome browser, online music application Spotify, online meeting application Zoom, video editing software DaVinci Resolve and a variety of Adobe software will all be native Supports Windows systems developed for Arm chips.

Among them, the only software vendor that can be considered a heavyweight is Adobe; after six or seven years, Microsoft and hardware vendors still have not been able to convince software vendors to invest resources and be willing to develop WOA (Windows on ARM) systems for Windows computers. Specifically rewrite another version of native software. In short, there has been little progress, and the amount of native software is still pitifully small.

Conclusion

Remember one principle – whether technological products can change the world, hardware is never the problem, the key is software.

It is unlikely that anyone would be willing to spend money from his own pocket to buy a computer that cannot run the software he is currently using; and it is a computer that is much more expensive than the market price. If it were you, would you?

Disclaimer

The content of this site is the author’s personal opinions and is for reference only. I am not responsible for the correctness, opinions, and immediacy of the content and information of the article. Readers must make their own judgments.
I shall not be liable for any damages or other legal liabilities for the direct or indirect losses caused by the readers’ direct or indirect reliance on and reference to the information on this site, or all the responsibilities arising therefrom, as a result of any investment behavior.