At this year's "Science and Technology Spring Festival Gala" CES, AI PC stood in a well-deserved C position. Both chip manufacturers and PC brands have set their sights on this field, and it has become one of the hottest concepts in the technology circle overnight.
In fact, in the second half of last year, a number of PC brands released or previewed their so-called AI PC products. Microsoft has decided to tweak the keyboard of the new PC and introduce the physical keys of the AI assistant Copilot to computers running Windows 11, and the first devices will be available this month. You know, the last time the keyboard layout of Windows PCs changed was 30 years ago.
In addition, the forecasts of many institutions are also very optimistic. According to the latest report of Canalys, the global share of AI-compatible PCs will account for 19% in 2024 and grow rapidly to 37% in 2025.
Whether it is the recognized technical content or the positive attitude of the industry, it seems that the PC industry in 2024 will be bright, but this iteration driven by the supply chain is also worth thinking about.
iPhone moment for PC?
The AI model has made many technology product tracks have the saying of the iPhone moment, and the PC industry is even a little late.
In September last year, Intel first proposed the concept of AI PCs and announced the launch of the industry's first AI PC acceleration program. Three months later, the new Core Ultra mobile processor was officially announced, using Intel's first on-chip AI accelerator for clients, the Neural Network Processing Unit (NPU).
Iterations of the upstream supply chain are quickly transferred downstream. Lenovo, Acer, Asus and other big brands have officially announced new products equipped with Intel's new processors, in the face of this market opportunity that has been waiting for 40 years, the major PC manufacturers obviously cannot remain reserved, and work together to push the AI PC to the first wave of climax.
Theoretically, the current product form and user habits, the story of AI PC makes perfect sense.
PCs have the same high-frequency interaction attributes as mobile phones, smart speakers, smart glasses and other terminal devices, and as a full-scene productivity tool, AI will be more obvious on PC to improve efficiency. In addition, compared with other terminal devices, PCs can also carry more diverse AI ecosystems due to their advantages in storage capacity, device space, and core hardware.
At present, from the perspective of the relevant applications of AI large models, PC is still the optimal carrier of large models. From ChatGPT to Microsoft 365 Copilot, killer apps often appear first on the PC, and AI has an immediate impact on efficiency on such productivity devices.
Because PCs will be simpler to deal with issues such as power consumption and computing power, even laptops still have room for improvement due to the iteration of GPU performance and the emergence of the latest NPU architecture. Lenovo's first AI computers were concentrated in notebooks, which is also proof of this.
In short, PCs, especially laptops, not only meet the hardware requirements of the scale computing power required by large models, but also do not excessively affect users' habits of using computer-type products.
Hua Chuang Securities believes that low latency + privacy protection will drive the demand for localization of large models, and the edge computing power of AI PCs is expected to solve the pain points of the industry and become the first large-scale landing AI terminal. The implementation of some AI applications on the PC user side is expected to drive a large number of PC replacement needs, or drive the PC industry into a new round of upward cycle.
IDC predicts that China's PC market will maintain a stable growth trend in the next five years due to the arrival of AI PCs, and the total market size of desktops, laptops, and tablets will increase from 68 million units in 2023 to more than 80 million units in 2027, an increase of nearly 18%. NVIDIA CEO Jensen Huang also pointed out that in the next 10 years, new AI PCs will replace traditional PCs, with a market value of trillions of dollars.
However, prediction is prediction, and reality is reality.
The evolution and development process of AI PC is divided into two stages: AI Ready and AI On. In the AI Ready stage, AI PCs are mainly upgraded to chip computing architectures with basic local hybrid AI computing power, while in the AI On stage, AI PCs have complete core features, providing epoch-making AI innovation experiences, providing personal AI assistant services in general scenarios based on a richer AI application ecosystem, and fine-tuning services for personal large models in edge private domain environments.
Today's first wave of craze has been set off, but the change of processor can only be regarded as an AI ready PC, which has hardware capabilities, but cannot apply AI independently, and the follow-up needs the support of upper-level model developers and application software manufacturers. Therefore, how much waves can be caused in the dull PC market depends on the landing of follow-up products.
Today's AI PCs are like the original iPhones, which are amazing enough, but it's not easy for every user to have and recognize them. The linkage between the upstream and downstream of the PC is a huge driving force, but the more important thing is the cost of the AI PC, whether it can inspire and benefit every user.
Reshaping the PC industry, cost first?
To meet the needs of users and allow everyone to enjoy the exclusive AI capability extension in the era of large models, it naturally puts forward higher requirements for the software and hardware of PCs.
With the upgrade of Intel or AMD, there may be a 10% to 20% price difference at the time of release of regular PC processors, but after half a year or a year, the price will basically be the same as the previous generation of products with the same configuration. AI PCs need to be paired with higher computing power, whether it is paired with a graphics card or NVIDIA GPU or computing power card, or integrated NPU into the CPU, the cost will increase significantly. For example, NVIDIA's card with about 75Tops of computing power may cost 7 or 8,000 yuan, in which case, the overall cost may be doubled.
However, considering that the products currently landed are all mid-to-high-end, the profit margin is not small, so the price change is not large. Not long ago, the first batch of laptop products equipped with Intel AI processor Meteor lake were launched, and the starting price of 14999 yuan is the same as that of the previous generation of the same series.
It is worth mentioning that in addition to the CPU, memory, hard disk, and heat dissipation are the most likely to be upgraded in the future. For example, a regular thin and light notebook can be used for about 6 hours at most, while a device equipped with an AI computing card may expect a battery life of 4 hours, or even only 2~3 hours when running at full power.
However, these are only paper parameters tested by hardware under certain conditions, and the actual upper limit depends on the specific product and ecological landing. Huo Jinjie, President of IDC China, said that the birth of AI PC marks that the terminal industry represented by PC is about to enter an era of great industry change, and all roles in the entire PC industry ecosystem will change because of AI PC.
Of course, the specific effect will vary from model to model, which is a test of the adaptation between AI hardware manufacturers and basic large models, AI PC may still be more biased towards B-end company procurement for a period of time after the launch, and the large-scale emergence of C-end demand may require PC manufacturers to give a sufficiently attractive case and effect.
Another additional point to consider is how to share the high costs of large model iterations among vendors. It is difficult to significantly reduce the training iteration and running cost of large models in the short term, but how to implement and iterate large model products on personal computers, and whether they have a buyout system or a subscription system, are all questions.
Judging from the products in the official mall, at present, Lenovo's newly released Pro 16 IMH, ThinkPad X1 Carbon AI and other AI PCs, its AI concept is still focused on the hardware provided by Intel, including the Core Ultra processor and the NPU AI engine, the only specific scenarioAI conference function will be upgraded in the form of OTA, and the concept of "private large model" and other applications is not clearly mentioned.
The current AI model, if you only look at the capabilities of the chip, is completely leeway to do larger-scale operations and develop some additional functions. However, it is still a difficult problem to balance the computation that can be provided by the large model runtime with the system resources that the computation itself needs.
For laptops, deploying large models on-premise means that larger storage capacity and higher hard drive costs are required, while for thin and light business laptops, cloud deployment not only requires higher costs, but also creates a paradox of security, and how to balance it is also a test for PC manufacturers. After all, since the development of the PC, many market perceptions have formed a strong inertia, such as size, weight, design and even price range, there is no particularly large room for adjustment.
Frankly speaking, there is no doubt that AI PC is already a new trend in PC applications, and the cost problem will not be unsolvable after iteration.
The time has not yet come
Although many research institutions are optimistic that AI PCs will promote the arrival of PC replacement waves, this is not a foolproof driving force.
The current PC market as a whole is on a downward trend, and AI PC, as a new product form, although it has improved localized computing power, is only used as a personal assistant, which has limited promotion for the replacement of the overall market. AI products such as ChatGPT show an optimistic trend with a high initial number of registrations, but users will find their limitations after becoming proficient in using them, and they cannot completely replace all jobs and applications. As a result, AI PCs may attract some early adopters to switch devices, but most people won't switch devices just because of this feature.
Consumers are more inclined to upgrade at the normal pace of obsolescence. Due to the impact of global economic restrictions and small product breakthroughs in recent years, the replacement cycle has been extended. It is expected that by 2024, there may be a period of replacement growth, when combined with the AIPC concept, the market may see relatively significant growth, but part of it is still the user group of conventional replacement.
However, for the industry, the good news is that the PC industry has the momentum to get out of the trough after experiencing labor pains.
According to the latest data released by IDC, global PC shipments in Q3 2023 will be 68.2 million units, an increase of 11% month-on-month, and the year-on-year decline will narrow to 8%.
However, it should not be ignored that whether you are willing to pay for the extra functions related to AI PCs, or change your phone because of the functions of offline large models, depends on how the functions of the PC end side will help consumers. For example, for students, the arguments and opinions that need AI PC to help summarize and summarize materials are actually only needed at that moment or a certain period of time, or for part-time workers, PowerPoint and Word need to become smarter, then they definitely need the assistance of an end-side AI PC function.
In short, if the functions on the AI PC are more relevant to the daily work, study, and life of the consumer group, then the probability of being willing to change the phone is higher. Therefore, the killer application or the actual application effect of the end-side large model determines the replacement power in the market.
With the integration of AI technology on the PC side, the dependence on software will increase, and the ecological binding will be tighter.
In the future, the differentiation of AIPC will be mainly reflected in software capabilities and user experience. Just like the smart speaker market, there is not much difference in hardware, but there are advantages in software optimization and experience. The superposition of software and hardware will bring about a leap in the industry, and the AI On stage will have the core characteristics of the AI PC. Therefore, starting from next year, the competition between manufacturers will be more reflected in the optimization and experience of software, and the market will gradually open up at the same time as competition.
Finally, when we discuss how capital and the industry view this AI PC revolution, please don't forget that Moore's Law and the global division of labor in the industrial chain still dominate the development of the PC industry, although their essence is to serve consumers, but also bring many variables and questions to the development of the industry.
Source: Titanium Media
Author: Science and Technology News
Original link: https://www.tmtpost.com/6874659.html